A short-term load forecasting method and device

Abstract

The present invention belongs to the technical field of load forecasting, and particularly relates to a short-term load forecasting method and device. The method includes: obtaining a load forecasting model for each load component in the network-supplied load; using the load forecasting models of each load component to obtain the predicted values of each load component; obtaining a network-supplied load forecasting model; and taking the predicted values of each load component as the input of the network-supplied load forecasting model to obtain the predicted value of the network-supplied load. The technical solution provided by this application can not only perform feature analysis and prediction on each load component, improving the refinement degree of prediction, but also effectively decompose each load component in the network-supplied load, which is beneficial for dispatching personnel to understand the composition components of the network-supplied load and their change rules, and improving the accuracy of network-supplied load forecasting.

Description

Technical Field

[0001] The present invention belongs to the technical field of load forecasting, and in particular relates to a short-term load forecasting method and device. Background Art

[0002] The short-term load forecast of the power system is the basis for the preparation of power generation plans and the security verification and analysis of the power grid. Its forecast level will directly affect the economy, stability and safety of the power system operation. In recent years, with the continuous upgrading and adjustment of my country's industrial structure, the continuous increase in the proportion of distributed power sources and flexible loads, and the differences in regional meteorology, the load characteristics of some regions have changed significantly. Especially during holidays or when the weather changes suddenly, the load fluctuates greatly, and the meteorology in various cities is quite different. Different types of user loads show different load characteristics in scenarios such as holidays or sudden weather changes. Therefore, the existing load forecasting faces great difficulties and challenges.

[0003] The method of using only the total system load as the prediction target historical sample for prediction model training is difficult to fully reflect the load information of each power user group. It is difficult to conduct more detailed positioning analysis for some abnormal load fluctuations. The prediction accuracy is limited and it is difficult to meet the refined prediction requirements.

[0004] Although the traditional industry classification method is used to classify user-side loads, the system-level load has been effectively refined. However, since each industry contains users with different electricity consumption characteristics, it is also difficult to explore the changing patterns of total loads in different industries. In addition, the industry load base is greatly reduced relative to the system load. Therefore, the prediction effect of the system load prediction method is limited. The feasibility of the method of predicting by industry and then aggregating the system-level load prediction is limited to a certain extent due to the randomness of the prediction error distribution of each industry.

[0005] Generally speaking, existing technologies mostly use indirect explanations to forecast loads, and cannot directly explain the causes of load changes from a physical perspective. That is, by obtaining relevant data and using relevant prediction models in artificial intelligence to predict load forecasting results, it is difficult to better explain the mechanism of load changes in this way. Summary of the invention

[0006] In view of this, the purpose of the present invention is to overcome the deficiencies of the prior art and provide a short-term load forecasting method and device to solve the problem that the prior art cannot accurately predict the grid load.

[0007] According to a first aspect of an embodiment of the present application, a short-term load forecasting method is provided, the method comprising:

[0008] Obtaining the load forecasting model of each load component in the grid load;

[0009] Use the load prediction models of the respective load components to obtain the predicted values of the respective load components;

[0010] Obtain a network-supplied load prediction model;

[0011] Use the predicted values of the respective load components as the input of the network-supplied load prediction model to obtain the predicted value of the network-supplied load.

[0012] Preferably, the respective load components include: load data of large power users at 35 kV and above, output load data of local power plants, and load data of 10 kV public dedicated lines.

[0013] Preferably, the obtaining of the load prediction models of the respective load components in the network-supplied load includes: establishing a first load prediction model;

[0014] The establishment of the first load prediction model includes: when the load component is the load of large power users at 35 kV and above, collect the historical load data and historical load influencing factor data of large power users at 35 kV and above in the historical time period;

[0015] Use the historical load influencing factor data as the input layer training samples of the PSO-LSSVM model, and use the historical load data of large power users at 35 kV and above in the historical time period as the output layer training samples of the PSO-LSSVM model for training to obtain the first load prediction model.

[0016] Preferably, the obtaining of the load prediction models of the respective load components in the network-supplied load includes: establishing a second load prediction model;

[0017] The establishment of the second load prediction model includes: when the load component is the output load of the local power plant, collect the historical load data and historical load influencing factor data of the output of the local power plant in the historical time period;

[0018] Use the historical load influencing factor data as the input layer training samples of the PSO-LSSVM model, and use the historical load data of the output of the local power plant in the historical time period as the output layer training samples of the PSO-LSSVM model for training to obtain the second load prediction model.

[0019] Preferably, the obtaining of the load prediction models of the respective load components in the network-supplied load includes: establishing a third load prediction model;

[0020] The establishment of the third load prediction model includes:

[0021] Step 1: When the load component is the 10kV public dedicated line load, collect the historical load data and historical load influencing factor data of each line in the 10kV public dedicated line during the historical time period;

[0022] Step 2: Let the historical load data of each line in the 10kV public dedicated line during the historical time period be the basic data, and use the basic data to obtain the typical load curves of each line in the 10kV public dedicated line;

[0023] Step 3: Let the cluster center K = 1;

[0024] Step 4: Use the K-means clustering algorithm to cluster the typical load curves of each line in the 10kV public dedicated line to obtain the clustering result;

[0025] Step 5: Determine whether the cluster center K is greater than or equal to M. If K is greater than or equal to M, execute Step 6; if K is less than M, let K = K + 1 and return to Step 4; where M is a preset value and M is a positive integer;

[0026] Step 6: Use the elbow method to obtain the optimal cluster center, and use the clustering result corresponding to the optimal cluster center as the final clustering result;

[0027] Step 7: Use the correlation method to calculate the correlation results between various influencing factor data in the historical load data and historical load influencing factor data of the 10kV public dedicated line corresponding to each cluster in the optimal clustering result, and sort all the correlation results. Let the top three ranked correlation results correspond to a certain type of influencing factor as the main influencing factor of this cluster;

[0028] Step 8: Use the historical load influencing factor data corresponding to each cluster in the clustering result as the input layer training samples of the PSO-LSSVM model, and use the historical load data of the 10kV public dedicated line corresponding to each cluster in the clustering result as the output layer training samples of the PSO-LSSVM model for training to obtain the third load prediction model for various influencing factors, that is, the third load prediction model for each cluster.

