Adaptive interval-based demand side power load partitioning method and system

Abstract

The application discloses a demand side power load division method and system based on adaptive intervals, relates to the technical field of power system dispatching control, and solves the problem that the traditional method of first division and then review is prone to interval backtracking and repeated calculation when being implemented, and is difficult to meet the one-time issuing requirement for implementation; by embedding the feasible region constraint of voltage margin and current margin in the interval generation process, and by using physical quantities such as voltage influence intensity and current influence intensity constructed by sensitivity and node margin to train the auto-encoder to obtain low-dimensional features, the interval division can remain sensitive to the risk of exceeding the limit when being formed; based on the low-dimensional features and the sensitivity threshold, a load conflict graph is constructed, coloring is performed on the graph to obtain a peak-shaving period, nodes in the same interval that may still produce a superposition effect are decoupled on the time axis; combined with the sensitivity continuously corrected interval and time sequence updated by online identification, a structured interval dispatching parameter set is finally output, and the implementability and stability are ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system dispatching control, and more particularly to a demand side power load division method and system based on adaptive intervals. BACKGROUND

[0002] In the demand side management of distribution networks, common load clustering often relies on power consumption curves or business portraits for similarity clustering, and then safety checking and dispatching are performed in the execution stage through optimal power flow or rule base. This path does not explicitly consider the voltage and current constraints of the distribution network when the interval is formed, nor does it map the coupling relationship of topology, line impedance and distributed power injection to the feature space that can be used for optimization, resulting in concurrent peak shifting under the same feeder or the same bus, which easily induces power flow migration and adjacent line over-limit. At the same time, line parameter aging, switch state change and scene load fluctuation will cause the linearization model to deviate, and the traditional method of first division and then review is prone to interval rollback and repeated calculation when landing execution, which is difficult to meet the one-time issuance requirement for execution.

[0003] In view of the above problems, the present application provides a solution. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide a demand side power load division method and system based on adaptive intervals to solve the problems raised in the background art.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0006] The demand side power load division method based on adaptive intervals comprises the following steps:

[0007] Obtain the topology, node voltage, current and load power time series data of the distribution network, construct a power flow linearization equation based on LinDistFlow approximation, calculate the sensitivity matrix of voltage to active and reactive power and the sensitivity matrix of current to active and reactive power, and use recursive least squares to identify the sensitivity matrix online;

[0008] Extract the voltage influence intensity, current influence intensity, voltage margin and current margin characteristics of the node, input the characteristics into the trained graph autoencoder for dimension reduction, and generate a low-dimensional feature vector that retains the topology and physical coupling information of the distribution network;

[0009] Construct a load conflict graph based on the low-dimensional feature vector, the edges of the conflict graph represent the risk of over-limit when two loads are adjusted concurrently, and use a graph coloring algorithm to assign off-peak periods to the nodes of the load conflict graph according to a predetermined rule to form an off-peak scheme;

[0010] In combination with the staggered peak scheme and the voltage margin and current margin constraints, a differentiable clustering algorithm is used to divide the load intervals, and a load interval division result is output.

[0011] In a preferred embodiment, the online identification includes: presetting a sensitivity matrix element drift threshold value, comparing a current value with a historical reference value in real time, controlling the distributed power supply to inject a small perturbation within a preset amplitude interval when the difference exceeds the threshold value, collecting node voltage and current data after the perturbation, and updating the sensitivity matrix by recursive least squares.

[0012] In a preferred embodiment, the training of the graph autoencoder includes: using node features and topological adjacency of the power distribution network for a preset historical time length as a training set, corresponding to the number of node features, setting a preset number of hidden layers and using linear rectification activation, and simultaneously reconstructing the features and adjacency matrix at the output end, and optimizing the mean square error loss to a preset convergence threshold.

[0013] In a preferred embodiment, the construction of the load conflict graph includes: estimating the influence of concurrent adjustment of any two loads on branch current and node voltage according to the sensitivity matrix, establishing an edge between the corresponding node pairs when any branch is occupied or any node margin is lower than a preset safety threshold, and obtaining an edge set of the load conflict graph.

[0014] In a preferred embodiment, the graph coloring algorithm uses a greedy strategy and combines capacity constraints: staggered peak periods are assigned in descending order of node degree of the load conflict graph, nodes with edges are prohibited from falling into the same period, the number of staggered peak periods is within a preset interval, and the length of the period is determined according to a preset rule of capacity limit and scheduling period.

[0015] In a preferred embodiment, the initialization of the differentiable clustering includes: setting interval prototype vectors according to the number of staggered peak periods, initializing the prototypes by calculating the mean of the low-dimensional feature vectors of the loads in the corresponding periods, calculating the belonging probability of each load to each interval using a Softmax function, and constraining the sum of the belonging probabilities to be equal to one.

