Power distribution network and flexible resource mutual regulation and control system for source-load space-time distribution prediction

Abstract

The invention discloses a power distribution network and flexible resource mutual regulation and control system for source load space-time distribution prediction, and relates to the technical field of intelligent regulation and control of a power system, and the system comprises an acquisition preprocessing module, a prediction modeling module, an analysis and evaluation module, a regulation and control optimization module and a monitoring correction module. According to the method, the convolutional neural network and the graph neural network are fused to realize joint modeling of source-load spatial-temporal characteristics, and a power distribution network intelligent prediction and regulation system with a closed-loop self-correction function is constructed in combination with a flexible mathematical model of renewable energy and an energy storage device and an optimal regulation algorithm based on A3C reinforcement learning. According to the method, the multi-source data processing quality and prediction precision can be effectively improved, quantitative utilization and dynamic coordination of flexible resources are realized, and the stability and self-adaptive capability of system operation are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent regulation and control of power systems, and in particular to a source-load space-time distribution prediction distribution network and flexible resource mutual regulation and control system. BACKGROUND

[0002] With the increasing proportion of new energy power generation, especially the large-scale access of distributed power sources such as photovoltaic and wind power in distribution networks, the operation challenges faced by traditional distribution networks are increasingly prominent. On the one hand, the output of distributed power sources is greatly affected by weather, season and other factors, resulting in intensified node power imbalance, line flow fluctuation and voltage deviation; on the other hand, although flexible resources such as energy storage devices and adjustable loads in distribution networks have certain regulation capabilities, there is a lack of systematic collaborative regulation mechanism in the prior art, resulting in the failure of flexible resources to fully exert their optimization potential.

[0003] The existing distribution network dispatching method mainly relies on static control based on prediction or traditional optimization algorithm, and the response to the uncertainty of future load and distributed power output is insufficient, which is prone to problems such as large power balance deviation, frequent voltage out-of-limit, high operation cost, etc. In addition, the existing method lacks a closed-loop correction mechanism in real-time operation, and cannot dynamically compensate for prediction deviations or sudden load changes, limiting the improvement of the intelligent level of distribution networks.

[0004] Therefore, there is an urgent need for a source-load space-time distribution prediction distribution network and flexible resource mutual regulation and control system that can realize real-time perception, accurate prediction, dynamic optimization and closed-loop correction, to realize efficient collaboration of distributed power sources, energy storage and adjustable loads, and to improve the stability and reliability of distribution network operation. SUMMARY

[0005] In view of the above problems, the present application is proposed.

[0006] Therefore, the problem to be solved by the present application is how to accurately predict the source-load space-time distribution of each node in the case of large-scale access of distributed power sources in distribution networks, and make full use of flexible resources such as energy storage devices and adjustable loads to realize comprehensive optimization of power balance and voltage stability in the whole network.

[0007] To solve the above technical problems, the present application provides the following technical solutions:

[0008] In a first aspect, the present application provides a source-load space-time distribution prediction distribution network and flexible resource mutual regulation and control system, comprising: a collection preprocessing module for collecting multi-source data of a distribution network, performing anomaly detection, missing repair and normalization processing on the multi-source data, and constructing a high-quality time series data set;

[0009] The prediction modeling module is used to construct a source-load intelligent prediction model based on the high-quality time-series dataset using a convolutional neural network model, perform feature dimensionality reduction and noise reduction, and output the temporal feature representation of each node through a fully connected layer; and fuse the time-series features of the convolutional neural network with the spatial features of the graph structure to obtain the source-load spatiotemporal distribution prediction results of each node.

[0010] The analysis and evaluation module is used to establish a mathematical model of the flexibility of renewable energy and energy storage devices based on the spatiotemporal distribution prediction results of source and load, and to calculate the available up- and down-adjustment capacity of each node, forming a time series flexibility capability curve.

[0011] The regulation and optimization module is used to construct a multi-objective comprehensive optimization model. It adopts a reinforcement learning algorithm based on asynchronous advantage Actor-Critic to solve the regulation strategy. The strategy optimization and value assessment are realized through multi-threaded parallel interaction between the Actor and Critic networks.