[0029] Preferably, the collection of the historical load data of each line in the 10kV public dedicated line during the historical time period includes:

[0030] Combine the historical load data of the lines with transfer supply relationships in the 10kV public dedicated line during the historical time period to obtain combined data;

[0031] The historical load data of the lines without transfer supply relationships in the 10kV public dedicated line during the historical time period and the combined data constitute the historical load data of each line in the 10kV public dedicated line during the historical time period.

[0032] Preferably, in step 2, obtaining the typical load curves of each line of the 10 kV public and dedicated lines by using the basic data includes:

[0033] Step 21: When the load value at a certain moment point in a certain piece of data in the basic data is 0, remove this piece of data to obtain the preprocessed basic data;

[0034] Among them, the historical load data of each line in the 10 kV public and dedicated lines every day within the historical time period is one piece of data;

[0035] Step 22: Using the preprocessed basic data, calculate the average load of each moment point of each line in the 10 kV public and dedicated lines, and generate the initial load curve of each line with the average load of each moment point of each line in the 10 kV public and dedicated lines;

[0036] Step 23: Using the preprocessed basic data, generate the daily load curve of each line in the 10 kV public and dedicated lines within the historical time period;

[0037] Step 24: Calculate the deviation rate of each moment point in the daily load curve of each line in the 10 kV public and dedicated lines within the historical time period from each moment point in the initial load curve;

[0038] When there is a deviation rate of each moment point in the daily load curve from each moment point in the initial load curve greater than the preset deviation rate threshold of each moment point, remove this daily load curve to obtain the typical load curves of each line of the 10 kV public and dedicated lines.

[0039] Preferably, obtaining the predicted values of the respective load components by using the load prediction models of the respective load components includes:

[0040] Using the load influencing factor data at the future time to be predicted as the input of the load prediction models of the respective load components to obtain the predicted values of the respective load components.

[0041] Preferably, obtaining the network supply load prediction model includes:

[0042] Collect the historical load data of each load component and the historical load data of the network supply load within the historical time period;

[0043] Using the historical load data of each load component within the historical time period as the input layer training samples of the PSO-LSSVM model, and using the historical load data of the network supply load within the historical time period as the output layer training samples of the PSO-LSSVM model for training to obtain the network supply load prediction model.

[0044] According to the second aspect of the embodiments of the present application, there is provided a short-term load prediction device, and the device includes:

[0045] A first acquisition module, configured to acquire a load prediction model for each load component in the network-supplied load;

[0046] A second acquisition module, configured to use the load prediction models of the respective load components to acquire predicted values of the respective load components;

[0047] A third acquisition module, configured to acquire a network-supplied load prediction model;

[0048] A fourth acquisition module, configured to use the predicted values of the respective load components as inputs to the network-supplied load prediction model to acquire a predicted value of the network-supplied load.

[0049] Preferably, the respective load components include: load data of large power users at 35 kV and above, output load data of local power plants, and load data of 10 kV public dedicated lines.

[0050] Preferably, the first acquisition module includes: a first model establishment unit, configured to establish a first load prediction model;

[0051] The first model establishment unit is specifically configured to:

[0052] When the load component is the load of large power users at 35 kV and above, collect historical load data and historical load influencing factor data of large power users at 35 kV and above in a historical time period;

[0053] Use the historical load influencing factor data as the input layer training samples of the PSO-LSSVM model, and use the historical load data of large power users at 35 kV and above in the historical time period as the output layer training samples of the PSO-LSSVM model for training to acquire the first load prediction model.

[0054] Preferably, the first acquisition module further includes: a second model establishment unit, configured to establish a second load prediction model;

[0055] The second model establishment unit is specifically configured to:

[0056] When the load component is the output load of the local power plant, collect historical load data and historical load influencing factor data of the output of the local power plant in a historical time period;

[0057] Use the historical load influencing factor data as the input layer training samples of the PSO-LSSVM model, and use the historical load data of the output of the local power plant in the historical time period as the output layer training samples of the PSO-LSSVM model for training to acquire the second load prediction model.

[0058] Preferably, the first acquisition module further includes: a third model establishment unit configured to establish a third load prediction model;

[0059] The third model establishment unit includes:

[0060] An acquisition subunit configured to, when the load component is the 10kV public dedicated line load, acquire the historical load data and historical load influencing factor data of each line in the 10kV public dedicated line within a historical time period;

[0061] A first acquisition subunit configured to use the historical load data of each line in the 10kV public dedicated line within the historical time period as basic data, and use the basic data to obtain the typical load curves of each line in the 10kV public dedicated line;

[0062] A first determination subunit configured to set the cluster center K = 1;

[0063] A clustering subunit configured to perform clustering on the typical load curves of each line in the 10kV public dedicated line by using the K-means clustering algorithm to obtain a clustering result;

[0064] A judgment subunit configured to judge whether the cluster center K is greater than or equal to M. If K is greater than or equal to M, then execute the second acquisition subunit; if K is less than M, then set K = K + 1 and return to the clustering subunit; where M is a preset value and M is a positive integer;

[0065] A second acquisition subunit configured to obtain the optimal cluster center by using the elbow method, and use the clustering result corresponding to the optimal cluster center as the final clustering result;

[0066] A second determination subunit configured to use the correlation method to calculate the correlation results between various influencing factor data in the historical load data and historical load influencing factor data of the 10kV public dedicated line corresponding to each cluster in the optimal clustering result, and sort all the correlation results, and set the certain type of influencing factor corresponding to the top three correlation results as the main influencing factor of the cluster;

[0067] A training subunit configured to use the historical load influencing factor data corresponding to each cluster in the clustering result as the input layer training samples of the PSO-LSSVM model, and use the historical load data of the 10kV public dedicated line corresponding to each cluster in the clustering result as the output layer training samples of the PSO-LSSVM model for training to obtain the third load prediction models of various influencing factors, that is, the third load prediction models of each cluster.