[0016] In a preferred embodiment, the objective function of the differentiable clustering includes: a distance term of the low-dimensional feature vector of the load and the prototype vector of the interval to which it belongs, and a logarithmic barrier penalty term constructed by the voltage margin and the current margin, and an adaptive matrix estimation type optimizer is used to iterate within a preset learning rate interval until a preset convergence criterion is reached.

[0017] In a preferred embodiment, the method further comprises: generating an interval scheduling parameter set based on the load interval division result, the parameter set containing interval identification, node set, planned adjustment amount of each node and safety threshold, and outputting the parameter set to a control unit through a communication interface for subsequent execution and docking.

[0018] In a preferred embodiment, the output is verified by a communication authentication module before output, the verification including key check and identity matching; after authorization, the interval scheduling parameter set is issued by an instruction transmission module, node voltage and current data are collected in real time by a data monitoring module during execution, and when the safety threshold is exceeded, event information is stored by a data recording module and an abnormal code is returned.

[0019] The adaptive interval-based demand-side power load division system comprises:

[0020] A receiving module configured to obtain power distribution network operation data and model parameters, the operation data including power time series of each node voltage, current and each demand-side load, and the model parameters including power distribution network topology and line parameters, and if part of the parameters are unknown, they are recorded as to-be-identified parameters;

[0021] A processing module configured to perform physical modeling processing on the data obtained by the receiving module, calculate the sensitivity matrix of voltage to active power and reactive power and the sensitivity matrix of current to active power and reactive power based on the approximately linearized power flow equation of LinDistFlow, and online identify the sensitivity matrix by using recursive least squares, and then combine the node voltage influence strength, current influence strength, voltage margin and current margin to construct a physical feature vector, and generate a low-dimensional feature vector retaining topology and physical coupling information by dimension reduction through a graph autoencoder;

[0022] A determination module configured to set a physical constraint rule according to the feasible region limit of power distribution network voltage and current, input the low-dimensional feature vector generated by the processing module, divide the load interval by using a differentiable clustering algorithm, minimize the interval internal feature divergence and the risk of exceeding the limit in the clustering objective function, and output the load interval division result;

[0023] A sending module configured to generate a scheduling instruction according to the load interval division result output by the determination module, determine the instruction execution order according to a preset priority strategy, send the scheduling instruction link to the corresponding control execution unit, monitor the node voltage and branch current in real time, and record the operation log containing the control execution unit identification information, operation content and time information synchronously.

[0024] The adaptive interval-based demand-side power load division method and system have the following technical effects and advantages:

[0025] The application can keep sensitive to the risk of exceeding the limit when interval division is formed, by embedding the feasible region constraints of voltage margin and current margin in the interval generation process, and obtaining low-dimensional features by training the graph auto-encoder with physical quantities such as voltage influence strength and current influence strength constructed by sensitivity, and node margin; the load conflict graph is constructed based on the low-dimensional features and the sensitivity threshold, and the peak-shaving period is obtained by coloring the graph, so that the nodes in the same interval that may still produce superposition effect are decoupled on the time axis; the sensitivity is continuously corrected by combining the online identification updated sensitivity and the interval and timing, and finally the structured interval scheduling parameter set is output, to ensure the implementability and stability under different working conditions and topology switching conditions. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 The application is based on the adaptive interval demand side power load division method flowchart;

[0027] Figure 2 The application is a matching control unit and scheduling flowchart;

[0028] Figure 3 The application is based on the adaptive interval demand side power load division system structure diagram. DETAILED DESCRIPTION

[0029] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments.

[0030] The application provides a demand side power load division method based on adaptive intervals, referring to Figure 1 , Figure 1 The application provides a flowchart of the interval method, and the method comprises:

[0031] In step S101, power distribution network operation data and model parameters are obtained; the operation data includes node voltage, current and power time series of each demand side load; the model parameters include power distribution network topology structure and line parameters such as impedance, and if part of the parameters are unknown, they are recorded as to-be-identified parameters. The operation data is obtained through a collection terminal, a SCADA system or a smart meter, and an undirected topological graph model of the power distribution network is established.