[0012] The monitoring and correction module is used to monitor the operating status of the distribution network in real time, calculate the deviation rate between the measured operating data and the predicted source load results, and automatically trigger the closed-loop correction mechanism when the prediction deviation of any node exceeds the set threshold.

[0013] As a preferred embodiment of the power distribution network and flexible resource interoperability system for source-load spatiotemporal distribution prediction described in this invention, the following steps are included: collecting multi-source data from the power distribution network and performing anomaly detection, missing data repair, and normalization processing on the multi-source data, including:

[0014] Anomaly detection is performed on the collected raw dataset using a multi-layered discrimination strategy: a sliding window difference method is used to detect abrupt changes in the time series; the Z-score statistical method is used to identify deviations in spatial nodes; the Pearson correlation coefficient method is used to detect logical anomalies in the correlation between devices; when anomalies are detected, the occurrence node and time are recorded, and corrections are made using adjacent time mean interpolation or multivariate regression prediction.

[0015] To address data loss caused by sensor failures or communication packet loss, a two-layer repair strategy based on temporal correlation and spatial correlation is employed: at the temporal level, linear interpolation, exponential smoothing, or LSTM time prediction are used to compensate for short-term missing data; at the spatial level, weighted averaging of adjacent node data or graph convolutional interpolation based on network topology is used. After the repaired dataset passes consistency verification, it enters the normalization stage.

[0016] As a preferred embodiment of the power distribution network and flexible resource interoperability system for source-load spatiotemporal distribution prediction described in this invention, the system comprises: based on the high-quality time-series dataset, a convolutional neural network model is used to construct a source-load intelligent prediction model for feature dimensionality reduction and denoising, and a fully connected layer is used to output the time-series feature representation of each node, including:

[0017] Convolutional operations are performed on time-dimensional data in convolutional layers to extract local temporal features and trend changes;

[0018] The convolutional output is processed by a pooling layer to reduce dimensionality, filtering out high-frequency noise and retaining the main changing features, resulting in a smoothed node feature sequence. The feature sequence output from the pooling layer is then input into a fully connected layer, where weight parameters are used to comprehensively map the features at different time steps, generating a temporal feature representation for each node. Through end-to-end training, historical power and output data of the nodes are used as supervision signals to jointly optimize the parameters of the convolutional and fully connected layers. The calculation formula for the convolution operation is as follows:

[0019] ;

[0020] Where f is the activation function of the convolutional layer, and k is the number of convolutional kernels. The weight matrix of the convolution kernel. Let b be the i-th input time series data, where i ranges from 1 to n, n is the number of time series data, and b is the bias value. For the corresponding The output value of the convolution kernel.

[0021] As a preferred embodiment of the power distribution network and flexible resource interoperability system for source-load spatiotemporal distribution prediction described in this invention, the method involves fusing the time-series features of a convolutional neural network with the spatial features of a graph structure to obtain the source-load spatiotemporal distribution prediction results for each node, including:

[0022] The fused node joint features are input into the output layer to generate load and distributed energy output prediction results for each node in the future prediction period, forming a node-level source-load spatiotemporal distribution prediction sequence; the time step information and node identifier are retained in the output prediction sequence.

[0023] As a preferred embodiment of the power distribution network and flexible resource interoperability system for source-load spatiotemporal distribution prediction described in this invention, wherein: based on the source-load spatiotemporal distribution prediction results, a flexible mathematical model for renewable energy and energy storage devices is established, including:

[0024] Based on the projected renewable energy output, the equivalent controllable portion of renewable energy is determined according to the system's set confidence capacity, while the remaining portion is considered as uncontrollable load and its associated flexibility. The flexibility of renewable energy supply and consumption is as follows:

[0025] ;

[0026] ;

[0027] in, The flexibility that renewable energy can provide For the confidence capacity of renewable energy, Let t represent the amount of flexible resources consumed by renewable energy at time t. For the moment of renewable energy contribution;

[0028] Based on the charge / discharge state and capacity limitations of the energy storage device, calculate the up and down flexibility that the energy storage device can provide at time t:

[0029] ;

[0030] ;

[0031] in, , Energy storage devices Provides flexibility in both positive and negative directions at all times; , This refers to the maximum charging and discharging power of the energy storage device. for Energy stored in a time-of-use energy storage device; , These are the upper and lower limits of the energy storage capacity, respectively. This is the scheduling time interval.

