Power distribution network and virtual power plant interaction evaluation method fusing entropy evaluation method and LSTM

Abstract

The invention discloses a power distribution network and virtual power plant interaction evaluation method fusing an entropy evaluation method and an LSTM, and the method comprises the steps: constructing a multi-dimensional evaluation index system based on a hierarchical structure; the hierarchical structure comprises a target layer, a performance layer, an element layer and an index layer; the index layer is composed of evaluation indexes representing all elements; generating an initial weight of the evaluation index based on an improved entropy method; according to the improved entropy method, an interval decision matrix is formed based on probability distribution of analysis interval values to capture uncertainty of the evaluation indexes; optimizing the initial weight of the evaluation index based on an LSTM network to obtain an optimized weight; and based on the optimized weight and the corresponding evaluation index, interactive evaluation of the power distribution network and the virtual power plant is carried out. According to the method, the LSTM network and the entropy evaluation method are fused, and accurate and adaptive evaluation of power distribution network-virtual power plant interaction is realized through a multi-dimensional index system and dynamic weight optimization.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart grid, in particular to a power distribution network and virtual power plant interaction evaluation method combining entropy value method and LSTM. BACKGROUND

[0002] At present, the world is actively promoting sustainable development and climate action, which has prompted renewable energy to rapidly penetrate the power system, and carbon neutrality has become the core driving force for the transformation of energy systems. Under this trend, the traditional centralized power generation mode is gradually being replaced by a distributed and flexible mode, which is characterized by the widespread application of distributed energy resources (Distributed Energy Resource, hereinafter referred to as DER). Among many DER-related technologies, virtual power plants (Virtual Power Plant, hereinafter referred to as VPP) aggregate and manage dispersed energy resources such as solar photovoltaics, wind turbines, and energy storage systems, enabling them to operate in a coordinated and consistent manner. Although VPPs perform well in improving system flexibility and adaptability, their interaction with the power distribution network has also raised a series of technical, economic, and operational challenges.

[0003] State Grid Corporation of China and China Southern Power Grid (hereinafter referred to as CSG) are promoting the transformation to a sustainable energy system. State Grid has invested heavily in smart grid technology and distributed energy integration. Similarly, CSG focuses on flexible resource management and demand-side response, committed to enhancing the resilience and efficiency of the power grid. These initiatives demonstrate the important role of VPPs in modernizing the power system and optimizing the use of renewable energy.

[0004] As the penetration rate of renewable energy continues to rise, bidirectional power flow and voltage fluctuations have occurred in the power distribution network, which requires changes in the planning and operation of the power distribution network. Unlike traditional power plants, renewable energy has intermittency and geographical dispersion, which can affect voltage characteristics, protection systems, and equipment life. From an economic perspective, although the integration of VPPs reduces the cost of upgrading the power grid, it complicates the cost-benefit analysis and market mechanisms. From an operational perspective, the increasing variability and uncertainty in power generation and consumption patterns place higher demands on advanced monitoring, control, and forecasting systems.

[0005] Currently, traditional evaluation methods such as Failure Mode and Effects Analysis (FMEA) and Cost-Benefit Analysis (CBA) have limitations in describing the multi-dimensional and dynamic characteristics of the interaction between the distribution network and the VPP. Although the Analytic Hierarchy Process (AHP) and fuzzy evaluation methods integrate multiple indicators, they have weak ability to handle dynamic and uncertain data. Machine learning techniques, especially Long Short-Term Memory (LSTM), have shown great advantages in handling sequence data and capturing time-dependent relationships, but further exploration is needed in this specific field.

[0006] Therefore, the present application is proposed. SUMMARY

[0007] The technical problem to be solved by the present application is that the existing evaluation methods for the interaction between the distribution network and the virtual power plant cannot comprehensively consider multi-dimensional evaluation indicators, have limitations in multi-dimensional and dynamic characteristics, are disturbed by subjective factors in weight distribution, and cause low evaluation accuracy. The present application aims to provide an evaluation method for the interaction between the distribution network and the virtual power plant that integrates entropy method and LSTM. The method integrates LSTM network and entropy method, optimizes the multi-dimensional index system and dynamic weight, realizes accurate and adaptive evaluation of the interaction between the distribution network and the virtual power plant (VPP), and is suitable for complex power grid environments with high proportion of new energy access, especially in power system planning and operation decision-making under the goal of carbon neutrality.

[0008] The present application is realized by the following technical solutions: In a first aspect, the present application provides an evaluation method for the interaction between the distribution network and the virtual power plant that integrates entropy method and LSTM. The method includes: Based on the hierarchical structure, a multi-dimensional evaluation index system is constructed. The hierarchical structure includes the target layer, the performance layer, the element layer, and the index layer. The index layer is composed of evaluation indicators representing each element. Based on the improved entropy method, the initial weight of the evaluation index is generated. The improved entropy method forms an interval decision matrix based on the probability distribution of the analysis interval value to capture the uncertainty of the evaluation index. Based on the LSTM network, the initial weight of the evaluation index is optimized to obtain the optimized weight. Based on the optimized weight and the corresponding evaluation index, the interaction between the distribution network and the virtual power plant is evaluated.

[0009] Further, the evaluation indexes include a first index, a second index, and a third index; The first index is an economic dimension index, which is used to evaluate the cost-effectiveness and financial feasibility of integrating the virtual power plant VPP into the power distribution network; the economic dimension index includes dynamic indexes and static indexes; The second index is a safety dimension index, which is used to evaluate the reliability and safety of the power distribution network and the virtual power plant VPP during operation; the safety dimension index includes load indexes and reliability indexes; The third index is a flexibility dimension index, which is used to measure the adaptability of the power system to changing conditions; the flexibility dimension index includes transmission-distribution coordination indexes and source-grid-load-storage coordination indexes.

