Optimizing deployment model of flexible interconnection systems for new energy 6/10 kV medium-voltage grids

Abstract

Introduction

The rapid evolution of new energy technologies has enriched the distribution grid with enhanced diversity and increased electrical capacity facilitated by distributed renewable energy sources. Nevertheless, the prevalent issue of unstable power supply from these sources has compromised the grid's reliability in delivering electricity, thereby posing a threat to power quality assurance [1–3]. The inherent variability and oscillation linked with renewable energy sources present substantial hurdles in assimilating high-frequency power electronic systems into the grid. With the expanding infiltration of new energy sources in the power system, the distribution grid is anticipated to gradually confront challenges associated with the low inertia and inadequate damping of new energy power [4]. Additionally, these complexities, exacerbated by escalating power demands, manifest in occurrences of power flow reversal and surpassing system thresholds [5], further intensified by the substantial power transfer to flexible electronic devices such as electric vehicles and energy storage units via high-penetration methodologies. This transition prompts adjustments in auxiliary power and voltage control mechanisms, potentially leading to equipment self-ignition and other safety hazards that significantly endanger grid security and the operational safety of adaptable power systems [6, 7].

To address this issue, researchers have delved into linking the direct current buses of disparate distribution lines to establish an interconnected system fostering energy exchange among diverse branches of distribution networks, termed the flexible interconnection system. In practical scenarios, this system, adaptable to varying voltage requisites across distribution grid settings, proves effective across low, high, and medium voltage ranges [8]. Studies highlight the advantages of medium-voltage flexible interconnection systems in load equilibrium, power flow optimization, and substantial economic benefits (EBs) within the actual deployment of 6/10 kV distribution grids integrated with renewable energy sources [9]. Furthermore, to counteract the volatility of renewable energy output, susceptible to fluctuations due to environmental and regional influences, flexible interconnection systems can inject traditional electrical power into interconnected lines to offset reactive power in the distribution grid. This aids in achieving equilibrium control, harmonics management, bolstering system stability and power quality, enhancing power supply reliability, and reaping economic advantages [10]. In practical grid settings, flexible interconnection systems can alleviate transmission congestion in distribution grids through soft open points, providing continuous power compensation, addressing voltage coordination challenges, and furnishing dependable power flow regulation strategies [11]. Nonetheless, this approach may encounter constraints in terms of response time and longevity, notably when managing high-impact currents. Closed-loop soft open point interconnection systems can further assimilate the variability of impact currents, particularly in instances of precarious renewable energy supply. This framework enables prompt grid power adjustments through flexible power flow allocation and topological reconfiguration to exchange power within secure parameters for flexible power devices operating in high-penetration control modes [12, 13].

While the flexible interconnection system shows promise for distribution grids, its investment and operational expenses are notably high [14]. Recent research highlights that the cost management of these systems currently lags behind that of voltage regulation equipment. For small-scale renewable energy distribution setups, employing multiple voltage regulation mechanisms can achieve comparable reactive power compensation and voltage balance regulation. It is only in scenarios involving intricate distribution systems, diverse application contexts, and elevated operational costs that the adoption of the flexible interconnection system can yield significant economic advantages. To address this issue, scholars have introduced an integrated energy storage flexible interconnection system. By enabling network operators to dynamically align active grids with flexible electronic devices and incorporating renewable energy generation uncertainty through nonconvex nonlinear equations, this system adeptly handles power flow constraints and no-load losses of flexible electronic devices under powered circumstances [15]. Moreover, researchers have investigated innovative interconnection system voltage regulation tactics to achieve automatic voltage tuning at soft open points. This strategy initially employs a rolling optimization algorithm for device scheduling within the grid, succeeded by formulating a multi-objective optimization framework for interconnection systems to govern actual reactive power. Simulation outcomes underscore the substantial efficacy of this approach in stabilizing voltage fluctuations within active distribution grids [16]. The intricate fluctuations in power flow, as renewable energy generation adjusts within distribution grids, contribute to heightened uncertainty in power loads. Despite endeavors to optimize equipment setups, this intricacy persists in challenging the reliability of distribution grids.