[0032] As a preferred embodiment of the distribution network and flexible resource interoperability system for source-load spatiotemporal distribution prediction described in this invention, the following steps are taken: calculating the available upward and downward adjustment capacity of each node to form a time-series flexibility capability curve, including:

[0033] The upward and downward capacity adjustments of each node at each moment are arranged in chronological order to form node-level upward and downward adjustment capacity curves. The source device ID, calculation assumptions, and confidence index of each time step are recorded in the capacity curves. Based on the time series capacity curves of each node, a network-wide flexibility adjustment capacity matrix is ​​generated.

[0034] As a preferred embodiment of the distribution network and flexible resource interoperability system for source-load spatiotemporal distribution prediction described in this invention, the system comprises: constructing a multi-objective integrated optimization model; employing a reinforcement learning algorithm based on asynchronous advantage Actor-Critic to solve the control strategy; and achieving strategy optimization and value assessment through multi-threaded parallel interaction between the Actor and Critic networks, including:

[0035] An optimization function is established with the objectives of power balance, voltage stability, and system operation economy. At the same time, it is combined with the line capacity, voltage constraints, and the adjustment limits of energy storage and adjustable load to form joint constraints at the node level and system level. The Actor network is used to generate node control strategies, that is, output adjustment actions under given conditions.

[0036] Critic networks are used to evaluate the value of actions taken and calculate reward signals;

[0037] A multi-threaded approach is adopted, with each thread independently running a pair of Actor-Critic networks to interact with the environment to generate policies and value assessments. After completing a certain training step, each thread asynchronously uploads its parameters to the global network for synchronous updates.

[0038] As a preferred embodiment of the distribution network and flexible resource interoperability system for source-load spatiotemporal distribution prediction described in this invention, the following is included: real-time monitoring of the distribution network operating status; calculation of the deviation rate between measured operating data and predicted source-load results; and automatic triggering of a closed-loop correction mechanism when the prediction deviation of any node exceeds a set threshold, including: Calculate the deviation rate between the predicted results and the actual results:

[0039] ;

[0040] in, Let be the measured active power value of the nth node. This represents the predicted active power value for the nth node.

[0041] If ΔP n If the prediction result of that node is found to be too biased when the threshold is exceeded, a closed-loop correction mechanism is automatically triggered; if If the value is 0, then a safety correction strategy is adopted.

[0042] In a second aspect, embodiments of the present invention provide a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of the distribution network and flexible resource interoperability system for source-load spatiotemporal distribution prediction as described in the first aspect of the present invention are implemented.

[0043] Thirdly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of the distribution network and flexible resource interoperability system for source-load spatiotemporal distribution prediction as described in the first aspect of the present invention are implemented.

[0044] The beneficial effects of this invention are as follows: By fusing convolutional neural networks and graph neural networks to achieve joint modeling of the spatiotemporal characteristics of source and load, and combining a flexible mathematical model of renewable energy and energy storage devices with an optimization and control algorithm based on A3C reinforcement learning, a smart prediction and control system for distribution networks with closed-loop self-correction function is constructed. This method can effectively improve the quality and prediction accuracy of multi-source data processing, realize the quantitative utilization and dynamic coordination of flexible resources, and improve the stability and adaptability of system operation. Attached Figure Description

[0045] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Fig. 1 A schematic diagram of the structure of a distribution network and a flexible resource interoperability system for predicting the spatiotemporal distribution of source and load; Fig. 2 This is a computer equipment diagram for a distribution network and flexible resource interoperability system for predicting the spatiotemporal distribution of source and load. Detailed Implementation

[0047] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0048] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0049] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0050] Example 1

[0051] Reference Figs. 1-2 This is the first embodiment of the present invention. This embodiment provides a distribution network and flexible resource mutual control system for predicting the spatiotemporal distribution of source and load. The system consists of an acquisition and preprocessing module, a prediction modeling module, an analysis and evaluation module, a control optimization module, and a monitoring and correction module.