[0010] Further, the dynamic indexes include internal rate of return, net present value, and payback period, and the static indexes include unit asset power supply capacity, unit asset power supply, and unit asset power supply capacity; The load indexes include satisfaction rate, load shedding probability, load shedding frequency, and expected load shedding value, and the reliability indexes include system average interruption length, system average interruption frequency, customer average interruption length, and power supply availability rate; The transmission-distribution coordination indexes include voltage grade line length ratio, voltage grade substation quantity ratio, and substation capacity ratio; the source-grid-load-storage coordination indexes include interactive reserve capacity ratio, elastic load ratio, and self-generation self-use power gap rate.

[0011] Further, the improved entropy method includes: The evaluation indexes are represented as interval values to form an interval decision matrix; The interval decision matrix is standardized to obtain a standardized interval decision matrix; According to the standardized interval decision matrix, the information entropy of each evaluation index is calculated; According to the information entropy, the weight of the corresponding evaluation index is calculated and normalized; According to the information entropy of each evaluation index and the corresponding normalized weight, the total interval entropy of the power system is calculated.

[0012] Further, the standardization processing of the interval decision matrix includes: For benefit-type indexes in the evaluation indexes, the first standardization formula is used to standardize the interval decision matrix; the first standardization formula is wherein, represents the maximum value function, is the interval value represented by the evaluation index; For cost-type indexes in the evaluation indexes, the second standardization formula is used to standardize the interval decision matrix; the second standardization formula is wherein, min function.

[0013] Further, the initial weight of the evaluation index is optimized based on the LSTM network to obtain an optimized weight, including: The LSTM network is constructed, and the LSTM network is trained and optimized; The training and optimization include: inputting the training data into the LSTM network for forward propagation to calculate the output; and using a dynamic weighted loss function combining the mean square error and the mean absolute error to calculate the loss between the predicted weight and the actual weight, and calculating the gradient through back propagation, using the Adam optimizer to update the network parameters to obtain the optimized weight.

[0014] Further, the LSTM network is also verified; The verification includes: comparing the optimized weight with the weight determined by the expert scoring to verify the optimized weight of the LSTM network; and verifying the LSTM network once in a preset number of training cycles to fine-tune the hyperparameters and prevent overfitting.

[0015] In the second aspect, the application further provides a power distribution network and virtual power plant interactive evaluation system fusing the entropy value method and the LSTM, which comprises: An evaluation index construction unit is configured to construct a multi-dimensional evaluation index system based on a hierarchical structure; the hierarchical structure comprises a target layer, a performance layer, an element layer, and an index layer; the index layer is composed of evaluation indexes representing each element; An initial weight generation unit is configured to generate an initial weight of the evaluation index based on an improved entropy value method; the improved entropy value method is to form an interval decision matrix based on the probability distribution of the analysis interval value to capture the uncertainty of the evaluation index; A weight optimization unit is configured to optimize the initial weight of the evaluation index based on the LSTM network to obtain an optimized weight; An interactive evaluation unit is configured to perform interactive evaluation of the power distribution network and the virtual power plant based on the optimized weight and the corresponding evaluation index.

[0016] Further, the improved entropy value method includes: The evaluation index is represented as an interval value to form an interval decision matrix; The interval decision matrix is standardized to obtain a standardized interval decision matrix; The information entropy of each evaluation index is calculated according to the standardized interval decision matrix; The weight of the corresponding evaluation index is calculated according to the information entropy, and the weight is normalized; The total interval entropy of the power system is calculated according to the information entropy of each evaluation index and the corresponding normalized weight.

[0017] In a third aspect, the present application further provides a computer-readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the power distribution network and virtual power plant interactive evaluation method of the fusion entropy method and LSTM.

[0018] Compared with the prior art, the present application has the following advantages and beneficial effects: 1. The power distribution network and virtual power plant interactive evaluation method of the fusion entropy method and LSTM, the present application method fuses LSTM network and entropy method, realizes accurate and adaptive evaluation of power distribution network-virtual power plant (VPP) interaction through multi-dimensional index system and dynamic weight optimization, is suitable for complex power grid environment with high proportion of new energy access, and has important application value in power system planning and operation decision under the carbon neutralization target.

[0019] 2. The power distribution network and virtual power plant interactive evaluation method of the fusion entropy method and LSTM, the present application constructs a comprehensive multi-dimensional evaluation index system covering three core dimensions of economy, safety and flexibility, accurately and systematically quantifies various complex factors involved in the interaction process between power distribution network and virtual power plant, provides solid and detailed data support for power system planning and operation, effectively fills the gap of the existing evaluation system in multi-dimensional consideration, and helps decision makers to comprehensively examine system performance from different angles.

[0020] 3. The power distribution network and virtual power plant interactive evaluation method of the fusion entropy method and LSTM, the interval decision matrix is formed based on the probability distribution of the analysis interval value to improve the entropy weight method, when determining the initial weight, not only the dispersion degree of each index is considered, but also the correlation between them and the ideal solution is fully evaluated, so that a more objective and reasonable initial weight is generated. This improvement effectively avoids the interference of subjective factors on weight distribution in the traditional method, ensures the scientificity and reliability of the weight, lays a good initial condition for the subsequent optimization process based on deep learning, and helps to improve the accuracy and reliability of the whole evaluation system.

[0021] 4、The power distribution network and virtual power plant interactive evaluation method fusing entropy value method and LSTM of the application, for the optimization link of the LSTM network, a series of frontier improvement measures are integrated, including accurate evaluation of feature importance, which can effectively screen out the most influential feature information for power distribution network and virtual power plant interactive evaluation; data augmentation technology is used to expand the training data set, improve the model's adaptability to different scenarios; a deep integrated architecture is constructed to enhance the model's ability to mine complex data patterns; an attention mechanism is introduced to enable the model to focus on key information and improve the sensitivity to important features; an adaptive dynamic weight adjustment strategy is implemented to flexibly balance global trend fitting and local detail refinement according to the training process, accelerating model convergence; and multi-objective optimization regularization is used to prevent overfitting and improve the model's generalization performance. The comprehensive use of these optimization methods significantly enhances the LSTM network's ability to capture complex dynamic relationships in historical data, making it more adaptable and generalizable in different operating conditions. BRIEF DESCRIPTION OF DRAWINGS

[0022] The drawings described herein are used to provide further understanding of the embodiments of the application, form a part of the application, and do not constitute a limitation on the embodiments of the application. In the drawings: Figure 1 The flowchart of the power distribution network and virtual power plant interactive evaluation method fusing entropy value method and LSTM of the application; Figure 2 The schematic diagram of the multi-dimensional evaluation index system of the application; Figure 3 The loss curve during training of the application; Figure 4 The system structure block diagram of the power distribution network and virtual power plant interactive evaluation system fusing entropy value method and LSTM of the application. DETAILED DESCRIPTION

[0023] To make the purpose, technical scheme and advantages of the application clearer and more explicit, the application will be further described in detail below in combination with examples and drawings, the illustrative embodiments of the application and their descriptions are used to explain the application, and do not constitute a limitation on the application.