The optimization of interconnection systems and the stochastic planning of flexible devices are recognized as effective strategies for managing uncertainties in distribution grids. Referenced scholars [17] argue that while enhancing the flexibility of active and reactive power flow can bolster grid stability, issues of coordination and allocation among interconnected flexible devices persist, including soft open points and network switches. To address these challenges, a stochastic scenario planning model is proposed, focusing on flexible devices and employing a second-order cone programming algorithm to synchronize renewable energy operations under varying loads. Nonetheless, this method shows diminished robustness postgrid fault recovery. Researchers delve into the self-healing capabilities of interconnection systems, outlining the flexible topology to enhance fault recovery and operational potential of multi-AC network closed-loop systems. Empirical findings demonstrate a significant improvement in self-healing capacity [18]. Furthermore, in managing fluctuations from renewable sources, experts propose an optimization method for a deeply penetrative interconnection scheme. Initially establishing a two-stage optimization model to reduce operational costs, they introduce a mixed-integer second-order cone programming approach for system deployment optimization, outperforming other methods [19]. Another research group presents a multistate interconnection strategy to optimize energy resource integration and enhance grid absorption capacity. By formulating a multi-objective function and using analytic hierarchy and entropy weight methods to determine weight factors, the deployment strategy ensures efficient power absorption control within safety constraints [20].

The deployment model and control strategies of the system have a substantial influence on line compensation and balance control within distribution grids. The complex interconnections within distribution grids give rise to various line challenges when transitioning across different application scenarios, encompassing discrepancies in interconnection line load rates, power distribution imbalances, and difficulties in main transformer load mode switching, among others. These diverse challenges necessitate tailored deployment and control approaches for flexible interconnection systems. To tackle the voltage disparity issue in feedback line interconnections for power transmission within low-voltage active distribution grids, scholars propose a multi-bus-based flexible interconnection tactic. This strategy establishes bus interconnections for multiterminal voltage source converters and integrates them with energy storage devices. Comparative simulation studies reveal the favorable economic outcomes of this method, surpassing other control strategies [21]. However, this approach may trigger feeder line overload dilemmas during flexible multistate and multimode transitions. To mitigate this issue, researchers introduce a distribution grid feeder line interconnection system control strategy founded on virtual synchronous generators to enhance load balancing capabilities during multimode operations. This method categorizes feeder line loads into distinct operational modes, scrutinizes the power switching logic of various operating modes to compute active power, and simulates feeder line load mode switches and main transformer overloads within a virtual distribution grid until a stable switching mode control strategy is iteratively ascertained. Empirical findings demonstrate significant effectiveness in practical implementations [22]. Addressing the deployment challenges of high-penetration flexible interconnection systems within high-power electronic contexts in energy systems, scholars advocate a novel perspective. They argue that the influence of new energy generation and flexible electronic devices on the stability of distribution grid interconnection systems under intense penetration should not solely focus on exploring equipment-level and system-level characteristics but should also devise deployment strategies for high-penetration models based on power system stability frameworks to effectively enhance dual high-penetration mode optimization strategies [23].

Hence, a novel approach is proposed by researchers utilizing a deep learning model founded on the transformer architecture to forecast the stability of short-term voltage imbalances within distribution grids. This methodology incorporates a Wasserstein generative adversarial network with gradient penalty to generate unstable abnormal data within distribution grids, conducts cluster analysis on load balancing and reactive power alterations observed within the interconnection system, and formulates a target loss function predicated on the correlation within a stable system state. Predicted values pertaining to line compensation and balance within distribution grids under varying new energy integrations are derived, subsequently optimizing the deployment mode of the interconnection system based on these forecasts. Through numerical assessments executed on the IEEE 39-node test system, the efficacy of this approach is demonstrated in effectively capturing the unstable characteristics of distribution grids and furnishing a dependable guide for enhancing the deployment of flexible interconnection systems [24].

While the aforementioned methods showcase a variety of optimization strategies for deploying flexible interconnected systems in diverse application scenarios, addressing power fluctuations from distributed sources over time remains marginally insufficient. To tackle these challenges, this study endeavors to capture typical scenario characteristics for site optimization using a deep neural network framework and to generate standard interconnected operational scenarios for assessing the planning impacts of various interconnection strategies and the grid's reliability. Furthermore, a comprehensive exploration of cost variables and operational planning strategies within the interconnected distribution network system is conducted over a temporal scale to enhance the optimization of flexible interconnection site layouts and deployment mode strategies for the new energy 6/10 kV medium-voltage distribution network.