[0052] Specifically, the data acquisition and preprocessing module is used to collect multi-source data from the power distribution network, perform anomaly detection, missing data repair and normalization on the multi-source data, and construct a high-quality time series dataset.

[0053] Furthermore, outlier detection is performed on the collected raw dataset using a multi-layered discrimination strategy: a sliding window difference method is used to detect abrupt changes in the time series; the Z-score statistical method is used to identify deviations in the spatial nodes; the Pearson correlation coefficient method is used to detect logical anomalies in the correlation between devices; when an outlier is detected, its occurrence node and time are recorded, and correction is performed using adjacent time mean interpolation or multivariate regression prediction.

[0054] To address data loss caused by sensor failures or communication packet loss, a two-layer repair strategy based on temporal correlation and spatial correlation is employed: at the temporal level, linear interpolation, exponential smoothing, or LSTM time prediction are used to compensate for short-term missing data; at the spatial level, weighted averaging of adjacent node data or graph convolutional interpolation based on network topology is used. After the repaired dataset passes consistency verification, it enters the normalization stage.

[0055] Specifically, the prediction modeling module is used to construct a source-load intelligent prediction model based on the high-quality time-series dataset using a convolutional neural network model, perform feature dimensionality reduction and denoising processing, and output the temporal feature representation of each node through a fully connected layer; by fusing the time-series features of the convolutional neural network with the spatial features of the graph structure, the prediction results of the spatiotemporal distribution of source-load for each node are obtained.

[0056] Furthermore, convolutional operations are performed on the time-dimensional data in the convolutional layers to extract local temporal features and trend changes. The convolutional output is then processed by pooling layers for dimensionality reduction, filtering out high-frequency noise and retaining the main change features to obtain a smoothed node feature sequence. The feature sequence output from the pooling layer is input into the fully connected layer, where features at different time steps are comprehensively mapped using weight parameters to generate the temporal feature representation of each node. Through end-to-end training, historical power and output data of nodes are used as supervision signals to jointly optimize the parameters of the convolutional and fully connected layers. The calculation formula for the convolution operation is as follows:

[0057] ;

[0058] Where f is the activation function of the convolutional layer, and k is the number of convolutional kernels. The weight matrix of the convolution kernel. Let b be the i-th input time series data, where i ranges from 1 to n, n is the number of time series data, and b is the bias value. For the corresponding The output value of the convolution kernel.

[0059] Furthermore, the fused node joint features are input into the output layer to generate load and distributed energy output prediction results for each node in the future prediction period, forming a node-level source-load spatiotemporal distribution prediction sequence; the time step information and node identifier are retained in the output prediction sequence.

[0060] It should be noted that a high-quality time-series dataset that has undergone anomaly repair and normalization is used as input. Each node contains multi-dimensional input features, such as historical load power, distributed energy output, energy storage status, and meteorological variables (e.g., temperature, illumination, wind speed). In the convolutional layer, a one-dimensional convolution operation is performed on the time dimension to extract local temporal features and trend changes. Specifically, the resource capacity assessment module is used to calculate the adjustment capacity of energy storage units and controllable loads based on the source-load spatiotemporal distribution prediction results, and to perform node sensitivity analysis on flexible resources.

[0061] To reduce data dimensionality and noise interference, an average pooling layer is added after the convolutional layer output to downsample adjacent convolutional outputs, preserving key variation features and filtering out high-frequency noise. For example, the average of the convolutional results every 5 time steps is taken to generate a smoothed node feature sequence.

[0062] Specifically, the analysis and evaluation module is used to establish a mathematical model of the flexibility of renewable energy and energy storage devices based on the spatiotemporal distribution prediction results of source and load, and to calculate the available up- and down-adjustment capacity of each node, forming a time series flexibility capability curve.