[0024] The existing power distribution network and virtual power plant interactive evaluation method cannot comprehensively consider multi-dimensional evaluation indexes, has limitations in multi-dimension and dynamic characteristics, and is also affected by subjective factors in weight distribution, causing problems such as low evaluation accuracy.

[0025] To address the above issues, this invention designs an interactive evaluation method for distribution networks and virtual power plants that integrates entropy value method and LSTM. This method incorporates a multi-dimensional evaluation index system and employs a weighted method based on entropy value method and LSTM, enabling dynamic adjustment of weights to reflect actual operating conditions. This comprehensive framework integrates economic, safety, and flexibility factors, providing decision-makers with practical suggestions for optimizing system performance. Leveraging historical data and machine learning techniques, this method offers an effective solution for power system planning and operation, particularly in the context of achieving carbon neutrality and renewable energy integration in the power grid.

[0026] Example 1 like Figure 1 As shown, this invention integrates the entropy method with LSTM for interactive evaluation of distribution networks and virtual power plants. The method includes: Step 1: Based on the hierarchical structure, construct a multi-dimensional evaluation index system; the hierarchical structure includes the target layer, performance layer, element layer, and index layer; the index layer consists of evaluation indicators representing each element. Existing technologies have established various power system assessment methodologies. In safety assessment, methods such as the N-1 criterion and load shedding probability are used to evaluate system stability and power supply quality; economic assessment utilizes methods such as net present value and internal rate of return to measure project cost-effectiveness. This invention, focusing on carbon emissions and renewable energy integration capacity, introduces indicators such as renewable energy integration rate and carbon emission intensity. Simultaneously, power quality indicators and equipment utilization efficiency indicators are used to measure system operating efficiency, while indicators such as energy storage capacity and load response capability are used to assess system flexibility and coordination.

[0027] At the standards and specifications level, internationally, standards such as ISO 14001 and IEC 62271 provide guidelines for assessing reliability, economy, and environmental performance; domestically, national standards such as GB / T 2589 and GB / T 12325 focus on assessment criteria for economy, power quality, and renewable energy inclusion; and in terms of industry standards, standards such as DL / T 783 and DL / T 1365 emphasize assessment support for system flexibility and coordination.

[0028] like Figure 2 As shown, this invention constructs a comprehensive evaluation index system for the interaction between distribution networks and virtual power plants (VPPs). This system adopts a hierarchical structure, including an objective layer, a performance layer, an element layer, and an index layer. The objective layer aims to create an evaluation method for the interaction between distribution networks and virtual power plants. The performance layer includes three key elements: economy, safety, and flexibility. The element layer further refines into specific evaluation elements, such as static / dynamic economy, power load safety, reliability, transmission and distribution performance, and source-grid-load-storage coordination. The index layer consists of specific indicators representing each element.

[0029] The evaluation indexes include a first index, a second index and a third index; Specifically, the first index is an economic dimension index, which is used to evaluate the cost-effectiveness and financial feasibility of integrating the virtual power plant VPP into the power distribution network; the economic dimension index includes dynamic indexes and static indexes; the dynamic indexes include an internal rate of return (IRR), a net present value (NPV) and a payback period (PP); the net present value (NPV) refers to the difference between the present values of cash inflows and outflows within a certain period of time, and can be used to judge the economic feasibility of a project. The internal rate of return (IRR or r) refers to the discount rate that makes the net present value (NPV) of all cash flows of a project equal to zero, and can reflect the investment efficiency and profitability of the project, and its calculation formula is derived from the net present value equation. The payback period (PP) refers to the time required to recover the initial investment cost. The shorter the payback period, the lower the investment risk and the better the liquidity of funds.

[0030] The static indexes include a unit asset power supply capacity (UAPSC), a unit asset power supply volume (UAPSV) and a unit asset power supply capacity (UAPSCap); the unit asset power supply capacity (UAPSC) is used to measure the power supply capacity of each unit asset, and reflects the utilization efficiency of the asset. The unit asset power supply volume (UAPSV) represents the power supply volume of each unit asset, and shows the improvement of power supply efficiency. The unit asset power supply capacity (UAPSCap) reflects the enhancement degree of the power supply capacity of each unit asset.

[0031] The above indexes are represented as follows: (1) (2) (3) (4) (5) (6) Wherein, represents the net cash flow at time ; is the internal rate of return; is the initial investment cost; is the inflation rate at time ; is the tax rate at time ; is the internal rate of return at time ; is the total power supply volume at time ; is the asset value at time ; depreciation rate at time asset utilization rate at time total energy supply at time asset value at time transmission efficiency at time energy loss rate at time technology progress factor at time policy support coefficient at time

[0032] Specifically, the second index is a safety dimension index for evaluating the reliability and safety of the distribution network and the virtual power plant (VPP) in operation; the safety dimension index includes a load index and a reliability index; the load index covers indicators such as a satisfaction rate, a load shedding probability, a load shedding frequency, and an expected load shedding value; and the reliability index includes indicators such as a system average outage duration, a system average outage frequency, a customer average outage duration, and a power supply availability rate.