Optimizing site placements for interconnected operational scenarios

To address the planning challenges within the flexible interconnection system of new energy 6/10 kV medium-voltage distribution grids, we have introduced generative adversarial networks [25]. The aim is to generate representative data samples for medium-voltage distribution grids and simulate the voltage difference challenges faced by various flexible interconnection device topologies. These generated data samples can also be used to evaluate the planning effectiveness and reliability of flexible devices in different interconnection strategies. Initially, historical data on new energy generation and distribution grid loads are preprocessed. The data are then categorized based on set flexible interconnection features and input into different layers of generative adversarial networks according to representative power consumption scenarios of distribution grid loads. Subsequently, through a two-stage interconnection system planning model, the first stage obtains the output results of the generative adversarial network. In the second stage, the output placement optimization results of the generative adversarial network are validated based on distribution grid power consumption scenarios. Finally, by mapping and correlating the outputs of the two stages, the optimal planning decisions for medium-voltage flexible interconnection site placements can be derived.

To explore optimal interconnection system planning solutions within the 6/10 kV medium-voltage system, a time-scale-based flexible voltage and reactive power control model for distribution grids has been introduced [26]. This model initially categorizes deterministic and stochastic scenarios based on real-world applications of flexible electronic devices. This division is usually based on the degree of uncertainty of potential operating conditions and environmental factors. Deterministic scenarios involve known and predictable parameters and conditions, such as stable weather patterns or reliable grid connections. On the other hand, uncertain scenarios include those that are affected by multiple factors and are difficult to accurately predict or control, such as unstable weather conditions or power system failures. Deterministic scenarios can be directly utilized for generating input feature datasets for generative adversarial networks. On the other hand, stochastic scenarios necessitate alignment with distribution grids featuring high levels of energy integration. Using a 15-minute cycle time scale as a standard for net power load circulation, the model tests their random fluctuation characteristics. Subsequently, based on the fundamental principles of voltage and reactive power control in the original distribution grid, the model segregates these scenarios into hourly and 15-minute time scales. Each time scale interfaces with the respective reactive power control voltage system to regulate operational voltages, mitigating voltage overload and imbalance issues [27]. During this process, it is crucial to consider the response speed compatibility when selecting voltage control devices for reactive power regulation. If the response is too sluggish, it may fail to manage voltage deviations caused by the integration of resistive components over time scales. Additionally, prolonged and high-frequency operations of flexible electronic devices can amplify voltage fluctuations, leading to situations where inverters in resistive components are unable to provide reactive power support over time scales.

When selecting reactive power control voltage regulation equipment, it is crucial to reduce the net power load value of the interconnected system operation scenario to ensure system safety. In our method design, we will optimize power distribution and scheduling, dynamically adjust reactive power, and introduce energy storage facilities to reasonably distribute loads in the interconnected system to avoid overload conditions, thereby reducing the net power load value. We will then implement an effective load management strategy to adjust the load according to system needs through dynamic load regulation and load-side response to reduce the net power load value and improve the flexibility of the system. In addition, we will establish a real-time monitoring system to detect system load conditions and power fluctuations in a timely manner, so as to take rapid response measures to adjust system operation and ensure that the net power load value is within a safe range.

To address this, we have taken into account the uncertainty in load demands for new energy generation within the 6/10 kV medium-voltage system. By selecting representative uncertain scenarios and employing stochastic planning, we have utilized a data-driven scenario method based on generative adversarial networks to capture patterns of new energy generation and load consumption. In this context, we define the new energy generation system as j with power generation denoted by |${x}_{jt}$| and |$j=1,2,...,N$|⁠. Within time frame |$t\in T$|⁠, a series of input samples z are generated by the model, all adhering to the real distribution |$Z\to{R}_Z$|⁠. By employing Gaussian resolution, we ensure that the real distribution |${R}_Z$| strictly follows the normal distribution |${R}_X$|⁠. Additionally, we introduce a generator neural network |$G\left(z;{\theta}^{(G)}\right)$| and a discriminator neural network |$D\left(x;{\theta}^{(D)}\right)$| to transform the input samples “z” into a form resembling the features of a normal distribution set |${R}_X$| simultaneously. The distributions |${\theta}^{(G)}$| and |${\theta}^{(D)}$| represent the weights of the generator and discriminator networks. During the training of the generator network, input sample z undergoes upsampling to reconstruct scenarios of new energy generation in the 6/10 kV medium-voltage system, ultimately mapping them into the following mathematical function expression process.

where the generated distribution |${R}_G$| also follows the normal distribution. When training the discriminator network, the input samples come from the feature set after downsampling from distribution |${\theta}^{(D)}$|⁠. The continuous value |${P}^{real}$| of the feature output after the discriminator network satisfies the true distribution |${R}_Z$|⁠, and the mapping relationship is shown below.