[0063] Furthermore, based on the predicted renewable energy output, the equivalent controllable portion of renewable energy is determined according to the system's set confidence capacity, while the remaining portion is considered as uncontrollable load and its associated flexibility. The flexibility of renewable energy supply and consumption is as follows:

[0064] ;

[0065] ;

[0066] in, The flexibility that renewable energy can provide For the confidence capacity of renewable energy, Let t represent the amount of flexible resources consumed by renewable energy at time t. For the moment of renewable energy contribution;

[0067] Based on the charge / discharge state and capacity limitations of the energy storage device, calculate the up and down flexibility that the energy storage device can provide at time t:

[0068] ;

[0069] ;

[0070] in, , Energy storage devices Provides flexibility in both positive and negative directions at all times; , This refers to the maximum charging and discharging power of the energy storage device. for Energy stored in a time-of-use energy storage device; , These are the upper and lower limits of the energy storage capacity, respectively. This is the scheduling time interval.

[0071] Furthermore, the upward and downward capacity adjustments of each node at each moment are arranged in chronological order to form node-level upward and downward adjustment capability curves; the source device ID, calculation assumptions, and confidence index of each time step are recorded in the capability curves; and a network-wide flexibility adjustment capability matrix is ​​generated based on the time series capability curves of each node.

[0072] It should be noted that, for example, if the installed capacity of a wind farm is 1000kW and the confidence level is 80%, then P cl =800 kW; when the output at predicted time t is Pres,t=900kW:

[0073] F res =800 kW,D res,t =+100 kW;

[0074] This indicates that there is an uncertain output of 100kW at this moment, which needs to be absorbed by other flexible resources;

[0075] For an energy storage device with a capacity of 500kWh and a maximum charge / discharge power of 200kW:

[0076] When E t =250kWh, Δt=1h, then:

[0077] F up,t =min(200,(500-250) / 1)=200kW;

[0078] F down,t =min(200,(250-0) / 1)=200 kW;

[0079] This indicates that the energy storage capacity can be adjusted up or down by 200 kW at this moment, which can fully respond to the system's adjustment needs.

[0080] Specifically, the regulation and optimization module is used to construct a multi-objective comprehensive optimization model. It adopts a reinforcement learning algorithm based on asynchronous advantage Actor-Critic to solve the regulation strategy. The strategy optimization and value assessment are achieved through multi-threaded parallel interaction between the Actor and Critic networks.

[0081] Furthermore, an optimization function is established with the objectives of power balance, voltage stability, and system operation economy. At the same time, it is combined with the line capacity, voltage constraints, and the adjustment limits of energy storage and adjustable load to form joint constraints at the node level and system level. The Actor network is used to generate node control strategies, that is, to output adjustment actions under a given state. The Critic network is used to evaluate the value of the actions taken and calculate the reward signal.

[0082] A multi-threaded approach is adopted, with each thread independently running a pair of Actor-Critic networks to interact with the environment to generate policies and value assessments. After completing a certain training step, each thread asynchronously uploads its parameters to the global network for synchronous updates.

[0083] It should be noted that the A3C algorithm improves training efficiency and strategy diversity through a multi-threaded asynchronous mechanism:

[0084] Each thread runs an independent environment instance (corresponding to a local power grid of different nodes or regions), generating state, action, and reward sequences through interaction with the environment; after completing training with a fixed step size n, each thread calculates the gradient. and The gradient updates are asynchronously uploaded to the global network; the global network integrates gradient updates from all threads to achieve global parameter synchronization, thereby enabling fast convergence in multiple scenarios.

[0085] Specifically, the monitoring and correction module is used to monitor the operating status of the distribution network in real time, calculate the deviation rate between the measured operating data and the predicted source load results, and automatically trigger the closed-loop correction mechanism when the prediction deviation of any node exceeds the set threshold.

[0086] Furthermore, the deviation rate between the predicted results and the actual results is calculated:

[0087] ;

[0088] in, Let be the measured active power value of the nth node. This represents the predicted active power value for the nth node.

[0089] If ΔP n If the prediction result of that node is found to be too biased when the threshold is exceeded, a closed-loop correction mechanism is automatically triggered; if If the value is 0, then a safety correction strategy is adopted.