[0032] In step S102, the obtained data is processed by physical modeling to obtain processed data representing the feasible region of the power distribution network; wherein the processed data includes an electrical sensitivity matrix obtained based on LinDistFlow approximation and physical characteristic representation of each node. Specifically, the radial topology and line impedance parameters of the power distribution network are used to linearize the power flow equation, to establish an approximate linear relationship between node injection power and voltage and current changes: , .in , This is a sensitivity matrix for voltage and current to changes in active and reactive power. When some line parameters are missing or drift, the elements of the sensitivity matrix can be identified online through small disturbance tests. For example, if a small power step is introduced by a distributed power source or load device during certain periods, the sensitivity matrix estimate is continuously corrected using a recursive least squares algorithm until it converges to the true value. Then, based on the sensitivity matrix, the physical influence characteristics of each node are calculated, including voltage influence strength and current influence strength. Voltage influence strength refers to the degree of influence of the node's injected power change on the voltage of each neighboring node, and current influence strength refers to the degree of influence of the node's injected power change on the current of each related branch. The above influence characteristics, along with the node's local voltage margin and branch current margin, are combined to form the physical feature vector of each node. The physical feature vector of each node is dimensionality-reduced using a graph embedding algorithm to obtain the low-dimensional representation vector h of the node. During this dimensionality reduction process, a graph autoencoder is trained to reconstruct the original features and adjacency relationships, ensuring that the node representation vector h retains the grid topology and physical coupling information. The processed data obtained in this step includes physical semantic information such as sensitivity matrix and node representation, ensuring that subsequent steps can fully perceive the electrical feasible domain of the distribution network.

[0033] Step S103: Based on preset physical constraint rules, determine the load interval division scheme corresponding to the processed data. The physical constraint rules are based on the feasible region constraints of the distribution network voltage and current, and impose constraints on the load adjustment actions executed simultaneously. On this basis, differentiable clustering or optimization algorithms are used to divide the demand-side loads into intervals, and loads with similar electrical impact characteristics and which will not exceed limits when executed concurrently are grouped into the same interval. Specifically, the representation vector h of the node to which each load belongs, obtained in step S102, is used as the clustering input to initialize several interval prototype vectors. Through iterative optimization, each load is softly assigned to a certain interval with a certain probability. During the optimization process, an objective function is set to simultaneously minimize the feature divergence and limit-exceeding risk within the interval: on the one hand, the representation vectors of loads within the same interval are constrained to be close to each other while the distance between different intervals is large; on the other hand, the physical feasible region constraints are incorporated into the clustering objective to penalize interval combinations that may cause voltage or current to exceed the safety margin. Specifically, a certain interval is estimated through the aforementioned sensitivity matrix. The concurrent adjustment actions specifically include changes in the active and reactive power of each node. , This has an impact on the overall network voltage and branch current. , Calculate the voltage of each node relative to the upper limit under the action of this interval. remaining margin And the current in each branch relative to the thermal stability limit. remaining margin The minimum voltage margin and minimum current margin for the entire range are introduced into the objective function. A logarithmic barrier is used to penalize situations where the margin of any node or branch approaches zero, thereby guiding the clustering results to avoid combinations with high risk of exceeding limits. By solving this objective function, the range label for each load is determined. This enables automatic interval partitioning based on physical feasible domain awareness.

[0034] Step S104: A scheduling instruction link is sent to the corresponding control execution unit to prompt the execution unit to perform concurrent scheduling control on the processed data. Specifically, based on the interval division result of step S103, a set of scheduling task instructions corresponding to each interval is generated, with one instruction for each interval. Then, the execution order of tasks in each interval is determined according to a preset priority strategy, and the instructions are sent to the corresponding control equipment or scheduler for load adjustment operations in this order. The scheduling instruction is used to prompt the control execution unit to perform the following operations to complete the load adjustment within the interval: click the instruction link and verify identity or permissions; if the verification is successful, enter the execution interface and adjust the power of the load within the interval according to the predetermined plan. By directly pushing the scheduling instruction to the relevant control personnel or automatic control device, the rapid issuance and execution of tasks are achieved. During the instruction execution process, the system monitors the voltage of key nodes and the branch current in real time to ensure that the adjustment operation remains within a safe threshold. If an abnormal trend is detected, the system can promptly send a warning or automatically stop the relevant operation.

[0035] The execution entity for the above steps can be the backend process of a distribution network dispatch control system or energy management platform. By comprehensively processing distribution network data through these steps, intelligent automation of load optimization dispatch is achieved while ensuring that voltage, current, and other data meet safety constraints. First, raw data of distribution network operation is collected, and the sensitivity parameters and representation vectors of each node's impact on voltage and current are obtained through physical model analysis. This standardized processing not only solves the problem of difficulty in quantifying the physical coupling relationship between different loads but also allows subsequent interval division and dispatch optimization to be established on a unified physical semantic space, significantly reducing errors that may arise from traditional experience-based division. Based on preset physical constraint rules, the system automatically matches each load adjustment task to the most suitable interval, significantly improving the accuracy and efficiency of decision-making compared to manual grouping based on experience. For example, load adjustment tasks involving heavy line loads are automatically assigned to intervals that reduce the power flow of that line, and reactive power compensation tasks involving voltage anomalies are assigned to intervals that support voltage. After receiving the dispatch instruction link, the control execution unit only needs to click to verify to jump to the execution interface and implement load adjustments. This seamless experience reduces the average scheduling response time, while the system records complete operation logs in real time for subsequent auditing and analysis.