In the design of the generator's loss function |$Los{s}_G$|⁠, when the updated weights are introduced into the neural network, the generator features |$G\left(z;{\theta}^{(G)}\right)$| in the sample are randomly extracted, and a batch of real data is input into the discriminator |$D\left(\mathrm{x};{\theta}^{(D)}\right):\mathrm{x}$|⁠. From the design of the discriminator's loss function |$Los{s}_D$|⁠, the difference in features between the generated scene and the real scene can be shown, and the mathematical expression equation of the loss function can be expressed by the following formula.

To obtain a larger discriminator output feature, the generator network can be adjusted to minimize the discriminator feature |$-D\left(G\left(\bullet \right)\right)$|⁠. For the generator G, if the discriminator objective function D contains the minimized generator feature |$D\left(G\left(\bullet \right)\right)$|⁠, the deep neural network parameters and weights need to be adjusted to achieve balance. By fusing |$Los{s}_G$| and |$Los{s}_D$|⁠, a two-element maximum game value function |$V\left(G,D\right)$| [28] can be obtained, and its mathematical meaning is expressed as follows.

The maximal game value function encompasses the maximum–minimum objective function, which in mathematical terms can be interpreted as the dual value of the Wasserstein distance [29]. In the interconnected operation scenario of site optimization, the mapping of these two objective functions can be reflected through the strategy selection and game process in the game model. By modeling the strategy selection and game process of different participants (such as flexible electronic devices, energy systems, etc.), the maximum–minimum objective function can be balanced and optimized in the interconnected operation scenario. This mapping method helps to comprehensively consider the trade-offs of different objectives in the site optimization process to achieve the overall optimal interconnected operation effect. We employ the Wasserstein distance to quantify the distance between the distribution of generated scenario samples and real data, with the aim of mapping the diversity of scenarios related to power generation and load demands within the 6/10 kV medium-voltage system. Assuming the presence of two random variables, X and Y, in this mapping, the detailed computational formula can be expanded based on the following equation.

where |${f}_X$| and |${f}_Y$| represent the boundary distributions of variables X and Y, and |$\varLambda \left({f}_X,{f}_Y\right)$| represents the set of all possible true joint distributions. |$\operatorname{inf}$| represents the variable solution that conforms to the true joint distributions X and Y and can obtain the minimum Wasserstein distance. In the generative adversarial network structure of this study, the solution of Wasserstein distance can be converted into the distance between the true distribution |${R}_Z$| obtained by the generator and the true normal distribution |${R}_X$| [30]. The detailed mathematical principle is shown in the following formula.

When using the generative adversarial network to generate a new energy 6/10 kV medium-voltage power generation scenario, first set the learning rate, batch size, number of iterations, and weights of the generator and discriminator according to the forecast requirements. Then adjust the generation requirements of the scenario appropriately according to the convergence of |${\theta}^{(D)}$|⁠, and compare the Gaussian distribution |${P}_z(z)$| of the example batch |${\left\{\left({x}^{(i)},{y}^{(i)}\right)\right\}}_i^m$| with the historical data |${P}_{data}(x)$|⁠. According to the matching of the comparison results, the discriminator network is updated by gradient descent, and the corresponding generator network parameters matching the scenario are generated, as shown in the following formula:

where |$\nabla$| represents the gradient operator, |$\alpha$| represents the learning rate, |$RMS\Pr op$| represents the root mean square (RMS) propagation algorithm, |$clip$| represents the parameter self-regulation algorithm, c represents the gradient amplitude, x, y, z represent the feature dimensions of the medium voltage power generation scenario, and m represents the number of feature scenarios. The objective functions |$G(x)$| and |$D(x)$| of the generator and discriminator neural networks have parameters |${\theta}^{(G)}$| and |${\theta}^{(D)}$|⁠. Each network structure contains a multilayer perception layer, a convolution layer, a normalization layer, and a maximum pooling layer, and the activation function is kept as rectified linear unit (ReLU). The training of the generator adopts the iterative mode of gradient ascent, while the training of the discriminator adopts the iterative mode of gradient descent. The weight parameters of the discriminator training application need to be trimmed to avoid the gradient explosion problem. In the 6/10 kV medium voltage system, considering the time-varying characteristics of renewable energy generation and load demand, additional information can be added to the generator and discriminator designers as feature conditions, so as to ensure that the generated layout optimization scenario is close to the application scenario of real flexible electronic devices [31].

To ensure alignment with real-world scenarios, we commence by collecting and analyzing data, establishing appropriate time-series models to characterize the temporal aspects of renewable energy generation and load requirements. Subsequently, during the generation of deployment optimization scenarios, we consider the real-time and dynamic aspects of renewable energy generation and load demands. Through scenario validation and simulation, we assess the alignment between the generated optimized deployment scenarios and actual application scenarios. Finally, we flexibly adjust and optimize deployment strategies based on real-world conditions and feedback data, accommodating the time-varying nature of renewable energy generation and load demands. This iterative process guarantees system stability and performance across varying operational conditions.