[0090] It should be noted that, to avoid the system accidentally triggering the correction mechanism, a prediction deviation rate threshold ΔP is set. th For example, if we take 10%, the system makes the following real-time judgment: when |ΔP n |≤ΔP th : The prediction accuracy is considered acceptable, and no correction is triggered; when |ΔPn |>ΔP th If the prediction deviation of this node is determined to be too large, a closed-loop correction mechanism will be automatically triggered.

[0091] When P npred =0: The deviation rate cannot be calculated. To prevent abnormal amplification, a safety correction strategy is adopted, which is to replace it with the average power of neighboring nodes or the time series trend prediction result.

[0092] Example 2

[0093] This embodiment is the second embodiment of the present invention. This embodiment provides a power distribution network and flexible resource mutual control system for predicting the spatiotemporal distribution of source and load. In order to verify the beneficial effects of the present invention, a simulation experiment is conducted for scientific demonstration.

[0094] Simulation scenario and dataset preparation

[0095] Baseline network: IEEE 33-node distribution network topology, with N=33 nodes; 20 photovoltaic (PV) power generation units (distributed across several nodes); 4 wind power units; 10 energy storage (ESS) units (connected to 8 nodes); and 30 controllable load (DR) groups (distributed across multiple nodes).

[0096] Simulation period: 7 days (168 hours), time step Δt = 5 minutes (12 steps per hour, 2016 steps in total); meteorological data: hourly landscape modeling is used to drive PV / wind power output;

[0097] Load baseline: constructed based on the historical typical daily load curve of each node, with 10%-30% random disturbance added to simulate uncertainty; Training set: the first 5 days (120 hours) are used to train the prediction model and calibrate the parameters;

[0098] Validation set: Day 6 is used for model parameter tuning.

[0099] S1: Collect historical load, PV / wind output, ESS status and weather data, perform missing data repair, anomaly detection, normalization and store in time series database;

[0100] S2: Use GNN (GCN, 3 layers) encoding space + Transformer encoder to predict two sets of predictions for the next 1 hour (12 steps) and 4 hours (48 steps);

[0101] S3: Calculate the ESS up / down capacity at each time step, introduce an uncertainty reduction factor, and form a capacity curve sequence;

[0102] S4: Establish a multi-objective optimization with constraints including line capacity, node voltage, ESS SOC and DR constraints. The solver adopts an improved hybrid intelligent algorithm.

[0103] S5: Compare the prediction with the actual measurement every 5 minutes. If a node |ΔP n |>ΔP th | Or, if the voltage exceeds the limit, it may trigger closed-loop re-optimization or an emergency action.

[0104] Model and algorithm parameters

[0105] Optimization Algorithm: IGA: Population size 120, 200 iterations, crossover rate 0.6, adaptive mutation rate, fitness score weighted by incorporating node influence: Fitness = 1 / (target + λ / ΣI);

[0106] RL: Offline training is based on the policy gradient algorithm. The state includes local prediction error and node score. Actions are minor device adjustments. Reward = -(α1ΔP + α2ΣΔU + α3Cost). IGA is used for global search. RL fine-tunes local solutions and serves as a candidate policy generator. Bias threshold ΔP th =10%; voltage upper and lower limits are ±5%.

[0107] In this embodiment, a typical 33-node distribution network containing distributed photovoltaic and energy storage units is used as the test object, and seven consecutive days of operating data are selected for simulation verification. The data from the first six days are used for model training and parameter adaptation, and the seventh day is used as the test day to verify the performance of the present invention.

[0108] The table below shows the statistical results of the test days in the 7-day simulation:

[0109]

[0110] Table 2 Simulation Results and Effects Description

[0111] Table 2 shows that the average power balance deviation of the present invention is reduced by 90.7% compared with the traditional method.

[0112] Due to the introduction of a closed-loop correction mechanism and node sensitivity weights, the system's voltage over-limit events are reduced by 92.8%, and operating costs are reduced by approximately 13%. The real-time response latency is significantly reduced to only 1.8 seconds, which can meet the real-time requirements of distributed control. The high number of closed-loop correction triggers indicates that the system has continuous dynamic adaptive capabilities and can maintain stable operation in a multi-source fluctuation environment.