[0036] The implementation effects of the technical solution mainly lie in three aspects: firstly, in terms of safety, by integrating the electrical feasible domain constraint into interval division, cooperating with identity permission verification before execution and whole-process log recording, the safety and controllability of the load concurrent adjustment process are guaranteed, and risks such as voltage out-of-limit or line overload are avoided. Secondly, in terms of efficiency, the load clustering and generation of dispatching instructions are automatically completed, and are directly sent to the execution unit, which greatly saves the time of manual analysis and step-by-step issuance of instructions, and realizes rapid response. Finally, in terms of application scenarios, the scheme can be widely applied to peak load shifting control, distributed power consumption optimization, power grid demand response and emergency control in fault scenarios of distribution networks, and has important significance for improving the intelligent dispatching level of power grids.

[0037] In an optional embodiment, the acquired data is subjected to physical modeling processing to obtain processed data representing the feasible domain of the distribution network, specifically comprising: calculating the Jacobian matrix of the LinDistFlow equation based on the distribution network topology and line parameters to obtain a voltage sensitivity matrix and a current sensitivity matrix . In the case of incomplete or changed line parameters, the recursive least squares method is used to identify the sensitivity matrix online to obtain updated sensitivity matrix estimates and their covariance. For example, during the operation of the distribution network, when a slight change in the current of a branch is detected, it can be inferred that the active or reactive power output of the nearby node has been disturbed, and the disturbance data is used to correct the corresponding elements of the sensitivity matrix, and after multiple iterations and convergence, more accurate sensitivity parameters are obtained. The processed data obtained by the above method can dynamically reflect the physical characteristics of the distribution network, providing an accurate basis for subsequent interval division.

[0038] In the above embodiment, the system constructs the physical representation vector of the node through the electrical influence characteristics, and uses the graph autoencoder to maintain the adjacency mode related to the topology. This means that when dividing the interval, the factor of "which loads are adjusted at the same time to make the current of which line rise significantly" can be considered, so as to avoid potential risks caused by grouping only according to the similarity of load curves. Specifically, the voltage influence intensity of node i on neighbor node j and the current influence intensity of node i on branch e are defined. Taking these influence intensities in the limited radius neighborhood as the main characteristics, combined with the voltage and current margin indicators of the node itself and the distributed power output and other information, the physical feature vector of the node is constructed. Then the autoencoding network is trained to enable the low-dimensional vector to reconstruct the original features and their adjacency relations, so that the physical factors such as topology and impedance are implicitly contained in the node representation vectors The node representation vectors obtained in this way have physical semantics when used for interval division, and can divide nodes with potential mutual influence far apart, thereby reducing the risk of concurrency.

[0039] In an optional embodiment, the load interval division scheme is determined according to preset physical constraint rules, including: in the case that the target load data does not contain a preset mark, determining a first control unit and a first scheduling process corresponding to the target load; in the case that the target load data contains a preset mark, determining a second control unit and a second scheduling process corresponding to the target load. The target load data is any one of the processed data, for example, an adjustment task of a certain load, the first control unit is the first execution node of the first scheduling process, the first scheduling process includes the second control unit, the second scheduling process includes the second control unit, and the control execution unit includes the first control unit and the second control unit. The above-mentioned preset mark can be used to identify a load adjustment task with special emergency degree or importance level, so as to realize overstep processing in the process. After obtaining the processed data, the system first judges whether each data, for example, a load adjustment task, contains a special mark, to decide to use different scheduling processes. If it does not contain a preset mark, it is processed according to the first scheduling process, and the task is assigned to the first control unit to perform preliminary adjustment; if it contains a preset mark, it is processed according to the second scheduling process, and the task is directly assigned to the second control unit of a higher level for processing, thereby simplifying the process. The second scheduling process is equivalent to a simplification or overstep of the first scheduling process. For example: a certain load adjustment task involves important safety risk elimination and needs to be processed in time, so the system adds a preset emergency mark to it, and the task will jump over a first control unit such as a regional control station and be directly sent to a power distribution dispatching center as a second control unit for processing, and the process ends after the final execution. Through this marking-based mechanism, overstep scheduling can be realized for special applications, and the response speed and pertinence are improved. Figure 2 is a schematic diagram of matching a control unit and a scheduling process in an embodiment of the present application, and the above two processes will be further described below. Figure 2 The above two processes will be further described below. Figure 2 As shown in the figure, the left path is a regular process without marking, which is reviewed by the regional control station as the first control unit after approval, and then sent to the second control unit, the power distribution dispatching center, for review; the right path is an overstep process with marking, which is directly processed by the power distribution dispatching center, and then the process ends. By judging whether the task data contains a preset mark, the system can flexibly select different processes, which can ensure the hierarchical approval of general tasks to prevent misoperation, and can also ensure that emergency tasks are quickly processed.