Optimization of interconnected systems deployment

The optimization of interconnected systems deployment relies on a voltage-reactive power control algorithm that coordinates multiple time scales. Initially, a highly adaptable power system model is established, encompassing various electrical devices, points of renewable energy integration, and grid topological structures, among other elements. Additionally, this model must account for the characteristics of the flexible interconnected systems within the 6/10 kV medium voltage range and the voltage-reactive power control strategies. Assuming that this model minimizes power losses |${W}_m$| without violating voltage constraints, it can be expressed as shown in the following equation.

where the objective function of the coordinated multi-time-scale power system model can be composed of power loss cost and voltage deviation penalty cost. |${c}_t^E$| and |${c}^{penalty}$| represent the penalty cost of electricity price and voltage deviation, respectively. |${P}_{i,t}^{loss}$| represents the power loss in the operation cost, |${Q}_{i,t}^{CB}$| represents the connected reactive power of the capacitor bank node, and |${Q}_{i,t}^{Inv}$| represents the connected reactive power of the inverter node in the distribution network interconnection system. |${P}_{i,t}^{NE}$| represents the connected active power of renewable energy generation. |${Z}_{i,t}^{NE}$| represents the power capacity limit of the interconnection system connected to renewable energy generation. |$\sigma$| represents the average deviation of active power input, which can be adjusted according to the capacity configuration cost of the interconnection system. |${V}_{\mathrm{min}}$| and |${V}_{\mathrm{max}}$| represent the lower and upper limits of node voltage set by the interconnection system under the premise of safety constraints.

To predict the intergrid interaction cost, we describe the power flow of the distribution network by simplifying the DistFlow model [32], as shown in Fig. 1, where the power nodes match the nodes deployed in the interconnection system. The mathematical principle of the power conduction relationship between nodes is shown in the following formula:

where |${P}_{i,t}$| and |${Q}_{i,t}$| represent the active power and reactive power between nodes i and i + 1; |${V}_{i,t}$| represents the voltage amplitude of node i; and |${r}_i$| and |${x}_i$| represent the resistance and reactance of the interconnected system branch between nodes i and i + 1. |${P}_{i,t}^L$| and |${P}_{i,t}^G$| represent the active load and power generation of node i. |${Q}_{i,t}^L$|and |${Q}_{i,t}^G$| represent the reactive load and power generation of node i. When the interconnected system is deployed, the reactive power between nodes is balanced and guided according to the grid interaction cost, and the transmission voltage of adjacent nodes is controlled for overcharge.

Figure 1

Mapping relationship between interconnected system deployment and power grid flow.

Open in new tabDownload slide

Moreover, the optimization objectives of interconnected systems deployment encompass reducing system losses and costs, enhancing grid stability, and optimizing voltage quality. In response, we have devised a system coordination planning strategy that incorporates fast-response local control at multiple time scales and slow adjustments. This strategy comprises two interconnected stages: the first stage aims to determine optimal deployment decisions for interconnections, while the second stage focuses on coordinating the operational cost control issues across multiple time scales. These two-stage stochastic optimization models for interconnected systems deployment can be represented by the following mathematical equations.

where |${F}_1(x)$| and |${F}_2\left({\gamma}_w\right)$| are the objective functions of the first and second phases, x represents the input variables of the first phase, and |${\gamma}_w$| represents the input variables of the second phase. The input variables of different phases vary slightly according to the application scenarios of the interconnected system deployment. We classify these variables as shown in Table 1. w represents the uncertainty factor of the input variables of the second phase, |${p}_w$| represents the probability of different interconnected system deployment scenarios, and the operation scenario is generated by the generative adversarial model in Section 2 and satisfies |${\sum}_w{p}_w=1$|⁠.