[0033] The satisfaction rate, that is, the N-1 rule satisfaction rate (N-1SR), is used to evaluate the power supply continuity when a single component fails. The load shedding probability (LSP) measures the power supply safety risk by calculating the probability that the system cannot meet the load demand. The load shedding frequency (LSF) reflects the frequency at which the system needs to shed load to maintain stability. The expected load shedding value (ELCV) is used to evaluate the load support capacity of the system during failure. The system average outage duration (SAOD) reflects the power restoration capability of the system after outage. The system average outage frequency (SAOF) indicates the degree of influence of power supply continuity by outage. The customer average outage duration (CAOD) evaluates the power supply service quality from the customer's perspective. The power supply availability rate (PSAR) directly reflects the reliability of power supply.

[0034] The above indexes are represented as follows: (7) (8) (9) (10) (11) (12) (13) (14)​​​​​​​​ where, is the total number of components (devices or elements that make up various parts of the power system); is the criticality coefficient of component , reflecting its importance in the system; is the failure probability of component ; is an indicator function that takes the value 1 if the condition is met, and 0 otherwise; is the set of failed components in scenario ; is the set of normally operating components in scenario ; is the probability that component can withstand additional load in scenario ; is the duration of the th load curtailment event; is the restoration rate of the system after the th load curtailment event; is the time since the start of the th load curtailment event; is the total time period; is the priority weight of load ; is the load curtailment value of load ; is the duration of the th outage; is the restoration rate after the th outage; is the time since the start of the th outage; is the load affected during the th outage; is the aging coefficient of component ; is the operating time of component ; is the number of outages experienced by customer ; is the duration of each outage experienced by customer ; is the restoration rate of customer ; is the time since the start of each outage experienced by customer ; is the availability state of the system at time (taking the value 1 for normal operation and 0 for outage); is the time interval.

[0035] Specifically, the third index is a flexibility dimension index, which is used to measure the adaptability of the power system to changing conditions; the flexibility dimension index includes a transmission-distribution coordination index and a source-grid-load-storage coordination index. The transmission-distribution coordination index includes a voltage level line length ratio, a voltage level substation quantity ratio, and a substation capacity ratio; the source-grid-load-storage coordination index includes an interactive reserve capacity ratio, an elastic load ratio, and a self-generation self-use electricity gap rate.

[0036] The voltage level line length ratio (VLLR) is used to evaluate the coordination between lines of different voltage levels in the power grid. The voltage level substation quantity ratio (VLSNR) analyzes the layout of substations. The substation capacity ratio (SCR) reflects the degree of improvement of power supply capacity. The interactive reserve capacity ratio (IRCR) is used to evaluate the flexibility support capability of the system. The elastic load ratio (ELR) measures the demand side management capability of the system. The self-generation self-use electricity gap rate (GCEDR) reflects the energy utilization efficiency.

[0037] The above indexes are represented as follows: (15) (16) (17) (18) (19) (20) wherein, is the number of voltage levels; is the line length at the voltage level is the total line length; is the aging coefficient of the line at the voltage level is the operation time of the line at the voltage level is the number of substations at the voltage level is the total number of substations; is the geographical diversity coefficient of the substations at the voltage level is the average distance between substations at the voltage level is the capacity of the substations at the voltage level is the total capacity of the substations; is the redundancy coefficient of the substations at the voltage level is the total capacity of the substations; is the redundancy coefficient of the substations at the voltage level is the total capacity of the substations; is the redundancy coefficient of the substations at the voltage level is the total capacity of the substations; is the redundancy coefficient of the substations at the voltage level is the total capacity of the substations; is the redundancy coefficient of the substations at the voltage level is the total capacity of the substations; Redundancy level of the substation; It refers to the quantity of energy; It is energy The spare capacity; It is the total backup capacity of the system; It is energy The dynamic adjustment coefficient; It is energy Changes in reserve capacity; It is the number of load points; It is the load point Elastic load at the location; It is the total elastic load; It is a load The response speed coefficient; It is a load Response time; It refers to the number of power generation units; It is a power generation unit The power shortage; It is the total power generation; It is a power generation unit The intermittent nature of power generation; It is a power generation unit Conversion efficiency.

[0038] Step 2: Generate the initial weights of the evaluation indicators based on the improved entropy method. The improved entropy method is based on analyzing the probability distribution of interval values ​​to form an interval decision matrix to capture the uncertainty of the evaluation indicators. Specifically, it is reflected in the following: First, interval values ​​are used in step (1) because the traditional entropy method is usually calculated based on a single point of fixed value (such as a specific numerical value). The default data is accurate and without fluctuations, which makes it difficult to reflect the inherent uncertainty of evaluation indicators (such as cost, reliability parameters, etc.) in the actual power system (such as measurement error, data fluctuation, prediction deviation, etc.). This invention adopts the interval method to solve the above problems of the traditional entropy method. Second, in step (2), benefit-type and cost-type indicators are distinguished and calculated using corresponding formulas.

[0039] In this embodiment, the improved entropy method in step 2 includes: (1) Express the evaluation indicators as interval values ​​to form an interval decision matrix; The evaluation metrics are expressed as range values ​​to account for uncertainty. Let... For the set of evaluation indicators, For the set of measurement objects. Each measurement object Indicators Represented as an interval Forming an interval decision matrix as follows: (twenty one) (2) Standardizing the interval decision matrix to obtain a standardized interval decision matrix; To eliminate the dimension effect and scale difference, the decision matrix is standardized. The standardization of the interval decision matrix includes: For the benefit type indexes (internal rate of return IRR, net present value NPV, unit asset power supply capacity UAPSC, unit asset power supply volume UAPSV, unit asset power supply capacity UAPSCap, system N-1 criterion satisfaction ratio N-1SR, power supply availability rate PSAR, voltage grade line length ratio VLLR, voltage grade substation quantity ratio VLSNR, substation capacity ratio SCR, interactive reserve capacity ratio IRCR, and elastic load ratio ELR) in the evaluation indexes, the first standardization formula is used to standardize the interval decision matrix. The first standardization formula is: (22) wherein, represents the maximum value function, is the interval value represented by the evaluation index; For the cost type indexes (payback period PP, load shedding probability LSP, load shedding frequency LSF, expected load shedding value ELCV, system average outage duration SAOD, system average outage frequency SAOF, customer average outage duration CAOD, and gap rate of self-generating and self-using electricity GCEDR) in the evaluation indexes, the second standardization formula is used to standardize the interval decision matrix. The second standardization formula is: (23) wherein, represents the minimum value function.