[0040] Optionally, in the case that the target load data does not contain the preset mark, the first control unit corresponding to the target load is determined, specifically including: keyword or feature recognition is performed on the target data to obtain a recognition result; and the first control unit corresponding to the recognition result is determined according to a preset correspondence table, wherein the preset correspondence table records the correspondence between different keyword or feature modes and control units. In the above embodiment, the appropriate control execution unit is automatically matched according to the preset recognition rule, which can improve the efficiency and accuracy of task allocation. The correspondence table is preset by power distribution network operation experience and rules, for example: if the data content of the detected load adjustment task contains the "voltage abnormality" related keyword, the correspondence table indicates that this task is allocated to the voltage control special team as the first control unit; for another example, if the content contains a large load transfer feature, the task is allocated to the load management special personnel for processing. Through the keyword or mode matching control unit method, intelligent classification and allocation of the load adjustment task are realized, subjective errors caused by manual judgment are avoided, and the overall scheduling efficiency can be further improved.

[0041] In an optional embodiment, the scheduling instruction link is sent to the control execution unit, specifically including: a group of scheduling instruction links corresponding to each processed data is generated, wherein each item of processed data corresponds to an instruction link; the priority order of each item of processed data is determined according to a preset priority strategy; and each item of processed data corresponds to a control execution unit, and the control execution unit includes a plurality of units. In other words, the system first generates a corresponding instruction link for each load adjustment task in the processed data, and stores these links in a to-be-sent queue; then the tasks in the queue are sorted according to the priority strategy to determine the processing order of the tasks; finally, each instruction is sent to the execution unit responsible for the task in sequence according to the sorting through the message or communication module. For example: at a certain moment, there are N load adjustment tasks to be executed, and the system generates N instruction links corresponding to the N tasks; if the priority strategy A is used to sort the N tasks, the system will send the N instructions to the respective responsible control personnel or device in the order obtained by the A strategy, ensuring that the tasks are executed in the intended order.

[0042] In an optional embodiment, the priority strategy comprises at least one of the following: a first-submission-first-execution strategy, sorting according to the time of generation of each task in the processed data in chronological order, the earlier the submission time, the higher the priority; a short-time-consumption-first strategy, sorting according to the estimated execution time of each task in the processed data, the shorter the estimated execution time, the higher the priority; and a balanced sorting strategy, calculating the balanced ratio of each task in the processed data as follows: (estimated execution time of the task + task waiting time) / estimated execution time of the task, the larger the balanced ratio, the higher the priority, wherein the task waiting time is the difference between the current time and the submission time of the task. The above embodiments provide three different sorting methods for selection. The first-submission-first-execution strategy completely processes according to the time order of the tasks entering the system, ensuring the fairness and first-come-first-served principle of scheduling, and its characteristics are simple and direct, but it may lack flexibility in some emergency situations. The short-time-consumption-first strategy sorts based on the execution time required by each task, and the tasks that are easy to complete and have short time consumption are executed first, which can complete more tasks in a short time, and is suitable for scenarios where a large number of tasks need to be executed, but its disadvantage is that important tasks with long time consumption may be delayed repeatedly. The balanced sorting strategy considers the waiting time and execution time consumption of each task through the above formula. When a task has been waiting for a long time, its balanced ratio will increase, thereby increasing the priority; and the tasks with short time consumption will also have a large ratio due to the small denominator, so this strategy takes into account fairness and efficiency. For example, at a certain time, assuming that the current time is 10:00, there are three tasks a, b, and c to be executed, among which a is submitted at 8:00, the waiting time is 120 minutes, b is submitted at 9:30, the waiting time is 30 minutes, and c is submitted at 9:00, the waiting time is 60 minutes, and the estimated execution time is a=10 minutes, b=30 minutes, and c=20 minutes, respectively. Then the balanced ratio of a is calculated as (10+120) / 10=13, the balanced ratio of b is (30+30) / 30=2, and the balanced ratio of c is (20+60) / 20=4. It can be seen that the balanced ratio is a>c>b, and the execution order of the tasks is a, c, and b in turn. The balanced sorting strategy has the advantages of considering the waiting time and execution speed of the tasks, avoiding any task waiting for too long, and ensuring that the system can prioritize tasks that are easy to complete, combining the advantages of the first two strategies. Among the above three sorting methods, the scheduling administrator can flexibly select according to different business needs: when the number of tasks to be executed is large and needs to be processed quickly, the short-time-consumption-first strategy can be selected; when the number of tasks is not large and needs to strictly follow the first-come-first-served principle, the first-submission-first-execution strategy is selected; and when a balance between fairness and efficiency is desired, the balanced sorting strategy is used. The provision of multiple priority sorting methods increases the flexibility of task scheduling and further improves the user experience.

