Power distribution network risk early warning and control method based on predictive power flow
Through multi-factor considerations, machine learning and distribution collaborative trend algorithms, combined with network reconstruction algorithms, the problem of insufficient load prediction of distribution networks is solved, accurate prediction and effective control of grid risks is achieved, and the safety of grid operation is improved. Sex and efficiency.
Patent Information
- Application Number
- CN202411012277.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-26
- Publication Date
- 2024-11-15
AI Technical Summary
The existing technology lacks accurate prediction of distribution network load, resulting in frequent grid accidents, low risk warning efficiency, and relying on manual experience makes it difficult to effectively control operating risks.
The multi-factor consideration machine learning method is used to predict the distribution network load, and combined with the transmission and distribution collaborative current algorithm, the optimal load transfer is achieved through the network reconstruction algorithm to achieve adaptive balance of the grid load.
It improves the accuracy of distribution network load prediction, realizes advance prediction and effective control of power grid risks, and improves the safety and efficiency of power grid operation.
Smart Images

Figure CN118970973A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of power system simulation, and in particular relates to a distribution network risk early warning and control method based on predicted power flow. Background Art
[0002] In recent years, with the impact of global warming, summer temperatures have soared year by year, residential electricity load has increased dramatically, and power grid equipment has been in a state of heavy overload for a long time during peak load periods. However, there is a certain cycle in the planning and transformation of power grids; at present, most of the solutions to the risk of equipment heavy overload are to adjust the operation mode of the power grid to keep the power load at a relatively balanced level to reduce the risk of power grid operation. However, affected by many factors such as meteorology, engineering construction, disasters at the site, distributed power sources, and new energy vehicles, power grid accidents usually occur under unknown circumstances. At present, there is a lack of accurate prediction of risks. After the risk is formed, most of the load transfer depends on manual experience, which has low efficiency, poor accuracy, and poor practical application effects, resulting in frequent power grid operation risks. Summary of the invention
[0003] The purpose of the present invention is to overcome the shortcomings of the existing technology and propose a distribution network risk warning and control method based on predicted flow. The present invention adopts a machine learning method that considers the influence of multiple factors to realize the prediction of the distribution network load, making up for the prediction distortion problem caused by insufficient consideration of influencing factors in the traditional distribution network load prediction; by combining the predicted data with the transmission and distribution coordinated flow algorithm to realize the distribution of the distribution network flow under the predicted state, the occurrence of distribution network risks can be effectively prevented; the heavy overload equipment of the power grid under the predicted state is used as the warning object and as the improvement target, and the optimal transfer of the distribution network load is finally realized through the network reconstruction algorithm, so as to achieve the adaptive balance of the power grid load.
[0004] The embodiment of the present invention proposes a distribution network risk warning and control method based on predicted power flow, including:
[0005] Using multi-source historical data related to the distribution network, a training set for line current prediction for each 10kV feeder in the distribution network and a training set for current prediction for each 10kV substation are constructed respectively;
[0006] The preset prediction algorithm is used to train each of the training sets respectively, and a line current prediction model of each 10kV feeder under the corresponding prediction algorithm and a substation current prediction model of each 10kV substation under the corresponding prediction algorithm are obtained respectively;
[0007] Select the prediction model with the highest fitting accuracy from the current prediction models of each line and each substation, and obtain the load prediction result of the prediction period based on the prediction model;
[0008] Based on the load forecasting results, the power flow distribution of the distribution network under the transmission and distribution coordination is calculated to achieve power grid risk warning and control in the forecast period.
[0009] In a specific embodiment of the present invention, the prediction algorithm includes: autoregressive model, moving average model, autoregressive sliding average model, autoregressive integrated sliding average model, random forest, gradient boosting tree.