[0040] (3) According to the standardized interval decision matrix, the information entropy of each evaluation index is calculated; The information entropy of the i-th index is calculated as follows: (24) wherein, , Here, represents the probability distribution of the standardized value. The information entropy reflects the information content of the index .

[0041] (4) According to the information entropy, the weight of the corresponding evaluation index is calculated and the weight is normalized; The weight is calculated according to the information entropy: (25) The weight reflects the weight of the i-th index in the evaluation index system. The relative importance of each indicator. To ensure the sum of the weights is 1, normalization is performed: (26) (5) Calculate the total interval entropy of the power system based on the information entropy of each evaluation index and the corresponding normalized weight.

[0042] The total entropy of the power system is calculated using the following formula: (27) The total entropy of the power system integrates the uncertainty contributions of all indicators, providing a comprehensive measure of system uncertainty.

[0043] To calculate the 20 indicators, data sources included internal company documents, data released by official agencies, and technical documents provided by equipment manufacturers. Data collection methods included extracting data from financial systems, monitoring systems, and policy documents, or through expert evaluation and historical data analysis.

[0044] For missing values, linear interpolation is used to maintain the temporal continuity of the data, while forward or backward imputation is used to fill in missing values ​​in static data with few missing values. To handle outliers, the interquartile range (IQR) method is used to identify and remove data points outside the reasonable range based on the upper and lower quartiles. Finally, the aforementioned 20 indicators are calculated according to formula (1-20). Subsequently, 3000 sets of indicators are obtained, as shown in Table 1 below.

[0045] Table 1

[0046] Based on the improved entropy weight method, the calculations are shown in Table 2: Table 2

[0047] Step 3: Optimize the initial weights of the evaluation indicators based on the LSTM network to obtain the optimized weights; and perform interactive evaluation of the distribution network and virtual power plant based on the optimized weights and the corresponding evaluation indicators.

[0048] In this embodiment, the initial weights of the evaluation metrics are optimized based on an LSTM network to obtain optimized weights, including: Construct an LSTM network, and train and optimize the LSTM network; Training and optimization include: inputting training data into an LSTM network for forward propagation to calculate the output; calculating the loss between the predicted weights and the actual weights using a dynamic weighted loss function that combines mean squared error and mean absolute error; calculating the gradient through backpropagation; updating the network parameters using the Adam optimizer to obtain the optimized weights.

[0049] As a further implementation, it also includes verifying the LSTM network; The verification includes: comparing the optimized weight with the weight determined by the expert scoring to verify the optimized weight of the LSTM network; and verifying the LSTM network once in a preset number of training cycles to fine-tune the hyperparameters and prevent overfitting.

[0050] In specific implementation, the specific process of step 3 is as follows: (1) Training evaluation index To optimize the training performance, the present application introduces a dynamic weighted loss function, which combines mean square error (MSE) and mean absolute error (MAE), whose expressions are shown in formulas (27) and (28) respectively. The designed weighted loss function (29) aims to adaptively balance the trade-off between global trend fitting and local detail refinement during training. The dynamic weight factor It plays a key role in this mechanism, which changes with the training round, making the error minimization priority consistent with the model learning progress.

[0051] (27) (28) (29) (30) Initially, it is set to a higher value of 0.9, giving priority to MSE, prompting the model to preferentially capture the global structure of the data. As the training progresses, it is gradually linearly decayed to 0.1 within T rounds. This gradual change enables the model to gradually shift its focus to MAE, which is more sensitive to local anomalies and data detail patterns. By systematically rebalancing the contributions of MSE and MAE, the dynamic weighted loss function ensures efficient convergence of the model while maintaining robustness to the inherent non-stationary characteristics of power system data.

[0052] This adaptive weighting strategy is particularly advantageous for training complex models such as LSTM networks, as they must learn complex temporal dependencies and nonlinear correlations from sequential data. Dynamic adjustment Not only does it accelerate convergence, but it also enhances the model's generalization ability to unseen operating scenarios, making it very suitable for dynamic power system assessment.

[0053] (2) Test evaluation index In the power grid industry, due to the complexity and sensitivity of system operation, any method or model must undergo rigorous expert evaluation before practical application to ensure the safety, reliability, and compliance of power grid operation. Unlike other fields, the power grid industry cannot test and adjust methods during application; instead, it must first verify them through expert scoring or validation. During the testing phase, the weights optimized by the LSTM network are compared with the weights determined by expert scoring to verify the accuracy and practical applicability of the LSTM network optimization weights, ensuring that they match the priorities and complexity of actual power grid operation.

[0054] The expert scoring process involves multiple rounds of feedback to ensure the scientific validity and consistency of the weighting allocation. Specifically, firstly, experts with extensive experience and expertise in the power system field are selected. Then, the significance, calculation method, and role of each evaluation indicator in the power grid assessment are explained to the experts. Each expert scores each indicator independently, and consensus is reached through multiple rounds of discussion and feedback to form a group average weight.

[0055] These expert weights serve as a benchmark for evaluating the accuracy of weight optimization in LSTM networks. Based on predicted weights... Validating weights with experts relative error threshold Calculate the match rate. For each metric... Relative error ( ) is defined by formula (31). If If the match is found, then it is considered a match. As shown in formula (32), the overall matching rate ( ) is the ratio of the number of indicators that meet this condition to the total number of indicators, where This represents the number of matching metrics, indicating the total number of metrics.