[0043] In an optional embodiment, the feasible domain-aware interval division method further includes: automatically recording and storing operation logs; wherein, the operation logs are generated during the process of each control execution unit entering the execution interface to perform load adjustment operations, and the operation logs include the identification information of the control execution unit, specific operation content, and time information, etc. By automatically recording the execution process of each load adjustment task, it can be ensured that every step of the scheduling process is traceable. When the adjustment results cause disputes or require tracing responsibility, the log records can provide a detailed chain of evidence. In addition, the collected log data can also be used for subsequent analysis and model training to continuously optimize interval division and scheduling strategies, and achieve continuous improvement of the system. In practical applications, the log recording module can asynchronously save the operation data to a secure storage medium to ensure that it does not interfere with real-time control.

[0044] It should be noted that the method steps provided in the above embodiments are divided into logical sequences. In practical applications, some steps can be executed in different orders by software or hardware as needed, or several steps can be combined. For example, steps S103 and S104 can be executed in parallel under certain circumstances, that is, corresponding instructions can be issued step by step while dividing the interval to shorten the overall response time. Those skilled in the art should understand that the description of the order of method steps in this application does not constitute a limitation.

[0045] This application also provides a demand-side power load allocation system based on adaptive intervals, such as Figure 3 As shown, Figure 3 This is a structural block diagram of the system provided in the embodiments of this application. The system includes:

[0046] The receiving module 301 is used to acquire power distribution network operation data and model parameters, which corresponds to step S101 of the above method.

[0047] Processing module 302 is used to perform physical modeling processing on the acquired data to obtain processed data that characterizes the feasible domain of the distribution network, corresponding to step S102 of the above method.

[0048] The determination module 303 is used to determine the load interval division scheme that matches the processed data according to the preset physical constraint rules, which corresponds to step S103 of the above method.

[0049] The sending module 304 is used to send a scheduling instruction link to the control execution unit, prompting the execution unit to perform concurrent scheduling control on the processed data, corresponding to step S104 of the above method.

[0050] In an optional embodiment, the processing module 302 comprises a modeling unit configured to calculate an electrical sensitivity matrix based on the power distribution network topology and the monitoring data, and to perform physical modeling on the operation data to obtain the processed data; the modeling unit can also use an online identification technique to iteratively correct the sensitivity matrix by using small perturbation data when the model parameters are incomplete, so as to ensure that the processed data is accurate and effective.

[0051] In an optional embodiment, the determining module 303 comprises a first determining unit configured to determine a first control unit and a first scheduling process corresponding to the target load when the target load data does not contain the preset mark, wherein the target load data is any one of the processed data; and a second determining unit configured to determine a second control unit and a second scheduling process corresponding to the target load when the target load data contains the preset mark. The first control unit is the first execution node in the first scheduling process, the first scheduling process comprises the second control unit, the second scheduling process comprises the second control unit, and the control execution unit comprises the first control unit and the second control unit. This module realizes the function of selecting different scheduling processes according to whether there is a special mark in the task data, and guarantees the over-level processing of the emergency task.

[0052] In an optional embodiment, the first determining unit comprises an identifying subunit configured to perform keyword or feature identification on the target data to obtain an identification result when the target load data does not contain the preset mark; and a mapping subunit configured to determine the first control unit corresponding to the identification result according to a preset correspondence table, wherein the preset correspondence table records the correspondence between different keyword or feature modes and control units. Through the cooperation of the identifying subunit and the mapping subunit, the ordinary load adjustment task can be automatically assigned to the appropriate regional control personnel or device, thereby improving the scheduling efficiency.

[0053] In an optional embodiment, the sending module 304 comprises a generating unit configured to generate a group of scheduling instruction links corresponding to the processed data, wherein each item of data in the processed data corresponds to an instruction link; a third determining unit configured to determine the priority order of each task in the processed data according to a preset priority strategy; and a sending unit configured to send the corresponding instruction links to each unit in the group of control execution units in sequence according to the priority order, wherein each item of data in the processed data corresponds to a control execution unit, and the control execution unit comprises a plurality of units. The sending module 304 can further comprise a priority strategy library, which internally stores a plurality of scheduling order decision strategies for the third determining unit to call, so as to meet the needs of different scenarios.