[0113] This embodiment also provides a computer device applicable to the distribution network and flexible resource interoperability system for source-load spatiotemporal distribution prediction, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the distribution network and flexible resource interoperability system for source-load spatiotemporal distribution prediction as proposed in the above embodiment.

[0114] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0115] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a power distribution network and flexible resource interoperability system for predicting the spatiotemporal distribution of source and load as proposed in the above embodiments.

[0116] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A distribution network and flexible resource interoperability system for predicting the spatiotemporal distribution of load and energy sources, characterized in that, include: The data acquisition and preprocessing module is used to collect multi-source data from the power distribution network, perform anomaly detection, missing data repair and normalization on the multi-source data, and construct a high-quality time series dataset. The prediction modeling module is used to construct a source-load intelligent prediction model based on the high-quality time-series dataset using a convolutional neural network model, perform feature dimensionality reduction and noise reduction, and output the temporal feature representation of each node through a fully connected layer; and fuse the time-series features of the convolutional neural network with the spatial features of the graph structure to obtain the source-load spatiotemporal distribution prediction results of each node. The analysis and evaluation module is used to establish a mathematical model of the flexibility of renewable energy and energy storage devices based on the spatiotemporal distribution prediction results of source and load, and to calculate the available up- and down-adjustment capacity of each node, forming a time series flexibility capability curve. The regulation and optimization module is used to construct a multi-objective comprehensive optimization model. It adopts a reinforcement learning algorithm based on asynchronous advantage Actor-Critic to solve the regulation strategy. The strategy optimization and value assessment are realized through multi-threaded parallel interaction between the Actor and Critic networks. The monitoring and correction module is used to monitor the operating status of the distribution network in real time, calculate the deviation rate between the measured operating data and the predicted source load results, and automatically trigger the closed-loop correction mechanism when the prediction deviation of any node exceeds the set threshold.

2. The distribution network and flexible resource interoperability system for source-load spatiotemporal distribution prediction as described in claim 1, characterized in that, The process of collecting multi-source data from the power distribution network and performing anomaly detection, missing data repair, and normalization on the multi-source data includes: Anomaly detection is performed on the collected raw dataset using a multi-layered discrimination strategy: a sliding window difference method is used to detect abrupt changes in the time series; the Z-score statistical method is used to identify deviations in spatial nodes; the Pearson correlation coefficient method is used to detect logical anomalies in the correlation between devices; when anomalies are detected, the occurrence node and time are recorded, and corrections are made using adjacent time mean interpolation or multivariate regression prediction. To address data loss caused by sensor failures or communication packet loss, a two-layer repair strategy based on temporal correlation and spatial correlation is employed: at the temporal level, linear interpolation, exponential smoothing, or LSTM time prediction are used to compensate for short-term missing data; at the spatial level, weighted averaging of adjacent node data or graph convolutional interpolation based on network topology is used. After the repaired dataset passes consistency verification, it enters the normalization stage.

3. The distribution network and flexible resource interoperability system for source-load spatiotemporal distribution prediction as described in claim 1, characterized in that: Based on the high-quality time-series dataset, a source-load intelligent prediction model is constructed using a convolutional neural network model for feature dimensionality reduction and denoising. The model then outputs the time-series feature representations of each node through a fully connected layer, including: Convolutional operations are performed on time-dimensional data in convolutional layers to extract local temporal features and trend changes; The convolutional output is processed by a pooling layer to reduce dimensionality, filtering out high-frequency noise and retaining the main changing features, resulting in a smoothed node feature sequence. The feature sequence output from the pooling layer is then input into a fully connected layer, where weight parameters are used to comprehensively map the features at different time steps, generating a temporal feature representation for each node. Through end-to-end training, historical power and output data of the nodes are used as supervision signals to jointly optimize the parameters of the convolutional and fully connected layers. The calculation formula for the convolution operation is as follows: ； Where f is the activation function of the convolutional layer, and k is the number of convolutional kernels. The weight matrix of the convolution kernel. Let b be the i-th input time series data, where i ranges from 1 to n, n is the number of time series data, and b is the bias value. For the corresponding The output value of the convolution kernel.