[0056] (31) (32) (3) LSTM training and validation The dataset obtained in Step 1 is divided into a training set, a validation set, and a test set in a 70:15:15 ratio. The training set is used to train the LSTM network, enabling it to learn latent patterns and relationships in the data. The validation set is used to evaluate model performance during training, helping to fine-tune hyperparameters and prevent overfitting. The remaining 15% of the data is reserved as the test set, serving as an independent dataset to evaluate how well the model matches expert opinions in terms of accuracy.

[0057] The hyperparameters for the LSTM network training are set as follows. In the training phase, the network model is initialized with the entropy-generated weights. The training data is input into the LSTM for forward propagation to calculate the output. A dynamic weighted loss function combining mean squared error (MSE) and mean absolute error (MAE) is adopted to calculate the loss between the predicted weights and the actual weights. The gradients are calculated by backpropagation, and the network parameters are updated using the Adam optimizer. The model is validated once every 5 training cycles to fine-tune the hyperparameters and prevent overfitting.

[0058] The hyperparameters are set as follows: learning rate: 0.001; number of hidden units: 30; batch size: 16; number of training cycles: 50; validation frequency: validation once every 5 training cycles; initial dynamic weight: 0.9; minimum dynamic weight: 0.1.

[0059] After 5000 iterations on the training data set, the loss curve is shown in Figure 3 Figure 3 The loss curve (Loss Value) of the training data set and the validation data set during the training of the model for 50 epochs is shown in

[0060] The minimum difference between the training loss and the validation loss at the end of the training further indicates that the model has achieved a good balance between bias and variance, ensuring reliable performance on known and new data. Overall, the loss curve proves the successful training of the LSTM network and its ability to accurately evaluate the interaction between the distribution network and the VPP. The final training results obtained by the LSTM network training are shown in Table 3.

[0061] Table 3

[0062] In this embodiment, the interaction between the distribution network and the virtual power plant is evaluated based on the optimized weights and corresponding evaluation indicators, including: (1) Normalization and reuse of evaluation indicators: Reuse the standardized results of the indicators in the improved entropy method (i.e., the standardized interval decision matrix elements ​), to ensure the index value in the interval [0, 1] (eliminate the dimensional effect). If there are new evaluation samples, they need to be reprocessed according to the same standardization rules (benefit-type index uses formula (22), cost-type index uses formula (23)) to ensure data consistency.

[0063] (2) Comprehensive evaluation score calculation: based on the weights optimized by LSTM (“Maximum weight” in Table 3) and the standardized index values, the comprehensive evaluation score of the interaction between the distribution network and the VPP is calculated using the weighted summation method, as follows:

[0064] Wherein: is the total number of evaluation indexes (20 in the document); is the weight of the th index after LSTM optimization; is the value of the th index after standardization; is the comprehensive evaluation score, with a value range of [0, 1], and the higher the score, the better the interaction effect.

[0065] (3) Calculation and analysis of hierarchical dimension scores: to refine the evaluation results, the hierarchical scores are calculated according to the “economic dimension, safety dimension, and flexibility dimension”, as follows: Economic dimension score: economic index ; Safety dimension score: safety index ; Flexibility dimension score: flexibility index ; (4) Evaluation level division and decision-making suggestions: according to the comprehensive score and the hierarchical scores, the interaction effect is divided into 4 levels, corresponding to different decision-making suggestions, as shown in Table 4: Table 4

[0066] (5) Dynamic evaluation and iterative optimization: due to the time sequence of the interaction between the distribution network and the VPP (such as load fluctuation and renewable energy output change), dynamic evaluation needs to be based on the time sequence processing capability of the LSTM network: New index data (such as real-time load satisfaction rate and unit asset power supply) are collected every hour / day, and steps (1) to (4) are repeated to calculate the real-time comprehensive score; By comparing the score changes in different time periods (such as peak and valley periods, seasonal alternation), the fluctuation rules of the interaction effect are identified, for example: If the flexibility The dynamic evaluation results are fed back to the LSTM network periodically (e.g., monthly), and the optimized weights are retrained to adapt the evaluation model to the long-term changes in the system operating state.

[0067] (6) Verification and correction of evaluation results: compare the comprehensive score and the layered score with the expert evaluation results, calculate the relative error (refer to formula 31). If the error , it indicates that the evaluation result is reliable; if the error , the index standardization process or the LSTM weight optimization parameters (such as adjusting the decay rate of the dynamic weighted loss function) need to be checked back, until the error meets the requirements.

[0068] The specific implementation is as follows: Firstly, the present application constructs a comprehensive and systematic multi-dimensional evaluation index system, which is the basis of the entire evaluation method and provides a scientific and comprehensive quantitative basis for subsequent analysis and decision-making. The index system adopts a hierarchical structure, covering the target layer, performance layer, element layer and index layer, and comprehensively covers the main influencing factors of the interaction between the distribution network and the virtual power plant, and conducts in-depth analysis from the three key dimensions of economy, safety and flexibility, accurately and systematically quantifying various complex factors involved in the interaction process between the distribution network and the virtual power plant.

[0069] After constructing a comprehensive multi-dimensional evaluation index system, how to objectively and reasonably determine the weight of each index becomes a key problem. The traditional method often has strong subjectivity in weight distribution, and the improved entropy method of the present application effectively solves this problem.

[0070] As an index to measure the degree of data disorder, the lower the entropy value, the higher the information content and the more important the decision-making significance. The improved entropy weight method of the present application captures uncertainty by analyzing the probability distribution of interval values, providing an objective weight allocation scheme for complex decision-making problems. This method is particularly suitable for evaluating the interaction between the distribution network and the VPP, because such evaluation involves a large amount of dynamic and uncertain data.

[0071] Specifically, the present application first represents the evaluation index as an interval value to consider uncertainty and forms an interval decision matrix; then standardizes the decision matrix to eliminate dimensional effects and scale differences; then calculates the information entropy of each evaluation index to reflect the information content of the evaluation index; calculates the weight according to the information entropy and normalizes the weight to ensure that the sum of the weights is 1; finally, the total interval entropy of the system is calculated to comprehensively measure the uncertainty of the system.