[0068] Preferably, the acquisition subunit is specifically configured to:

[0069] Combine the historical load data of the lines with transfer relationship in the 10kV public and dedicated lines within the historical time period to obtain combined data;

[0070] The historical load data of the lines in the 10kV public and dedicated lines that do not have a transfer relationship in the historical time period and the combined data constitute the historical load data of each line in the 10kV public and dedicated lines in the historical time period.

[0071] Preferably, the first acquisition subunit is specifically used for:

[0072] When a load value of a certain data in the basic data at a certain time point is 0, the data is removed to obtain the pre-processed basic data; wherein the historical load data of each line in the 10kV public dedicated line every day in the historical time period is one piece of data;

[0073] Utilizing the pre-processed basic data, calculating the average load of each line in the 10kV public dedicated line at each time point, and generating the initial load curve of each line based on the average load of each line in the 10kV public dedicated line at each time point;

[0074] Using the pre-processed basic data, generate a daily load curve for each line in the 10kV public dedicated line within a historical period;

[0075] Calculate the deviation rate of each time point in the daily load curve of each line in the 10kV public dedicated line during the historical period from each time point in the initial load curve;

[0076] When the deviation rate between each time point in the daily load curve and each time point in the initial load curve is greater than the preset deviation rate threshold of each time point, the daily load curve is eliminated to obtain the typical load curve of each line of the 10kV public dedicated line.

[0077] Preferably, the second acquisition module is specifically used to:

[0078] The load influencing factor data of the future time to be predicted is used as the input of the load prediction model of each load component to obtain the predicted value of each load component.

[0079] Preferably, the third acquisition module is specifically used for:

[0080] Collecting historical load data of each load component and historical load data of the network supply load within a historical time period;

[0081] Using the historical load data of each load component within the historical time period as the training samples for the input layer of the PSO-LSSVM model, and using the historical load data of the network-supplied load within the historical time period as the training samples for the output layer of the PSO-LSSVM model for training, to obtain the network-supplied load prediction model.

[0082] According to the third aspect of the embodiments of the present application, there is provided a readable storage medium, on which an executable program is stored, and when the executable program is executed by a processor, the steps of the above short-term load prediction method are implemented.

[0083] The beneficial effects that can be achieved by the present invention adopting the above technical solutions include: by obtaining the load prediction models of each load component in the network-supplied load, using the load prediction models of each load component to obtain the predicted values of each load component, and by obtaining the network-supplied load prediction model, using the predicted values of each load component as the input of the network-supplied load prediction model to obtain the predicted value of the network-supplied load, not only can the characteristics of each load component be analyzed and predicted, improving the prediction refinement degree; but also the effective decomposition of each load component in the network-supplied load is realized, which is beneficial for dispatchers to understand the composition components and their change rules of the network-supplied load, and improves the accuracy of the network-supplied load prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings without creative efforts.

[0085] Figure 1 is a flowchart of a short-term load prediction method shown according to an exemplary embodiment;

[0086] Figure 2 is a flowchart of the K-means clustering algorithm shown according to an exemplary embodiment;

[0087] Figure 3 is a load transfer relationship diagram of 10kV public dedicated lines shown according to an exemplary embodiment;

[0088] Figure 4 is a flowchart of obtaining the typical load curves of each 10kV public dedicated line using basic data shown according to an exemplary embodiment;

[0089] Figure 5 is a structural block diagram of a short-term load prediction device shown according to an exemplary embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0090] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be described in detail below. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other implementation manners obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope protected by the present invention.

[0091] Figure 1 is a flowchart of a short-term load forecasting method shown according to an exemplary embodiment. As Figure 1 shown, this method can be but is not limited to being used in a terminal, and includes the following steps:

[0092] Step 101: Obtain the load forecasting models of each load component in the network-supplied load;

[0093] Step 102: Use the load forecasting models of each load component to obtain the predicted values of each load component;

[0094] Step 103: Obtain the network-supplied load forecasting model;

[0095] Step 104: Use the predicted values of each load component as the input of the network-supplied load forecasting model to obtain the predicted value of the network-supplied load.

[0096] A short-term load forecasting method provided by an embodiment of the present invention, by obtaining the load forecasting models of each load component in the network-supplied load, using the load forecasting models of each load component to obtain the predicted values of each load component, by obtaining the network-supplied load forecasting model, using the predicted values of each load component as the input of the network-supplied load forecasting model to obtain the predicted value of the network-supplied load, can not only perform feature analysis and prediction on each load component, improving the prediction refinement degree; but also realizes the effective decomposition of each load component in the network-supplied load, which is beneficial for dispatchers to understand the composition components and their change rules of the network-supplied load, and improves the accuracy of the network-supplied load forecasting.

[0097] Further, each load component includes: load data of large power users with 35 kV and above, output load data of local power plants, and load data of 10 kV public special lines.

[0098] Further, step 101 of obtaining the load forecasting models of each load component in the network-supplied load includes: establishing a first load forecasting model;

[0099] Establishing the first load forecasting model includes: when the load component is the load of large power users with 35 kV and above, collecting the historical load data and historical load influencing factor data of large power users with 35 kV and above in the historical time period;

[0100] Using the historical load influencing factor data as the training samples for the input layer of the PSO-LSSVM model, and using the historical load data of large power users with 35 kV and above in the historical time period as the training samples for the output layer of the PSO-LSSVM model for training, to obtain the first load forecasting model.