[0010] In a specific embodiment of the present invention, the method further comprises:
[0011] When training the training set for line current prediction of each 10kV feeder, a feeder rule library is established, wherein the content of the feeder rule library includes: the load growth trend of the feeder in the same period, the load change trend at different temperatures, and the load change under maintenance;
[0012] When training the training set for each 10kV substation for substation current prediction, a substation rule base is established, and the content of the substation rule base includes: the load growth trend of the substation in the same period, the load change trend at different temperatures, and the load change in the presence of maintenance.
[0013] In a specific embodiment of the present invention, the method also includes: using the load forecast result as the input flow of the distribution network, combining with the current operation mode of the distribution network, using the forward and backward substitution method to distribute the input flow of the distribution network to each substation according to the historical usage ratio of the substation, and at the same time back-substituting the flow to the transmission network, the transmission network uses the branch flow method for iterative calculation, through multiple iterative calculations, until the transmission network and the distribution network are in a convergence state of flow balance, the distribution network flow distribution under transmission and distribution coordination is obtained.
[0014] In a specific embodiment of the present invention, the method further comprises:
[0015] Based on the load forecast results and the distribution network flow distribution, by comparing the difference between the current on each branch at each moment in the forecast period and the rated current, it is determined whether each device is in a severely overloaded state at the corresponding moment; during the comparison process, if it is determined that no device is in a severely overloaded state at any moment, the judgment will be continued at the next moment; if it is determined that a device is in a severely overloaded state at any moment, the device will be marked as a risky device at that moment.
[0016] In a specific embodiment of the present invention, the method further comprises:
[0017] Generate a load transfer plan for each risk device, including:
[0018] Search for the interconnection set feeders adjacent to any risky equipment and reconstruct the partitions by using the distribution network reconstruction division;
[0019] The load conditions of the feeder where the risky equipment is located and the load conditions of each interconnecting set feeder are recorded separately, and the overload amount of the risky equipment on the feeder where the risky equipment is located is recorded. The overload amount is transferred to each interconnecting set feeder to form a load transfer plan, and the transmission and distribution coordinated power flow algorithm is used to verify that no new overload problem will occur after the load transfer; wherein, the changed switch position is recorded during the transfer process to form the operation steps of each load transfer plan.
[0020] In a specific embodiment of the present invention, the method further comprises:
[0021] The load transfer plan for each risk device is evaluated, and after the evaluation, the optimal load transfer plan based on the risk device is marked.
[0022] The characteristics and beneficial effects of the present invention are:
[0023] The present invention proposes a distribution network risk warning and control method based on predicted power flow, which adopts a machine learning load forecasting scheme that integrates multi-source data, to make up for the prediction distortion problem caused by insufficient consideration of influencing factors in traditional distribution network load forecasting, and has the advantages of high prediction accuracy and strong adaptability to the environment. The present invention uses the predicted distribution network power flow distribution given by the transmission and distribution collaborative power flow algorithm based on the predicted load of the distribution network, effectively locates the risk points of the distribution network, and realizes the early prediction of risks. Finally, based on the load adaptive balancing optimization scheme proposed as the improvement target of the predicted risk point, it comprehensively considers the capacity constraints of the transmission and distribution network equipment, important users, and maintenance plans, and realizes the optimal transfer of the distribution network load, effectively improving the efficiency of load transfer that relies solely on manual analysis, with the characteristics of comprehensive risk control, and fully mobilizes the advantages of resources within the network. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] Figure 1 It is an overall flow chart of a distribution network risk warning and control method based on predicted power flow according to an embodiment of the present invention. DETAILED DESCRIPTION
[0025] The present invention proposes a distribution network risk warning and control method based on predicted flow. The technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.
[0026] The embodiment of the present invention proposes a distribution network risk warning and control method based on predicted power flow, including:
[0027] Using multi-source historical data related to the distribution network, a training set for line current prediction for each 10kV feeder in the distribution network and a training set for current prediction for each 10kV substation are constructed respectively;
[0028] The preset prediction algorithm is used to train each of the training sets respectively, and a line current prediction model of each 10kV feeder under the corresponding prediction algorithm and a substation current prediction model of each 10kV substation under the corresponding prediction algorithm are obtained respectively;
[0029] Select the prediction model with the highest fitting accuracy from the current prediction models of each line and each substation, and obtain the load prediction result of the prediction period based on the prediction model;
[0030] Based on the load forecasting results, the power flow distribution of the distribution network under the transmission and distribution coordination is calculated to achieve power grid risk warning and control in the forecast period.