[0054] In an optional embodiment, the priority strategy comprises at least one of the following: a first-submission-first-execution strategy (ordering according to the submission time of the task in the processed data); a short-time-consumption priority strategy (ordering according to the estimated execution time of each task, and the shorter the estimated execution time, the higher the priority); and a balanced-ordering strategy (calculating the balanced ratio of each task according to the formula (estimated execution time of the task + task waiting time) / estimated execution time of the task, and the greater the balanced ratio, the higher the priority, wherein the task waiting time is the difference between the current time and the task submission time). In the balanced-ordering strategy, the longer the waiting time or the shorter the execution time of each task, the greater the balanced ratio, so as to occupy a higher position in the ordering. For example, at the current time, if task a has waited for 120 minutes, task b has waited for 30 minutes, and task c has waited for 60 minutes, and the estimated execution times of a, b, and c are 10 minutes, 30 minutes, and 20 minutes respectively, then the balanced ratio of a is 13, the balanced ratio of b is 2, and the balanced ratio of c is 4, so the execution order will be a, c, and b in turn. Through the flexible selection of the above three strategies, a balance between efficiency priority and fair scheduling can be achieved according to the actual situation. The administrator can freely select the appropriate strategy according to the scene needs, complete more tasks in priority when the task quantity is large, ensure the fairness of first-come-first-served when the task is less, and use the balanced strategy when both need to be considered. The diversified priority strategy helps to improve the adaptability of the scheduling system in different application environments.

[0055] In an optional embodiment, the system further comprises a recording module configured to automatically record and store operation logs, wherein the operation logs are generated in the process of the control execution unit entering the execution interface to execute the load adjustment task, and the operation logs include identification information, operation content, and timestamp information of the control execution unit. The recording module stores the key process information of each task execution, which is convenient for auditing and analyzing the scheduling process in the future. If an exception or accident occurs, the operation details of each control unit can be traced through the logs to determine the responsibility.

[0056] It should be noted that the system provided in the above embodiments can be deployed in hardware according to the above module division, or a plurality of modules can be integrated in the same processing unit through software logic to realize. Similarly, the division of the modules is only for clear description of the scheme, and the specific name and division method of each functional module do not constitute a limitation to the present application. The method embodiment and the device embodiment of the present application belong to the same concept, and the specific implementation process has been described in the method part, which will not be repeated here.

[0057] The application further provides a computer readable storage medium, wherein instructions are stored, and when the instructions are executed by a processor, the processor executes the power grid physical feasible region aware sectioning method steps of any one of the above. The computer readable storage medium can include but is not limited to: a USB storage disk, a read-only memory, a random access memory, a mobile hard disk, a magnetic disk, an optical disk and other media capable of storing program codes. These instructions can be burned in the storage medium in advance, and when installed on a device with processing capability, the instructions can be called and run to realize the functions of the method of the application.

[0058] It should be noted that for each of the above method embodiments, in order to be simple, it is expressed as a series of actions, but it does not mean that the application limits the order of steps. Based on the idea of the application, some steps can change the execution order without affecting the function realization, or be executed in parallel. Secondly, those skilled in the art should also understand that each specific embodiment described in the specification is a preferred description of the technical solution of the application, and is not a limitation on the protection scope of the application. Any equivalent improvement or replacement made within the spirit and principles of the application should be included in the protection scope of the application.

[0059] Finally: the above only describes the preferred embodiments of the application and is not used to limit the application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the application should be included in the protection scope of the application.

Claims

1. An adaptive interval based demand side power load partitioning method, characterized by, The method comprises the following steps: obtaining power distribution network topology, node voltage, current and load power time series data, constructing a power flow linearization equation based on LinDistFlow approximation, calculating the sensitivity matrix of voltage to active and reactive power and the sensitivity matrix of current to active and reactive power, and using recursive least squares to identify the sensitivity matrix online; extracting the voltage influence intensity, current influence intensity, voltage margin and current margin characteristics of the node, and inputting the characteristics into the trained graph autoencoder for dimension reduction to generate a low-dimensional feature vector that preserves the topology and physical coupling information of the power distribution network; constructing a load conflict graph based on the low-dimensional feature vector, the edges of the load conflict graph representing the risk of exceeding the limit when two loads are adjusted simultaneously, and using a graph coloring algorithm to assign the load conflict graph nodes to the peak shaving period according to a predetermined rule to form a peak shaving scheme; combining the peak shaving scheme with the voltage margin and current margin constraints, using a differentiable clustering algorithm to divide the load intervals, and outputting the load interval division result; generating an interval scheduling parameter set based on the load interval division result, and outputting the parameter set to the control unit through a communication interface for subsequent execution and docking; constructing the load conflict graph includes: estimating the influence of concurrent adjustment of any two loads on branch current and node voltage according to the sensitivity matrix, and establishing an edge between the corresponding node pairs when any branch is occupied or any node margin is lower than a predetermined safety threshold to obtain an edge set of the load conflict graph; the graph coloring algorithm uses a greedy strategy combined with capacity constraints: the peak shaving period is assigned in descending order of node degree in the load conflict graph, nodes with edges are prohibited from falling into the same period, the number of peak shaving periods is within a predetermined range, and the length of the period is determined according to the predetermined rules of capacity constraints and scheduling period.