[0101] For example, assuming to predict the load of large power users with 35 kV and above at 0:00 on February 10th, the load influencing factor data at 0:00 on February 10th is input into the first load forecasting model to obtain the predicted load of large power users with 35 kV and above at 0:00 on February 10th.

[0102] Further, step 101 for obtaining the load forecasting models of each load component in the network supply load includes: establishing a second load forecasting model;

[0103] Establishing the second load forecasting model includes: when the load component is the output load of local power plants, collecting the historical load data and historical load influencing factor data of the output of local power plants in the historical time period;

[0104] Using the historical load influencing factor data as the training samples for the input layer of the PSO-LSSVM model, and using the historical load data of the output of local power plants in the historical time period as the training samples for the output layer of the PSO-LSSVM model for training, to obtain the second load forecasting model.

[0105] For example, assuming to predict the output load of local power plants at 0:00 on February 10th, the load influencing factor data at 0:00 on February 10th is input into the second load forecasting model to obtain the predicted output load of local power plants at 0:00 on February 10th.

[0106] Further, step 101 for obtaining the load forecasting models of each load component in the network supply load includes: establishing a third load forecasting model;

[0107] Establishing the third load forecasting model includes:

[0108] Step 11: When the load component is the 10 kV public dedicated line load, collect the historical load data and historical load influencing factor data of each line in the 10 kV public dedicated line in the historical time period;

[0109] Step 12: Let the historical load data of each line in the 10 kV public dedicated line in the historical time period be the basic data, and use the basic data to obtain the typical load curves of each line in the 10 kV public dedicated line;

[0110] Step 13: Let the cluster center K = 1;

[0111] Step 14: Use the K-means clustering algorithm to cluster the typical load curves of each line in the 10 kV public dedicated line to obtain the clustering result;

[0112] Step 15: Determine whether the cluster center K is greater than or equal to M. If K is greater than or equal to M, execute Step 16; if K is less than M, set K = K + 1 and return to Step 14; where M is a preset value and M is a positive integer.

[0113] Step 16: Use the elbow method to obtain the optimal cluster center, and take the clustering result corresponding to the optimal cluster center as the final clustering result.

[0114] Step 17: Use the correlation method to calculate the correlation results between various influencing factor data in the historical load data and historical load influencing factor data of the 10kV public dedicated lines corresponding to each cluster in the optimal clustering result, and sort all the correlation results. Let the top three ranked correlation results correspond to a certain type of influencing factor, which is the main influencing factor of the cluster.

[0115] Step 18: Use the historical load influencing factor data corresponding to each cluster in the clustering result as the input layer training samples of the PSO-LSSVM model, and use the historical load data of the 10kV public dedicated lines corresponding to each cluster in the clustering result as the output layer training samples of the PSO-LSSVM model for training to obtain the third load prediction model for various influencing factors, that is, the third load prediction model for each cluster.

[0116] It should be noted that the specific value of the "preset value M" in the embodiments of the present invention is not limited. In some embodiments, it can be set by those skilled in the art according to experimental data or expert experience, etc.

[0117] It can be understood that using the K-means clustering algorithm to cluster the typical load curves of each line of the 10kV public dedicated line aims to determine the main influencing factors of the historical load data of the 10kV public dedicated line corresponding to each cluster.

[0118] In some embodiments, the historical load influencing factors may include, but are not limited to: temperature, humidity, output in the previous two days, air pressure, rainfall, etc. Specifically, the historical load influencing factors can be obtained from, but are not limited to, the meteorological bureau.

[0119] For example, assuming that the load of the 10kV public dedicated line at 0:00 on February 10 is predicted, then the load influencing factor data at 0:00 on February 10 is input into the third load prediction model corresponding to each cluster to obtain the predicted load corresponding to each cluster. Let the load value obtained by adding the predicted loads corresponding to all clusters be the final predicted load of the 10kV public dedicated line at 0:00 on February 10.

[0120] To establish a prediction model for each load component, first, obtain all historical load data, relevant meteorological data, and the innovations of each user in various clusters. Then, use correlation calculation to determine the main influencing factors affecting each load component. Next, predict the load of each cluster based on the PSO-LSSVM algorithm. Finally, analyze and correct the prediction results and save the load prediction model for each cluster.

[0121] In some embodiments, assuming that the typical load curves of each line of the 10kV public dedicated line are used as the sample set, for the K-means algorithm, first randomly select K points from the sample set, and each point represents the initial clustering center of each cluster. Then, calculate the Euclidean distance from each remaining sample to the clustering center, assign it to the cluster closest to it, and recalculate the average value of each cluster. This process is repeated continuously until the squared error criterion function is minimized and unchanged. The definition formula of the squared error criterion is:

[0122]

[0123] where K is the number of clusters, cluster A is the number of samples in the A-th class, and ave A is the mean value of the samples in the A-th class.

[0124] The K-means clustering algorithm is a typical distance-based clustering algorithm that uses the distance between points as the similarity evaluation index, that is, it is considered that the closer the distance between two objects, the greater the similarity. Suppose the typical electricity consumption behavior patterns of power users X and Y contain N features. The formula for calculating the Euclidean distance between the two users is:

[0125]

[0126] In each iteration of the K-means clustering algorithm, it is necessary to determine whether each sample data is correctly divided into the cluster. If not, it is readjusted. After all data is adjusted, the cluster center is modified for the next iteration calculation. If each data sample is assigned to the correct cluster in a certain iteration, the clustering center is no longer adjusted. When the clustering center is stable and no longer changes, it is marked that the objective function converges and the algorithm ends. Finally, evaluate the clustering result. The flow chart of the K-means clustering algorithm is as Figure 2 shown.