[0072] The initial weight generated by the improved entropy weight method comprehensively considers the dispersion degree of each index and the correlation to the ideal solution, avoids the interference of subjective factors on weight distribution, and ensures the scientificity and reliability of the weight. This improves the better initial condition for the subsequent optimization process based on deep learning, and helps to improve the accuracy and reliability of the entire evaluation system.

[0073] On the basis of determining the initial weight, the application further optimizes the evaluation index by using LSTM to fully excavate the complex dynamic relationship in the historical data and improve the performance and adaptability of the evaluation model.

[0074] First, in the training of evaluation indexes, a dynamic weighted loss function is introduced, which combines mean square error (MSE) and mean absolute error (MAE). This dynamic weighted loss function can adaptively balance the trade-off between global trend fitting and local detail refinement during training, and through the evolution of the dynamic weight factor with the training round, the error minimization priority and the model learning progress are consistent. At the beginning, it is set to a higher value of 0.9, giving priority to MSE to prompt the model to preferentially capture the global structure of the data. As the training progresses, it is gradually linearly decayed to 0.1 within T rounds, so that the model gradually shifts its focus to the MAE which is more sensitive to local anomalies and data detail patterns. This adaptive weighting strategy is particularly beneficial for training complex models such as LSTM networks, as it can accelerate convergence while enhancing the model's robustness to the inherent non-stationary characteristics of power system data.

[0075] Secondly, in the test evaluation index stage, considering the particularity of the power grid industry, the application adopts a strict expert evaluation process. The weights optimized by the LSTM model are compared with the weights determined by expert scoring to verify the accuracy and practical applicability of the LSTM optimized weights. The expert scoring process involves multiple rounds of feedback to ensure the scientificity and consistency of the weight distribution. Based on the relative error threshold between the predicted weight and the expert verified weight, the matching rate is calculated to ensure that the LSTM optimized weight matches the priority and complexity of the actual power grid operation. This step not only verifies the accuracy of the model, but also enhances its credibility and reliability in practical application.

[0076] Finally, during the LSTM training and validation process, the application carefully sets the hyperparameters to ensure efficient training and optimization of the model. The model is initialized using entropy-generated weights, and the training data is input into the LSTM for forward propagation to calculate the output. A dynamic weighted loss function combining mean squared error (MSE) and mean absolute error (MAE) is used to calculate the loss between the predicted weights and the actual weights, and the gradients are calculated through backpropagation to update the network parameters using the Adam optimizer. The model is validated every 5 training cycles to fine-tune the hyperparameters and prevent overfitting. Through these rigorous training and validation steps, the LSTM model ensures high efficiency and accuracy in capturing the dynamic relationship between the power distribution network and the virtual power plant.

[0077] The experimental verification results of the application strongly prove the outstanding performance of the method in practical application scenarios. The method can dynamically adjust the weights of evaluation indicators with very high precision and reliability, and is highly consistent with the actual interaction between the power distribution network and the virtual power plant. Specifically, 95.0% of the relative errors of the indicators are controlled within 3%, and when the error threshold is relaxed to 5%, all evaluation results of the indicators meet the standard requirements. This shows that the method has achieved remarkable results in improving evaluation accuracy and adaptability, effectively solving the defects of traditional evaluation methods in capturing multi-dimensional dynamic characteristics, providing strong data-driven support for power system planning and operation decision-making, and strongly promoting the coordinated development of power distribution networks and virtual power plants in the process of new energy integration and carbon neutralization target realization. It has significant practicality and innovation, and is expected to be widely applied and popularized in the field of smart grid technology.

[0078] Embodiment 2 As shown in Figure 4 The difference between this embodiment and embodiment 1 is that this embodiment provides a power distribution network and virtual power plant interaction evaluation system that integrates entropy value method and LSTM. This system corresponds to the power distribution network and virtual power plant interaction evaluation method of embodiment 1 that integrates entropy value method and LSTM. The system includes: An evaluation index construction unit for constructing a multi-dimensional evaluation index system based on a hierarchical structure. The hierarchical structure includes a target layer, a performance layer, an element layer, and an index layer. The index layer is composed of evaluation indicators representing each element. An initial weight generation unit for generating initial weights of evaluation indicators based on an improved entropy value method. The improved entropy value method forms an interval decision matrix based on the probability distribution of the analysis interval value to capture the uncertainty of the evaluation indicators. A weight optimization unit for optimizing the initial weights of the evaluation indicators based on the LSTM network to obtain optimized weights. An interaction evaluation unit for performing interaction evaluation of the power distribution network and the virtual power plant based on the optimized weights and the corresponding evaluation indicators.

[0079] As a further implementation, the improved entropy method comprises: expressing the evaluation indexes as interval values to form an interval decision matrix; performing standardization processing on the interval decision matrix to obtain a standardized interval decision matrix; calculating the information entropy of each evaluation index according to the standardized interval decision matrix; calculating the weight of the corresponding evaluation index according to the information entropy and performing normalization processing on the weight; calculating the total interval entropy of the power system according to the information entropy of each evaluation index and the corresponding normalized weight.

[0080] The execution process of each unit can be performed according to the power distribution network and virtual power plant interactive evaluation method of the fusion entropy method and LSTM in Embodiment 1, and will not be described again in this embodiment.

[0081] Meanwhile, the application further provides a computer-readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the power distribution network and virtual power plant interactive evaluation method of the fusion entropy method and LSTM.