[0031] In a specific embodiment of the present invention, the distribution network risk warning and control method based on the predicted flow has the following overall process: Figure 1 As shown, the following steps are included:
[0032] 1) Obtain multi-source historical data related to the distribution network.
[0033] In this embodiment, based on the control cloud platform, all 10kV feeders and substations of the regional distribution network, regional meteorological data, 10kV feeder historical data, distribution automation data, marketing procurement data, holiday time series, maintenance plans, industry types, power grid operation modes, and important user data are obtained and used as the basic data for distribution network load forecasting in the following steps.
[0034] For example: There are 400 10kV feeders and 23,000 distribution transformers in a certain area, involving the next day's meteorological data of 11 districts and counties, the historical current collection values of 10kV feeders within one month, the current real-time distribution automation data, the usage data of the distribution transformer within one month, the maintenance plan for the next week, and the important users and industry type data of the current power grid.
[0035] 2) Using the multi-source historical data collected in step 1), a training set for line current prediction for each 10kV feeder in the distribution network and a training set for current prediction for each 10kV substation are constructed; the specific steps are as follows:
[0036] 2-1) Using the multi-source historical data collected in step 1), a corresponding training set for line current prediction is established for each 10kV feeder in the distribution network. The input of each training sample in the training set corresponding to each 10kV feeder is the daily 96-point historical current value, the 24-hour meteorological data of the district or county to which the feeder belongs the next day, the maintenance plan information of the next day, and the planned grid operation mode, and the output is the 96-point current value of the feeder the next day.
[0037] In this embodiment, the training set extracts historical data and meteorological distribution patterns in the same period of a month to predict the current value in the same period of the next day, and finally outputs the current value of the feeder in the predicted period corresponding to the period of the input data in the historical data, forming a predicted current curve based on 96 points of the feeder for the next day.
[0038] 2-2) Using the multi-source historical data collected in step 1), a corresponding training set for current prediction is established for each 10kV substation in the distribution network. The input of each training sample in the training set corresponding to each 10kV substation is the 96-point current value of the day, the 24-hour meteorological data of the district or county to which the substation belongs the next day, the maintenance plan information of the next day, and the planned grid operation mode, and the output is the 96-point current value of the substation the next day.
[0039] In this embodiment, the training set extracts historical usage data and meteorological distribution patterns in the same period of a month, predicts the current value in the same period of the next day, and outputs the current value of the substation in the predicted period corresponding to the period of the input data in the historical data, forming a 96-point predicted current curve based on the substation for the next day.
[0040] 3) Using the preset prediction algorithm, each training set obtained in step 2) is trained respectively to obtain the line current prediction model of each 10kV feeder under the corresponding prediction algorithm and the substation current prediction model of each 10kV substation under the corresponding prediction algorithm; the specific steps are as follows:
[0041] 3-1) Select a prediction algorithm.
[0042] To ensure the accuracy of load forecasting, this embodiment makes predictions from two different dimensions: distribution network area and distribution network feeder. By comparing multiple algorithms, the most suitable algorithm for different scenarios is determined, and appropriate corrections can be made for unreasonable forecast data. At present, common forecasting algorithms are roughly divided into two types: one is a forecasting method based entirely on the pure time series of historical data of the predicted object, and the other is a comprehensive analysis based on the historical data of the predicted object and other objects related to it.
[0043] Specifically, based on the pure time series prediction method, four methods can be adopted, including AR model (autoregressive model), MA model (moving average model), ARMA model (autoregressive moving average model), and ARIMA model (autoregressive integrated moving average model); based on the prediction method of the correlation data of the prediction object, two methods can be adopted, including RF model (random forest) and GBDT model (gradient boosting tree).