[0127] In some embodiments, the main index used by the elbow method is the sum of squared errors SSE (Sum of the Squared Errors).

[0128]

[0129] In the formula: C iis the i-th cluster class, p is a sample point of C i and m i is the centroid of C i (i.e., the mean of all samples in C i ), SSE is the clustering error of all samples, representing the quality of the clustering effect.

[0130] As the number of clusters K increases, the sample division will be more refined, and the aggregation degree of each cluster will gradually increase. Then, the sum of squared errors SSE will naturally gradually decrease.

[0131] When K is less than the true number of clusters, since the increase of K will greatly increase the aggregation degree of each cluster, the decline rate of SSE will be large. When K reaches the true number of clusters, the return of the aggregation degree obtained by increasing K will quickly become small. Therefore, the decline rate of SSE will suddenly decrease and then level off as K continues to increase. That is to say, the relationship graph between SSE and K is in the shape of an elbow, and the K value corresponding to this elbow is the true number of clusters of the data, that is, the optimal cluster center.

[0132] Furthermore, in step 11, the historical load data of each line in the 10kV public and dedicated lines during the historical time period is collected, including:

[0133] Combining the historical load data of the lines with transfer supply relationships in the 10kV public and dedicated lines during the historical time period to obtain combined data;

[0134] The historical load data of the lines without transfer supply relationships in the 10kV public and dedicated lines during the historical time period and the combined data constitute the historical load data of each line in the 10kV public and dedicated lines during the historical time period.

[0135] In some embodiments, as Figure 3 shown, when the 10kV public and dedicated line is operating normally, switch 1 is closed and switch 2 is open. Metering point 1 meters the data of load 1 and load 2, and metering point 2 meters the data of load 3. When switch 1 needs to be repaired, to ensure the normal power supply of load 2, switch 2 needs to be closed. At this time, metering point 1 meters the data of load 1, and metering point 2 meters the data of load 2 and load 3.

[0136] For 10kV public dedicated lines, due to events such as maintenance, it is necessary to change their original operation mode, which affects the metering of load data at each 10kV checkpoint, and further leads to the lack of predictability of load data. To avoid the impact of the power transfer relationship on load forecasting, it is necessary to preliminarily combine the 10kV public dedicated lines with power transfer relationships based on the historical operation mode ledger and line topology, and then adjust the combination of 10kV public dedicated lines based on the real-time maintenance plan. Thus, the combination of lines is completed. By considering the power transfer relationship between 10kV public dedicated lines, the lines with power transfer relationships are combined into independent power consumption units to obtain combined data, thereby effectively avoiding the impact of power transfer on the regularity of the collected load forecasting data. Therefore, regarding the power transfer relationship, assume that the first to t 1 lines are lines without power transfer relationships, and the t 1 +1 to t 1 +k-1 lines are lines with power transfer relationships. Then, the load of each line of the 10kV public dedicated line at the d-th time point is determined according to the following formula

[0137]

[0138] Among them, the load of the (t 1 +1)-th line of the 10kV public dedicated line at the d-th time point is determined according to the following formula

[0139]

[0140] The load of the (t 1 +2)-th line of the 10kV public dedicated line at the d-th time point is determined according to the following formula

[0141]

[0142] The load of the (t 1 +k-1)-th line of the 10kV public dedicated line at the d-th time point is determined according to the following formula

[0143]

[0144] In the above formula, d ∈ [1, D], where D is the total number of time points in a day; t 2 is the total number of branch lines in the (t 1 +1)-th line, is the load of the i 1 -th branch line in the (t 2 +1)-th line; t 3 is the total number of branch lines in the (t 1 +2)-th line, is the load of the (t1 + The load of the i-th 3 sub-line among the 2 lines; t k is the t-th 1 + The total number of sub-lines among the k - 1 lines, and is the load of the i-th 1 sub-line among the k - 1 lines at the t-th k time.

[0145] Furthermore, the cyclic average method is used to calculate the typical load curves of each line of the 10kV public and dedicated lines. As Figure 4 shown, in step 12, the typical load curves of each line of the 10kV public and dedicated lines are obtained by using the basic data, including:

[0146] Step 121: When the load value at a certain time point in a certain piece of data in the basic data is 0, remove this piece of data to obtain the preprocessed basic data;

[0147] Among them, the historical load data of each line of the 10kV public and dedicated lines within the historical time period is one piece of data;

[0148] Step 122: Use the preprocessed basic data to calculate the average load of each line of the 10kV public and dedicated lines at each time point, and generate the initial load curve of each line based on the average load of each line of the 10kV public and dedicated lines at each time point;

[0149] Step 123: Use the preprocessed basic data to generate the daily load curve of each line of the 10kV public and dedicated lines within the historical time period;

[0150] Step 124: Calculate the deviation rate between each time point in the daily load curve of each line of the 10kV public and dedicated lines within the historical time period and each time point in the initial load curve;

[0151] When there is a deviation rate between each time point in the daily load curve and each time point in the initial load curve that is greater than the preset deviation rate threshold for each time point, remove this daily load curve to obtain the typical load curves of each line of the 10kV public and dedicated lines.

[0152] Furthermore, step 102 includes:

[0153] Using the load influencing factor data at the future time to be predicted as the input of the load prediction model for each load component to obtain the predicted values of each load component.

[0154] Specifically, using the load influencing factor data at the future time to be predicted as the input of the first load prediction model to obtain the load prediction value of large power users above 35kV at the future time to be predicted.

[0155] Specifically, using the load influencing factor data at the future time to be predicted as the input of the second load prediction model, the predicted value of the local power plant output load at the future time to be predicted is obtained.