After integrating the complete interconnected system deployment variables, the two-stage stochastic optimization model can be refined, and the detailed mathematical model is shown in the following equation.

where the investment cost in the interconnected system is allocated to the daily cost Y within the planning period with an interest rate r, |$\eta$| represents the daily cost recovery rate of the planned operation of the interconnected system, |${u}_i^{NE}$| represents the binary variable of the interconnected system installation cost decision, |${S}_{\mathrm{max}}^{NE}$| represents new energy generation maximum upper limit on running costs. In addition, the linear product of binary variable |${u}_i^{NE}$| and continuous variable |${S}_i^{NE}$| can be converted into a linear problem using a nonlinear mixed integer second-order cone programming algorithm [33]. For a small number of uncertain scenarios, it can be regarded as a random variable, with additional weights to eliminate overall impact. Treat uncertain scenarios as random variables and use nonlinear mixed integer second-order cone programming algorithms to transform them into linear problems. This can significantly improve computing efficiency when generating interconnected system operating scenarios and increase the feasible solution space for random problems, which makes it easier to find the optimal solution that meets various constraints and helps ensure that the generated interconnected system operation scenario is feasible. It can also ensure the stability of optimization results. In addition, ignoring the threshold value of the system deployed on the distributed nodes of flexible electronic devices will change due to the substantial increase in time scale. Usually, random planning can be simulated to improve this change by generating specific operating scenarios of the interconnected system based on the generative adversarial network in Section 2. However, when planning adaptive system deployment costs, the cost control rate can be significantly improved through iterative solution methods. Therefore, when deploying medium-voltage flexible systems, special attention needs to be paid to the stability and efficiency of the system, which helps to improve the accuracy of cost problem solving and reliability [34].

Case studies and discussions

To validate the effectiveness of the interconnected systems deployment model at the 6/10 kV medium voltage level, we selected the IEEE 37-node and 123-node distribution grid test systems [35], as depicted in Fig. 2. The power and voltage units in the case testing were standardized at 1 MVA and 12.66 kV, respectively. Throughout the case study, we assumed a planning horizon of 15 years. To explore the impact of uncertain scenarios on flexible interconnected systems deployment, we crafted 10 scenarios for flexible electronic device demands and 10 scenarios for 6/10 kV medium-voltage renewable energy generation. Building upon this foundation, we used generative adversarial networks to iteratively combine scenarios, resulting in 100 uncertainty scenarios. After scenario generation, the cost control of flexible interconnected system deployment is optimized according to load demand, and 45,000 load demand training samples and 106,300 6/10 kV medium voltage renewable energy generation training samples are generated. During the model training process, the training and testing samples were divided into 80% and 20% proportions. The training set comprised 36 000 load demand samples and 85 040 renewable energy generation samples, while the testing set consisted of 9000 load demand samples and 21 260 renewable energy generation samples. To ensure fair comparisons with similar methodologies, all datasets were sourced from public datasets [36]. Model training and testing were conducted on a computer equipped with an NVIDIA GeForce RTX 2080 Ti GPU, 11 GB RAM, 2.4 GHz Intel Core i7 processor, and 12 GB memory. The programming language used was Python 3.7, and PyCharm Community Edition 2020.3.2 served as the integrated development environment. Details regarding neural network training are outlined in Table 2.

Figure 2

Topology of IEEE 37-node and 123-node test systems.

Open in new tabDownload slide

During the training of the deep neural network, the difference in Wasserstein distance between the actual and predicted values is illustrated in Fig. 3. From the initial 0 to 100 iterations, the gap between the actual and predicted values consistently increases. Between 100 to 280 iterations, the Wasserstein distance of the actual values notably surpasses that of the predicted values. This discrepancy can be attributed to class imbalances or uneven sample distributions within the training data, leading to suboptimal model performance in certain scenarios and consequently increasing the Wasserstein distance. Furthermore, early convergence issues exist, including problems with vanishing gradients. To address these challenges, adjustments were made to the training strategies and model architecture, along with enhancing the balance of the dataset. Subsequently, from 280 to 500 iterations, the predicted values begin to exceed the actual values, indicating overfitting concerns at this stage where the model's generalization capabilities are weakened. To mitigate this, regularization techniques and dataset cleaning were implemented to reduce overfitting. Beyond 500 iterations, there is a notable increase in the alignment between actual and predicted values, signifying that the interconnected systems deployment scenario model has met the output requirements post 500 iterations.

Figure 3

Difference in Wasserstein distance between true value and predicted value during training.