2. The adaptive interval-based demand-side power load division method according to claim 1, characterized in that: the online identification includes: presetting a sensitivity matrix element drift threshold, comparing the current value with the historical reference value in real time, controlling the distributed power source to inject a small disturbance within a predetermined amplitude range when the difference exceeds the threshold, collecting the node voltage and current data after the disturbance, and updating the sensitivity matrix using recursive least squares.

3. The adaptive interval-based demand side power load partitioning method according to claim 1, characterized in that: The training of the graph autoencoder includes: taking the node features and topology adjacency of the power distribution network within a predetermined historical time length as the training set, setting the input layer dimension corresponding to the number of node features, setting a predetermined number of hidden layers and using linear rectification activation, and simultaneously reconstructing the features and adjacency matrix at the output end, and optimizing the mean square error loss to a predetermined convergence threshold.

4. The adaptive interval-based demand side power load partitioning method according to claim 1, characterized in that, The initialization of the differentiable clustering includes: setting the interval prototype vector according to the number of peak shaving periods, initializing the prototype by calculating the mean value of the low-dimensional feature vector of the load in the corresponding period, calculating the belonging probability of each load to each interval using the Softmax function, and constraining the sum of the belonging probability to be equal to one.

5. The adaptive interval-based demand side power load partitioning method according to claim 4, characterized in that, The objective function of the differentiable clustering includes: the distance term of the load low-dimensional feature vector and the interval prototype vector to which it belongs, and the logarithmic barrier penalty term constructed by the voltage margin and the current margin, and the adaptive matrix estimation type optimizer is used to iterate within a predetermined learning rate interval until a predetermined convergence criterion is reached.

6. The adaptive bin-based demand side power load partitioning method according to claim 1, characterized in that, The method further comprises that the interval dispatching parameter set comprises an interval identifier, a node set, a planned adjustment amount of each node, and a safety threshold.

7. The adaptive interval-based demand side power load partitioning method according to claim 6, characterized in that, Before output, verification is performed by a communication authentication module, the verification comprising key check and identity matching; after authorization, the interval dispatching parameter set is issued by an instruction transmission module, during execution, node voltage and current data are collected in real time by a data monitoring module, when the safety threshold is exceeded, event information is stored by a data recording module and an abnormal code is returned.

8. A demand side power load partitioning system based on adaptive intervals for implementing the method of any of claims 1-7, characterized by, The method comprises: a receiving module configured to acquire power distribution network operation data and model parameters, the operation data comprising voltage, current of each node and power time series of each demand-side load, the model parameters comprising power distribution network topology structure and line parameters, if part of the parameters are unknown, they are recorded as to-be-identified parameters; a processing module configured to perform physical modeling processing on the data acquired by the receiving module, based on LinDistFlow approximate linearization power flow equation, to calculate sensitivity matrix of voltage to active power and reactive power, and sensitivity matrix of current to active power and reactive power, to perform online identification of the sensitivity matrix by using recursive least squares, and to combine node voltage influence strength, current influence strength, voltage margin, and current margin to construct a physical feature vector, and to generate a low-dimensional feature vector retaining topology and physical coupling information by using a graph autoencoder for dimension reduction; a determining module configured to set a physical constraint rule according to a feasible region limit of power distribution network voltage and current, to take the low-dimensional feature vector generated by the processing module as input, to divide the load interval by using a differentiable clustering algorithm, to minimize interval internal feature divergence and out-of-limit risk simultaneously by using a clustering objective function, and to output a load interval division result; a sending module configured to generate a dispatching instruction according to the load interval division result output by the determining module, to determine an instruction execution order according to a preset priority strategy, to send a dispatching instruction link to a corresponding control execution unit, and to monitor node voltage and branch current in real time, and to record an operation log comprising control execution unit identifier information, operation content, and time information synchronously.

Images