[0156] Specifically, using the load influencing factor data at the future time to be predicted as the input of the third load prediction model corresponding to all clusters, the predicted load values corresponding to each cluster are obtained, and the load value obtained by adding the predicted load values corresponding to all clusters is the predicted value of the 10kV public and dedicated line load at the future time to be predicted.

[0157] Further, step 103 includes:

[0158] Collect the historical load data of each load component and the historical load data of the network-supplied load within the historical time period;

[0159] Using the historical load data of each load component within the historical time period as the training sample of the input layer of the PSO-LSSVM model, and using the historical load data of the network-supplied load within the historical time period as the training sample of the output layer of the PSO-LSSVM model for training, to obtain the network-supplied load prediction model.

[0160] It should be noted that the embodiments of the present invention do not limit the "historical time period" and "time point". In some embodiments, it can be set by those skilled in the art according to experimental data or expert experience. For example, the historical time period is 3 years, and the load data is collected every 15 minutes, 96 data points are collected every day.

[0161] It should also be noted that the "PSO-LSSVM model" method involved in the embodiments of the present invention is well-known to those skilled in the art. Therefore, its specific implementation method will not be described in detail.

[0162] Specifically, actually when calculating the network-supplied load, it needs to be calculated according to the following formula:

[0163] P grid =P line +P large +P loss -P plant (8)

[0164] In the formula, P grid is the network-supplied load, P line is the 10kV public and dedicated line load, P large is the load of large power users above 35kV, P loss is the network loss, P plant is the output of the local power plant;

[0165] Since the network loss is mainly composed of line loss and the loss of each electrical equipment, which is related to the specific power flow and is an uncertain quantity, by obtaining the network supply load prediction model and using the predicted values of the 10kV public dedicated line load, the load of large power users above 35kV, and the output of local power plants as the input of the network supply load prediction model, the predicted value of the network supply load can be obtained to ensure the accuracy of the network supply load prediction.

[0166] A short-term load prediction method provided by an embodiment of the present invention obtains the load prediction models of each load component in the network supply load, uses the load prediction models of each load component to obtain the predicted values of each load component, obtains the network supply load prediction model, and uses the predicted values of each load component as the input of the network supply load prediction model to obtain the predicted value of the network supply load, realizing the effective decomposition of each load component in the network supply load, which is beneficial for dispatching personnel to understand the composition and change rules of the network supply load; at the same time, the characteristic analysis and prediction are carried out specifically for each classified total load, improving the prediction refinement degree; since the load prediction process is specific to the network power distribution line, compared with the existing method of directly using network supply prediction, it can combine external influencing factors, distribution network side line structure adjustment, user addition and other factors at the same time to correct the prediction result more accurately, thereby further improving the accuracy of load prediction.

[0167] An embodiment of the present invention also provides a short-term load prediction device, as Figure 5 shown. The device includes:

[0168] The first acquisition module is used to obtain the load prediction models of each load component in the network supply load;

[0169] The second acquisition module is used to obtain the predicted values of each load component by using the load prediction models of each load component;

[0170] The third acquisition module is used to obtain the network supply load prediction model;

[0171] The fourth acquisition module is used to use the predicted values of each load component as the input of the network supply load prediction model to obtain the predicted value of the network supply load.

[0172] Further, each load component includes: the load data of large power users above 35kV, the load data of the output of local power plants, and the load data of 10kV public dedicated lines.

[0173] Further, the first acquisition module includes: a first model establishment unit for establishing a first load prediction model;

[0174] The first model establishment unit is specifically used for:

[0175] When the load component is the load of large power users at 35 kV and above, collect the historical load data and historical load influencing factor data of large power users at 35 kV and above within the historical time period;

[0176] Use the historical load influencing factor data as the training samples for the input layer of the PSO-LSSVM model, and use the historical load data of large power users at 35 kV and above within the historical time period as the training samples for the output layer of the PSO-LSSVM model for training to obtain the first load prediction model.

[0177] Furthermore, the first acquisition module further includes: a second model establishment unit for establishing a second load prediction model;

[0178] The first model establishment unit is specifically used for:

[0179] When the load component is the output load of local power plants, collect the historical load data and historical load influencing factor data of the output of local power plants within the historical time period;

[0180] Use the historical load influencing factor data as the training samples for the input layer of the PSO-LSSVM model, and use the historical load data of the output of local power plants within the historical time period as the training samples for the output layer of the PSO-LSSVM model for training to obtain the second load prediction model.

[0181] Furthermore, the first acquisition module further includes: a third model establishment unit for establishing a third load prediction model;

[0182] The third model establishment unit includes:

[0183] An acquisition subunit for collecting the historical load data and historical load influencing factor data of each line in the 10 kV public dedicated line within the historical time period when the load component is the 10 kV public dedicated line load;

[0184] A first acquisition subunit for using the historical load data of each line in the 10 kV public dedicated line within the historical time period as the basic data and using the basic data to obtain the typical load curves of each line in the 10 kV public dedicated line;

[0185] A first determination subunit for setting the cluster center K = 1;

[0186] A clustering subunit for clustering the typical load curves of each line in the 10 kV public dedicated line using the K-means clustering algorithm to obtain a clustering result;

[0187] A judgment subunit for judging whether the cluster center K is greater than or equal to M. If K is greater than or equal to M, execute the second acquisition subunit; if K is less than M, set K = K + 1 and return to the clustering subunit; where M is a preset value and M is a positive integer;

[0188] A second acquisition subunit, configured to obtain an optimal cluster center by using the elbow method, and use the clustering result corresponding to the optimal cluster center as the final clustering result;

[0189] A second determination subunit, configured to use a correlation method to calculate the correlation results between various influencing factor data in the historical load data and the historical load influencing factor data of the 10 kV public dedicated lines corresponding to each cluster in the optimal clustering result, and sort all the correlation results, and set a certain type of influencing factor corresponding to the top three correlation results as the main influencing factor of the cluster;

[0190] A training subunit, configured to use the historical load influencing factor data corresponding to each cluster in the clustering result as the input layer training samples of the PSO-LSSVM model, and use the historical load data of the 10 kV public dedicated lines corresponding to each cluster in the clustering result as the output layer training samples of the PSO-LSSVM model for training, so as to obtain a third load prediction model for various influencing factors, that is, the third load prediction model for each cluster.