Open in new tabDownload slide

To assess the predictive performance of the interconnected systems deployment scenario model, we conducted tests using the most representative 6/10 kV medium-voltage renewable energy generation scenarios and flexible equipment load demand scenarios. The results of these tests are illustrated in Fig. 4. Analysis of the renewable energy generation scenarios indicates that the overall trends and differences between predicted and actual values are minimal, aligning with the model's predictive requirements. Particularly noteworthy is the clustering effect during the 400–500 iteration period. However, there are delays in the predictions post 600 iterations, suggesting the need to tailor the choice of prediction results from different iteration periods based on specific application requirements. Evaluation of the flexible equipment load demand scenarios reveals significant discrepancies between predicted and actual values during the 0–200 and 700–800 iteration periods. This discrepancy stems from varying electricity demands of the flexible equipment, which adjust power input based on demand at the beginning and end of power consumption cycles. This leads to fluctuations in the distribution of flexible equipment load demand data, resulting in notable differences between predicted and actual values during these periods. However, between 200 and 700 iterations, the alignment between predicted and actual values improves significantly, showcasing consistent trends that meet the model's predictive standards. In practical applications, it may be advisable to focus on the predictions from the middle segment and disregard those from the initial and final periods, as they are closer to reality. These observations underscore the overall effectiveness of the interconnected systems deployment scenario generation model established through generative adversarial networks.

Figure 4

Difference between the actual value and predicted value of the interconnected system deployment scenario model.

Open in new tabDownload slide

Furthermore, in validating the effectiveness of the interconnected systems deployment scenario generation model, we not only assessed the model's training time, convergence performance, generation quality, and generalization capabilities, but also conducted a safety evaluation by examining the voltage heat maps of the generated scenarios, as depicted in Fig. 5. The results indicate that the original scenarios' voltage heat maps exhibit a chaotic and disordered state, which can compromise the stability of the distribution network's voltage regulation system, leading to system overload, decreased efficiency, and shortened equipment lifespan. In contrast, the optimized voltage heat maps generated by our model display systematic variations. These optimized maps regulate voltage concentrations based on a 24-hour cycle, allowing for concentrated voltage adjustment during peak periods to stabilize the distribution network's voltage regulation system, and reducing the load on the distribution network system during low voltage periods to prolong equipment lifespan. Additionally, all voltages remain within safe operating limits, ensuring the safety of the interconnected systems deployment.

Figure 5

Comparison of voltage heat maps before and after generating the scenario.

Open in new tabDownload slide

Optimizing interconnected system deployments requires validation of the model's effectiveness within specific application scenarios. To this end, we hypothesized nine scenarios to evaluate the new energy penetration rate (NEPR), source loading ratio (SLR), voltage offset (VO), number of voltage overruns (NVO), and network energy loss (NEL) during interconnected system deployments. The results are presented in Table 3. When selecting nine flexible interconnection system application scenarios in the distribution network, considerations typically encompass factors such as representativeness, practical requirements, complexity, data availability, and technical challenges. Our chosen scenarios primarily involve load forecasting, distributed energy management, energy storage system optimization, microgrid operational control, electric vehicle charging management, intelligent distribution device control, power market transaction optimization, fault detection, and recovery, among others.

The results in Table 3 reveal that as the penetration rate of renewable energy generation increases, the number of voltage exceedances initially rises, then decreases to zero, while NELs show an increasing trend. The source load ratio was evaluated using the median and quartiles (0.25, 0.50, 0.75) as criteria. It is noted that at a source load ratio of 0.75, the number of voltage exceedances can be minimized, whereas at a ratio of 0.25, NELs can be kept to a minimum. In practical application scenarios, if substantial investments are made in voltage regulation equipment for the distribution network, the impact of voltage exceedances on interconnected system deployments can be disregarded in favor of selecting a planning solution with lower NELs. This may involve choosing a scenario with a penetration rate below 30% and a source load ratio of 0.25. Conversely, in situations where voltage regulation capabilities are limited, leading to issues like overvoltage and overload, greater attention should be placed on the voltage exceedance metric. In such cases, a scenario with a penetration rate exceeding 70% and a source load ratio of 0.75 would be more suitable. To further evaluate the energy conversion efficiency in various flexible device application scenarios and the voltage index in the power flow of the distribution network, we narrowed the time frame to 24 hours and calculated the proportional weights of different scenarios for each hour. The test results are illustrated in Fig. 6.

Figure 6

Energy conversion efficiency and power flow voltage trends for different interconnected system application scenarios.

Open in new tabDownload slide

The results from the above figure indicate that in different application scenarios, the original energy conversion efficiency was below 80%. After optimization through interconnected system deployment, the energy conversion efficiency improved to the range of 85% to 95%. Additionally, there was a reduction in power flow voltages; during working hours from 8 a.m. to 6 p.m., the power flow voltages decreased by 10 p.u., effectively alleviating the voltage load pressure on the distribution network. It is evident that interconnected system deployment optimization can effectively enhance the energy conversion efficiency of the distribution network, appropriately schedule network loads, and increase the stability and safety of the distribution network system. Furthermore, the optimization of interconnected system deployment aims to improve the EBs of distribution network operations. To assess the daily EB, EB rates (EBR), reduction in operating costs (ROC), reduction in interconnection costs (RIC), reduction in capacity configuration costs (RCCC), and safety levels (SLs) in the IEEE 37-node and 123-node test systems, we evaluated the EBs of the proposed model and method using five indicators. The SLs range from A to E, with A indicating the highest risk and E representing the lowest safety risk. The results are presented in Table 4.