[0082] Those skilled in the art should understand that the embodiments of the application can be provided as a method, a system, or a computer program product. Therefore, the application can be in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the application can be in the form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0083] The application is described with reference to flowcharts and / or block diagrams according to the method, device (system), and computer program product of the embodiments of the application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to produce a machine, so that the instructions executed by the computer or other programmable data processing devices produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The device that implements the functions specified in one or more flows and / or blocks Figure 1 The device that implements the functions specified in one or more flows and / or blocks

[0084] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the flow Figure 1 The functions of a flow or multiple flows and / or a block or multiple blocks in conjunction with the disclosed methods can be implemented on a single device or distributed across several devices. Figure 1

[0085] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions that are executed on the computer or other programmable apparatus provide steps for implementing the functions specified in the flow Figure 1 The functions of a flow or multiple flows and / or a block or multiple blocks in conjunction with the disclosed methods can be implemented on a single device or distributed across several devices. Figure 1

[0086] The above detailed description merely describes the specific implementation of the application, and is not used to limit the protection scope of the application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the application shall be included in the protection scope of the application.​​

Claims

1. A power distribution network and virtual power plant interaction evaluation method fusing entropy value method and LSTM, characterized in that, The method comprises: constructing a multi-dimensional evaluation index system based on a hierarchical structure; the hierarchical structure comprises a target layer, a performance layer, an element layer, and an index layer; the index layer is composed of evaluation indexes representing each element; generating initial weights of the evaluation indexes based on an improved entropy method; the improved entropy method is to form an interval decision matrix based on the probability distribution of the analysis interval value to capture the uncertainty of the evaluation indexes; optimizing the initial weights of the evaluation indexes based on an LSTM network to obtain optimized weights; and performing interactive evaluation of the power distribution network and the virtual power plant based on the optimized weights and the corresponding evaluation indexes.

2. The power grid and virtual power plant interactive evaluation method of claim 1, wherein The evaluation indexes include first indexes, second indexes, and third indexes; The first indexes are economic dimension indexes, used to evaluate the cost-effectiveness and financial feasibility of integrating the virtual power plant into the power distribution network; the economic dimension indexes include dynamic indexes and static indexes; The second indexes are safety dimension indexes, used to evaluate the reliability and safety of the power distribution network and the virtual power plant during operation; the safety dimension indexes include load indexes and reliability indexes; The third indexes are flexibility dimension indexes, used to measure the adaptability of the power system to changing conditions; the flexibility dimension indexes include transmission and distribution coordination indexes and source-grid-load-storage coordination indexes.

3. The power grid and virtual power plant interactive evaluation method of claim 2, wherein, The dynamic indexes include internal rate of return, net present value, and investment payback period, and the static indexes include unit asset power supply capacity, unit asset power supply quantity, and unit asset power supply capacity; The load indexes include satisfaction rate, load shedding probability, load shedding frequency, and expected load shedding value, and the reliability indexes include system average outage duration, system average outage frequency, customer average outage duration, and power supply availability; The transmission and distribution coordination indexes include voltage grade line length ratio, voltage grade substation quantity ratio, and substation capacity ratio; the source-grid-load-storage coordination indexes include interactive reserve capacity ratio, elastic load ratio, and self-generation and self-use power gap rate.

4. The power grid and virtual power plant interactive evaluation method of claim 1, wherein The improved entropy method comprises: representing the evaluation indexes as interval values to form an interval decision matrix; standardizing the interval decision matrix to obtain a standardized interval decision matrix; calculating the information entropy of each evaluation index according to the standardized interval decision matrix; calculating the weight of the corresponding evaluation index according to the information entropy and normalizing the weight; calculating the total interval entropy of the power system according to the information entropy of each evaluation index and the corresponding normalized weight.

5. The power grid and virtual power plant interactive evaluation method of claim 4, wherein, The standardization processing of the interval decision matrix comprises: For the benefit type index in the evaluation index, the interval decision matrix is standardized by using a first standardization formula; the first standardization formula is wherein, represents a maximum value function, is an interval value represented by the evaluation index. For the cost type index in the evaluation index, the interval decision matrix is standardized by using a second standardization formula; the second standardization formula is wherein, represents a minimum value function.

6. The power grid and virtual power plant interactive evaluation method of claim 1, wherein optimizing the initial weights of the evaluation indexes based on an LSTM network to obtain optimized weights, comprising: constructing an LSTM network and training and optimizing the LSTM network; The training and optimization include: inputting the training data into the LSTM network for forward propagation to calculate the output; and using a dynamic weighted loss function combining mean square error and mean absolute error to calculate the loss between the predicted weight and the actual weight, and calculating the gradient through back propagation, using the Adam optimizer to update the network parameters to obtain the optimized weights.

7. The power grid and virtual power plant interactive evaluation method of claim 6, wherein, It also includes verifying the LSTM network; The verification comprises: comparing the optimized weight with the weight determined by expert scoring to verify the optimized weight of the LSTM network; and verifying the LSTM network once per preset number of training cycles to fine-tune the hyperparameters and prevent overfitting.

8. The power distribution network and virtual power plant interactive evaluation system fusing entropy value method and LSTM, characterized in that, The system comprises: The evaluation index construction unit is configured to construct a multi-dimensional evaluation index system based on a hierarchical structure; the hierarchical structure comprises a target layer, a performance layer, an element layer, and an index layer; the index layer is composed of evaluation indexes representing each element; The initial weight generation unit is configured to generate initial weights of the evaluation indexes based on an improved entropy value method; the improved entropy value method is used to capture the uncertainty of the evaluation indexes by forming an interval decision matrix based on the probability distribution of the analysis interval value; The weight optimization unit is configured to optimize the initial weights of the evaluation indexes based on an LSTM network to obtain optimized weights; The interactive evaluation unit is configured to perform interactive evaluation of the power distribution network and the virtual power plant based on the optimized weights and the corresponding evaluation indexes.

9. The power grid and virtual power plant interactive evaluation system of claim 8, wherein, The improved entropy value method comprises: The evaluation indexes are represented as interval values to form an interval decision matrix; The interval decision matrix is standardized to obtain a standardized interval decision matrix; The information entropy of each evaluation index is calculated based on the standardized interval decision matrix; The weight of the corresponding evaluation index is calculated based on the information entropy, and the weight is normalized; The total interval entropy of the power system is calculated based on the information entropy of each evaluation index and the corresponding normalized weight.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 9. The computer program, when executed by a processor, implements the power distribution network and virtual power plant interactive evaluation method based on the fusion entropy value method and the LSTM according to any one of claims 1 to 7.

Images