4. The distribution network and flexible resource interoperability system for source-load spatiotemporal distribution prediction as described in claim 1, characterized in that: The fusion of the time-series features and graph structure spatial features of the convolutional neural network yields the source-load spatiotemporal distribution prediction results for each node, including: The fused node joint features are input into the output layer to generate load and distributed energy output prediction results for each node in the future prediction period, forming a node-level source-load spatiotemporal distribution prediction sequence; the time step information and node identifier are retained in the output prediction sequence.

5. The distribution network and flexible resource interoperability system for source-load spatiotemporal distribution prediction as described in claim 1, characterized in that, The process of establishing a flexible mathematical model for renewable energy and energy storage devices based on the spatiotemporal distribution prediction results of source and load includes: Based on the projected renewable energy output, the equivalent controllable portion of renewable energy is determined according to the system's set confidence capacity, while the remaining portion is considered as uncontrollable load and its associated flexibility. The flexibility of renewable energy supply and consumption is as follows: ； ； in, The flexibility that renewable energy can provide For the confidence capacity of renewable energy, Let t represent the amount of flexible resources consumed by renewable energy at time t. For the moment of renewable energy contribution; Based on the charge / discharge state and capacity limitations of the energy storage device, calculate the up and down flexibility that the energy storage device can provide at time t: ； ； in, , Energy storage devices Provides flexibility in both positive and negative directions at all times; , This refers to the maximum charging and discharging power of the energy storage device. for Energy stored in a time-of-use energy storage device; , These are the upper and lower limits of the energy storage capacity, respectively. This is the scheduling time interval.

6. The distribution network and flexible resource interoperability system for source-load spatiotemporal distribution prediction as described in claim 1, characterized in that: The calculation of the available up-adjustment and down-adjustment capacity for each node, forming a time series flexibility curve, includes: The upward and downward capacity adjustments of each node at each moment are arranged in chronological order to form node-level upward and downward adjustment capacity curves. The source device ID, calculation assumptions, and confidence index of each time step are recorded in the capacity curves. Based on the time series capacity curves of each node, a network-wide flexibility adjustment capacity matrix is ​​generated.

7. The distribution network and flexible resource interoperability system for source-load spatiotemporal distribution prediction as described in claim 1, characterized in that: The multi-objective comprehensive optimization model is constructed by employing a reinforcement learning algorithm based on asynchronous advantage Actor-Critic to solve the control strategy. Strategy optimization and value evaluation are achieved through multi-threaded parallel interaction between the Actor and Critic networks, including: An optimization function is established with the objectives of power balance, voltage stability, and system operation economy. At the same time, it is combined with the line capacity, voltage constraints, and the adjustment limits of energy storage and adjustable load to form joint constraints at the node level and system level. The Actor network is used to generate node control strategies, that is, output adjustment actions under given conditions. Critic networks are used to evaluate the value of actions taken and calculate reward signals; A multi-threaded approach is adopted, with each thread independently running a pair of Actor-Critic networks to interact with the environment to generate policies and value assessments. After completing a certain training step, each thread asynchronously uploads its parameters to the global network for synchronous updates.

8. The distribution network and flexible resource interoperability system for source-load spatiotemporal distribution prediction as described in claim 1, characterized in that: The real-time monitoring of the distribution network operation status involves calculating the deviation rate between the measured operation data and the predicted source load results. When the prediction deviation of any node exceeds a set threshold, a closed-loop correction mechanism is automatically triggered, including: calculating the deviation rate between the prediction results and the operation results. ； in, Let be the measured active power value of the nth node. This represents the predicted active power value for the nth node. If ΔP n If the prediction result of that node is found to be too biased when the threshold is exceeded, a closed-loop correction mechanism is automatically triggered; if If the value is 0, then a safety correction strategy is adopted.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the power distribution network and flexible resource interoperability system for predicting the spatiotemporal distribution of source and load as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the power distribution network and flexible resource interoperability system for predicting the spatiotemporal distribution of source and load as described in any one of claims 1 to 8.

Images