The results in Table 4 reveal that, in the context of the IEEE 37-node test system, our proposed optimization method for interconnected system deployment achieved an average daily EB of $449.09 across all application scenarios, marking a 46.95% increase in returns. Operating costs, interconnection costs, and capacity configuration costs decreased by 27.01%, 52.67%, and 25.29%, respectively, with a lower level of safety risk observed. In contrast, for the IEEE 123-node test system, the average daily EB across all scenarios amounted to $1551.27, reflecting a 61.50% increase in returns. However, the reductions in operating costs, interconnection costs, and capacity configuration costs were only 9.98%, 34.27%, and 14.25%, respectively. This discrepancy can be attributed to the larger scale and more intricate network topology of the 123-node system compared to the 37-node system. Therefore, the potential EBs of optimizing the 123-node system may be higher, as enhancements in larger-scale systems tend to yield more substantial cost improvements. These experimental findings underscore that the interconnected system deployment optimization proposed in this study delivers significant comprehensive EBs in typical operational scenarios. It demonstrates the capacity to notably enhance the safety performance and reliability of distribution networks.

Conclusion

This study delves into the advantages of employing flexible interconnected system deployment strategies within new energy 6/10 kV medium-voltage distribution networks, presenting an optimization approach for crafting and implementing adaptable interconnected system scenarios to elevate system reliability and cost-effectiveness. Initially, the investigation scrutinizes the prerequisites for closed-loop operations in new energy distribution networks alongside standard operational scenarios. It introduces a flexible interconnection operational scenario generation model grounded in generative adversarial networks to mitigate voltage disparities inherent in flexible interconnection device topologies. This model assesses various interconnection strategies and network reliability by generating representative scenarios. Subsequently, a coordinated interconnected system deployment model is outlined, harmonizing multiple time scales. Through a two-stage deployment of distinct economic and safety variables, this model ensures the operational reliability and economic advantages of the distribution network within the interconnected system. Validation utilizing the IEEE dual-node test system demonstrates that our method generates scenarios closely resembling real-world instances, showcasing robust convergence and generalization. The voltage heat maps of all generated scenarios exhibit systematic variations within secure load constraint boundaries. Moreover, this study provides scientific references for determining NEPRs and source load ratios for diverse deployment scenarios. Ultimately, the proposed optimization of interconnected system deployment within typical operational scenarios underscores significant comprehensive EBs, with the potential to notably enhance the safety and reliability of distribution networks.

Author contributions

Kan Feng (Conceptualization [equal], Data curation [equal], Formal analysis [equal], Funding acquisition [equal], Investigation [equal], Methodology [equal], Project administration [equal], Resources [equal], Software [equal], Supervision [equal], Validation [equal], Visualization [equal], Writing—original draft [equal], Writing—review & editing [equal]), Shijin Xin (Conceptualization [equal], Data curation [equal], Funding acquisition [equal], Investigation [equal], Methodology [equal], Project administration [equal], Resources [equal], Software [equal], Supervision [equal], Writing—original draft [equal], Writing—review & editing [equal]), Pan Zhang (Conceptualization [equal], Formal analysis [equal], Investigation [equal], Methodology [equal], Resources [equal], Software [equal], Supervision [equal], Visualization [equal], Writing—original draft [equal], Writing—review & editing [equal]), Jian Yang (Data curation [equal], Formal analysis [equal], Investigation [equal], Methodology [equal], Project administration [equal], Resources [equal], Supervision [equal], Writing—original draft [equal], Writing—review & editing [equal]), Changhui Fan (Data curation [equal], Formal analysis [equal], Methodology [equal], Project administration [equal], Validation [equal], Visualization [equal], Writing—original draft [equal], Writing—review & editing [equal]), and Chong Chen (Data curation [equal], Software [equal], Validation [equal], Visualization [equal], Writing—original draft [equal]).

Funding

This work is funded by the State Grid Gansu Electric Power Company 2024 Science and Technology Project: Research on Key Technologies and Typical Configurations of Flexible Interconnection of Medium-voltage Distribution Networks (No. B72703241203).

References