[0191] Further, the acquisition subunit is specifically configured to:

[0192] Combine the historical load data of the lines with transfer supply relationships in the 10 kV public dedicated lines within the historical time period to obtain combined data;

[0193] The historical load data of the lines without transfer supply relationships in the 10 kV public dedicated lines within the historical time period and the combined data constitute the historical load data of each line in the 10 kV public dedicated lines within the historical time period.

[0194] Further, the first acquisition subunit is specifically configured to:

[0195] When the load value of a certain piece of data in the basic data is 0 at a certain moment point, remove this piece of data to obtain the preprocessed basic data; wherein, the historical load data of each line in the 10 kV public dedicated lines every day within the historical time period is one piece of data;

[0196] Use the preprocessed basic data to calculate the average load of each moment point of each line in the 10 kV public dedicated lines, and generate an initial load curve for each line based on the average load of each moment point of each line in the 10 kV public dedicated lines;

[0197] Use the preprocessed basic data to generate a daily load curve for each line in the 10 kV public dedicated lines within the historical time period;

[0198] Calculate the deviation rate between each moment point in the daily load curve of each line in the 10 kV public dedicated lines within the historical time period and each moment point in the initial load curve;

[0199] When the deviation rate between each time point in the daily load curve and each time point in the initial load curve is greater than the deviation rate threshold of each preset time point, the daily load curve is eliminated, and the typical load curves of each line of the 10kV public dedicated line are obtained.

[0200] Further, the second acquisition module is specifically configured to:

[0201] Using the load influencing factor data of the future time to be predicted as the input of the load prediction models of each load component, the predicted values of each load component are obtained.

[0202] Further, the third acquisition module is specifically configured to:

[0203] Collect the historical load data of each load component and the historical load data of the network supply load within the historical time period;

[0204] Using the historical load data of each load component within the historical time period as the training sample of the input layer of the PSO-LSSVM model, and using the historical load data of the network supply load within the historical time period as the training sample of the output layer of the PSO-LSSVM model for training, a network supply load prediction model is obtained.

[0205] It can be understood that the device embodiments provided above correspond to the method embodiments above, and the corresponding specific contents can be referred to each other, and will not be elaborated here.

[0206] The embodiment of the present invention also provides a readable storage medium, on which an executable program is stored, and when the executable program is executed by a processor, the steps of the short-term load prediction method in the above embodiment are implemented.

[0207] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program code.

[0208] The present application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, and the combination of processes and / or blocks in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate for implementing in the processFigure 1 one process or multiple processes and / or blocks Figure 1 a device for the functions specified in one block or multiple blocks.

[0209] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured article including an instruction method, and the instruction method implements the processes Figure 1 one process or multiple processes and / or blocks Figure 1 the functions specified in one block or multiple blocks.

[0210] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide for implementing the processes Figure 1 one process or multiple processes and / or blocks Figure 1 the steps of the functions specified in one block or multiple blocks.

[0211] As described above, this is only a specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A short-term load forecasting method, characterized in that, the method includes: Obtaining the load forecasting models of each load component in the network-supplied load; Using the load forecasting models of each load component to obtain the predicted values of each load component; Obtaining the network-supplied load forecasting model; Taking the predicted values of each load component as the input of the network-supplied load forecasting model to obtain the predicted value of the network-supplied load; The obtaining of the load forecasting models of each load component in the network-supplied load includes: establishing the third load forecasting model; The establishing of the third load forecasting model includes: Step 1: When each load component in the network-supplied load is the 10kV public dedicated line load, collect the historical load data and historical load influencing factor data of each line in the 10kV public dedicated line within the historical time period; Step 2: Let the historical load data of each line in the 10kV public dedicated line within the historical time period be the basic data, and use the basic data to obtain the typical load curves of each line in the 10kV public dedicated line; Step 3: Let the cluster center K = 1; Step 4: Use the K-means clustering algorithm to cluster the typical load curves of each line in the 10kV public dedicated line to obtain the clustering result; Step 5: Determine whether the cluster center K is greater than or equal to M. If K is greater than or equal to M, execute Step 6; if K is less than M, let K = K + 1 and return to Step 4; where M is a preset value and M is a positive integer; Step 6: Use the elbow method to obtain the optimal cluster center, and take the clustering result corresponding to the optimal cluster center as the final clustering result; Step 7: Use the correlation method to calculate the correlation results between various influencing factor data in the historical load data and historical load influencing factor data of the 10kV public dedicated line corresponding to each cluster in the optimal clustering result, and sort all the correlation results. Let the top three ranked correlation results correspond to a certain type of influencing factor, which is the main influencing factor of this cluster; Step 8: Take the historical load influencing factor data corresponding to each cluster in the clustering result as the input layer training samples of the PSO-LSSVM model, and take the historical load data of the 10kV public dedicated line corresponding to each cluster in the clustering result as the output layer training samples of the PSO-LSSVM model for training to obtain the third load forecasting model of various influencing factors, that is, the third load forecasting model of each cluster; The collection of the historical load data of each line in the 10kV public dedicated line in Step 1 includes: Combining the historical load data of the lines with transfer supply relationships in the 10kV public dedicated line within the historical time period to obtain combined data; The historical load data of the lines without transfer supply relationships in the 10kV public dedicated line within the historical time period and the combined data constitute the historical load data of each line in the 10kV public dedicated line within the historical time period; The obtaining of the typical load curves of each line in the 10kV public dedicated line by using the basic data in Step 2 includes: Step 21: When there is a load value of 0 at a certain time point in a certain piece of data in the basic data, remove this piece of data to obtain the preprocessed basic data; Among them, the historical load data of each line in the 10kV public and special lines every day within the historical time period is one piece of data; Step 22: Using the preprocessed basic data, calculate the average load of each line in the 10kV public and special lines at each time point, and generate the initial load curve of each line using the average load of each line in the 10kV public and special lines at each time point; Step 23: Using the preprocessed basic data, generate the daily load curve of each line in the 10kV public and special lines within the historical time period; Step 24: Calculate the deviation rate of each time point in the daily load curve of each line in the 10kV public and special lines within the historical time period from each time point in the initial load curve; When there is a deviation rate of each time point in the daily load curve from each time point in the initial load curve greater than the preset deviation rate threshold for each time point, eliminate this daily load curve to obtain the typical load curve of each line in the 10kV public and special lines.