[0044] 3-2) Using each prediction algorithm selected in step 3-1), respectively train the training set corresponding to each substation obtained in step 2) to obtain a substation current prediction model for each substation under the corresponding prediction algorithm.
[0045] Among them, during training, this embodiment needs to initialize the substation rule base based on the historical usage data and meteorological distribution laws in all time periods of a month extracted for the first time. The contents involved include: the load growth trend of the substation in the same period, the load change trend under different temperatures, and the load change under maintenance conditions, and finally form a rule base based on each substation; when predicting again, compare whether there is a difference between the currently extracted change law and the initial substation rule base, and identify the part that exceeds the error range as a new rule, that is, update the rule base to achieve the purpose of training the model, and finally form a prediction of the distribution network substation current.
[0046] Afterwards, the network topology is calculated using the grid operation mode obtained from step 1); based on the network topology, the prediction results of each trained substation current prediction model are aggregated to the 10kV feeder to form line prediction data.
[0047] 3-3) Using each prediction algorithm selected in step 3-1), train the training set of each 10kV feeder obtained in step 2) to obtain a line current prediction model for each 10kV feeder under the corresponding prediction algorithm.
[0048] Among them, during training, this embodiment needs to initialize the feeder rule base based on the historical current data and meteorological distribution laws in all time periods of a month extracted for the first time. The contents involved include: the load growth trend of the feeder in the same time period, the load change trend at different temperatures, and the load change under maintenance. Finally, a rule base based on each feeder is formed. When the prediction is made again, the current extracted change law is compared with the initial substation rule base to see if there is a difference. The part exceeding the error range is identified as a new rule, that is, the rule base is updated to achieve the purpose of training the model and finally form a prediction of the distribution network feeder current.
[0049] 4) Select the prediction model with the highest fitting accuracy from the line current prediction model and the substation current prediction model obtained in step 3), and obtain the load prediction result of the prediction period based on the prediction model.
[0050] In this embodiment, by establishing various rule bases, the prediction model is continuously trained to automatically discover the load operation characteristics of the distribution network at different times, and the operation characteristics are recorded in the rule base as prediction objects, and load prediction is performed using the new rule base.
[0051] This embodiment gradually analyzes the key influencing factors of the prediction accuracy gap through long-term model training, continuously updates the rule base, enhances the machine learning capability, and ultimately achieves load forecasting accuracy within the error range.
[0052] Furthermore, in the actual prediction process, some input parameters may be missing at different time points, and the prediction results of the algorithm need to be compared. In this embodiment, through comparison between multiple algorithms, the prediction method based on pure time series is more dependent on the continuity of historical data, and the prediction method based on the correlation data of the prediction object can be applied to scenarios where some parameters are missing, thereby determining the results of the most suitable prediction algorithm used by each feeder and substation in different scenarios, and ultimately realizing the prediction of the distribution network load.
[0053] Finally, by comparing the feeder load forecast results obtained using the substation current forecast model and the feeder load forecast results obtained using the feeder current forecast model with the actual load trend, the load forecast data corresponding to the output of the forecast model with the highest current fitting accuracy is selected as the load forecast result for the corresponding forecast period (in a specific embodiment of the present invention, the load forecast at 96 o'clock the next day), and it is used as the input of the transmission and distribution coordinated flow.
[0054] 5) Based on the load forecast results, calculate the distribution network flow distribution under transmission and distribution coordination.
[0055] The load forecast result obtained in step 4) is used as the input flow of the distribution network. Combined with the current operation mode of the distribution network, the input flow of the distribution network is distributed to each substation according to the historical consumption ratio of the substation using the forward-backward substitution method, and the flow is simultaneously back-substituted to the transmission network. The transmission network uses the branch flow method for iterative calculation. After multiple iterative calculations, the transmission network and the distribution network are both in a convergent state of flow balance.
[0056] 6) Based on the load forecast results, early warning of power grid risks during the forecast period is carried out.
[0057] Based on the load forecast result obtained in step 4), combined with the power flow distribution obtained in step 5), by comparing the difference between the current on each branch and the rated current at each moment in the forecast period, it is determined whether each device is in a heavy overload state at the corresponding moment. During the comparison process, if it is determined that no device is in a heavy overload state at the current moment, the judgment will be continued at the next moment; if it is determined that a device is in a heavy overload state at the current moment, the device will be marked as a risk device at that moment until the risk analysis of the forecast period, i.e. 96 o'clock the next day, is completed to achieve the prediction of risks.
[0058] Furthermore, in this embodiment, the equipment can be marked according to the risk in the real-time grid operation process and used as the risk equipment in the current grid operation. The risk analysis system in the real-time state performs analysis in a periodic calculation manner.
[0059] 7) Generate a load transfer plan for each risk device marked in step 6); specifically including:
[0060] Based on the power supply feeder where each risky device marked in step 6) is located, the adjacent interconnection set feeder is found, and the distribution network reconstruction is used to partition and reconstruct the partition flow.
[0061] For example, if the feeder where the risk equipment is located is A, and the adjacent interconnection set feeders are B, C, and D, then there are three possible paths for the load transfer of the risk equipment on feeder A, namely A to B, A to C, and A to D;
[0062] For risky equipment, by recording the current load conditions of the four feeders A, B, C, and D and the overload of risky equipment on feeder A, the overload on feeder A is transferred to feeders B, C, and D to form a new transfer plan. The transmission and distribution coordinated power flow algorithm is used to verify that no new overload problem will occur after the load transfer.
[0063] Among them, the changed switch positions are recorded during the transfer process to form the operation steps for each plan;.
[0064] During the reconstruction process of this embodiment, it is found that the load transfer amount on each feeder is not completely balanced. During the transfer process, it is necessary to consider the equipment capacity constraints, important users, and maintenance plans on the backup feeders B, C, and D to ensure that the power supply reliability will not be reduced. The overload on feeder A cannot be completely transferred to any of the backup feeders. There is a situation where one transfers to multiple feeders. After combining multiple factors, the adaptive balance of the load is achieved.
[0065] 8) Evaluate the load transfer plan for each risk equipment obtained in step 7).
[0066] In this embodiment, based on each power transfer scheme obtained in step 7), a score is given according to the number of switch operations, the effectiveness of load transfer, and the multi-dimensional evaluation scoring method of power supply reliability. A heuristic algorithm is used to find the optimal and suboptimal solutions and then compare them, so as to select the optimal equipment control scheme, realize early prediction of risks and adopt effective control plans.
[0067] In this embodiment, the scoring method is as follows (refer to the existing method):
[0068] The total score of the power transfer scheme is set to 100, the effectiveness of the power transfer scheme accounts for 50%, the number of switch operations of the power transfer scheme accounts for 20%, and the load recovery situation accounts for 30%. Let the input switch operation number index be x and the load recovery situation index be rate, then the scores are as follows:
[0069] When the transfer plan is valid, the score of the transfer plan is score = 100 × 0.5 + (100-(x-1) × 10) × 0.2 + rate × 100 × 0.3;
[0070] When the transfer plan is invalid, the score of the transfer plan is score = (100-(x-1)×10)×0.2+rate×100×0.3;
[0071] Among them, the proportion of load restored through the power transfer plan to the power loss load caused by maintenance is the load recovery situation.
[0072] Ultimately, the transfer plan with the highest score is the optimal transfer plan.
[0073] Furthermore, the method described in this embodiment also includes:
[0074] Under the real-time operation risk analysis of the power grid, the load transfer logic is automatically triggered based on the risk equipment, and the transfer plan is visualized to assist in the regulation and operation decision-making.
[0075] Under the risk analysis under the predicted state of the power grid, the load transfer logic is triggered based on the risk equipment at each time point of 96 o'clock the next day, and the selected transfer plan is stored in the decision database for early avoidance of possible risks of the next day to ensure the safe operation of the power grid.
[0076] The above is a preferred embodiment of the present disclosure, which can be understood and implemented by ordinary technicians in the field without creative work. It should be pointed out that ordinary technicians in this technical field can make several improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.
Claims
1. A distribution network risk warning and control method based on predicted power flow, characterized in that: include: Using multi-source historical data related to the distribution network, a training set for line current prediction for each 10kV feeder in the distribution network and a training set for current prediction for each 10kV substation are constructed respectively; The preset prediction algorithm is used to train each of the training sets respectively, and the line current prediction model of each 10kV feeder under the corresponding prediction algorithm and the substation current prediction model of each 10kV substation under the corresponding prediction algorithm are obtained respectively; Select the prediction model with the highest fitting accuracy from the current prediction models of each line and each substation, and obtain the load prediction result of the prediction period based on the prediction model; Based on the load forecasting results, the power flow distribution of the distribution network under the transmission and distribution coordination is calculated to achieve power grid risk warning and control in the forecast period.
2. The method according to claim 1, characterized in that: The prediction algorithms include: autoregressive model, moving average model, autoregressive sliding average model, autoregressive integrated sliding average model, random forest, and gradient boosting tree.
3. The method according to claim 1, characterized in that The method further comprises: When training the training set for line current prediction of each 10kV feeder, a feeder rule library is established, wherein the content of the feeder rule library includes: the load growth trend of the feeder in the same period, the load change trend at different temperatures, and the load change under maintenance; When training the training set for each 10kV substation for substation current prediction, a substation rule base is established, and the content of the substation rule base includes: the load growth trend of the substation in the same period, the load change trend at different temperatures, and the load change in the presence of maintenance.
4. The method according to claim 1, characterized in that: The method also includes: using the load forecast result as the input flow of the distribution network, combining it with the current operation mode of the distribution network, using the forward-backward substitution method to distribute the input flow of the distribution network to each substation according to the historical consumption ratio of the substation, and simultaneously back-substituting the flow to the transmission network, the transmission network adopts the branch flow method for iterative calculation, and through multiple iterative calculations, until the transmission network and the distribution network are both in a convergent state of flow balance, the distribution network flow distribution under the transmission and distribution coordination is obtained.
5. The method according to claim 1, characterized in that: The method further comprises: Based on the load forecast results and the distribution network flow distribution, by comparing the difference between the current on each branch at each moment in the forecast period and the rated current, it is determined whether each device is in a severely overloaded state at the corresponding moment; during the comparison process, if it is determined that no device is in a severely overloaded state at any moment, the judgment will be continued at the next moment; if it is determined that a device is in a severely overloaded state at any moment, the device will be marked as a risky device at that moment.
6. The method according to claim 5, characterized in that The method further comprises: Generate a load transfer plan for each risk device, including: Search for the interconnection set feeders adjacent to any risky equipment and reconstruct the partitions by using the distribution network reconstruction division; The load conditions of the feeder where the risky equipment is located and the load conditions of each interconnecting set feeder are recorded separately, and the overload amount of the risky equipment on the feeder where the risky equipment is located is recorded. The overload amount is transferred to each interconnecting set feeder to form a load transfer plan, and the transmission and distribution coordinated power flow algorithm is used to verify that no new overload problem will occur after the load transfer; wherein, the changed switch position is recorded during the transfer process to form the operation steps of each load transfer plan.
7. The method according to claim 6, characterized in that The method further comprises: The load transfer plan for each risk device is evaluated, and after the evaluation, the optimal load transfer plan based on the risk device is marked.
Citation Information
Patent Citations
Active power distribution network situation awareness system and method based on distributed monitoring and multi-source information fusion
CN107274115A
Power distribution network operation situation prediction method based on big data
CN111144638A
Distribution transformer load prediction and heavy overload early warning method and system
CN115587672A