2. The method according to claim 1, wherein, each load component includes: load data of large power users above 35kV, output load data of local power plants, and load data of 10kV public and special lines.

3. The method according to claim 1, wherein, the load prediction model for obtaining each load component in the network supply load includes: establishing a first load prediction model; The establishing of the first load prediction model includes: when the load component is the load of large power users above 35kV, collecting the historical load data and historical load influencing factor data of large power users above 35kV within the historical time period; Using the historical load influencing factor data as the input layer training samples of the PSO-LSSVM model, and using the historical load data of large power users above 35kV within the historical time period as the output layer training samples of the PSO-LSSVM model for training to obtain the first load prediction model.

4. The method according to claim 1, wherein, the load prediction model for obtaining each load component in the network supply load includes: establishing a second load prediction model; The establishing of the second load prediction model includes: when the load component is the output load of local power plants, collecting the historical load data and historical load influencing factor data of the output of local power plants within the historical time period; Using the historical load influencing factor data as the input layer training samples of the PSO-LSSVM model, and using the historical load data of the output of local power plants within the historical time period as the output layer training samples of the PSO-LSSVM model for training to obtain the second load prediction model.

5. The method according to claim 1, wherein, the obtaining of the predicted values of each load component using the load prediction models of each load component includes: Using the load influencing factor data of the future time to be predicted as the input of the load prediction models of each load component to obtain the predicted values of each load component.

6. The method according to claim 1, wherein, the obtaining of the network supply load prediction model includes: Collect the historical load data of each load component and the historical load data of the network supply load within a historical time period; Use the historical load data of each load component within the historical time period as the input layer training samples of the PSO-LSSVM model, and use the historical load data of the network supply load within the historical time period as the output layer training samples of the PSO-LSSVM model for training to obtain the network supply load prediction model.

7. A short-term load prediction device Characterized in that The device includes: A first acquisition module for acquiring the load prediction models of each load component in the network supply load; The acquisition of the load prediction models of each load component in the network supply load includes: establishing a third load prediction model; The establishment of the third load prediction model includes: Step 1: When each load component in the network supply load is a 10kV public dedicated line load, collect the historical load data and historical load influencing factor data of each line in the 10kV public dedicated line within a historical time period; Step 2: Let the historical load data of each line in the 10kV public dedicated line within the historical time period be the basic data, and use the basic data to obtain the typical load curves of each line in the 10kV public dedicated line; Step 3: Let the cluster center K = 1; Step 4: Use the K-means clustering algorithm to cluster the typical load curves of each line in the 10kV public dedicated line to obtain a clustering result; Step 5: Determine whether the cluster center K is greater than or equal to M. If K is greater than or equal to M, execute Step 6; if K is less than M, let K = K + 1 and return to Step 4; where M is a preset value and M is a positive integer; Step 6: Use the elbow method to obtain the optimal cluster center, and use the clustering result corresponding to the optimal cluster center as the final clustering result; Step 7: Use the correlation method to calculate the correlation results between various influencing factor data in the historical load data and historical load influencing factor data of the 10kV public dedicated line corresponding to each cluster in the optimal clustering result, and sort all the correlation results. Let the top three ranked correlation results correspond to a certain type of influencing factor as the main influencing factor of the cluster; Step 8: Use the historical load influencing factor data corresponding to each cluster in the clustering result as the input layer training samples of the PSO-LSSVM model, and use the historical load data of the 10kV public dedicated line corresponding to each cluster in the clustering result as the output layer training samples of the PSO-LSSVM model for training to obtain the third load prediction models of various influencing factors, that is, the third load prediction models of each cluster; The collection of the historical load data of each line in the 10kV public dedicated line in Step 1 includes: Combining the historical load data of the lines with a transfer supply relationship in the 10kV public dedicated line within the historical time period to obtain combined data; The historical load data of the lines without a transfer supply relationship in the 10kV public dedicated line within the historical time period and the combined data constitute the historical load data of each line in the 10kV public dedicated line within the historical time period; The use of the basic data to obtain the typical load curves of each line in the 10kV public dedicated line in Step 2 includes: Step 21: When the load value of a certain piece of data in the basic data is 0 at a certain moment point, remove this piece of data to obtain the preprocessed basic data; Among them, the historical load data of each line in the 10kV public special line every day within the historical time period is one piece of data; Step 22: Use the preprocessed basic data to calculate the average load of each line in the 10kV public special line at each moment point, and use the average load of each line in the 10kV public special line at each moment point to generate the initial load curve of each line; Step 23: Use the preprocessed basic data to generate the daily load curve of each line in the 10kV public special line within the historical time period; Step 24: Calculate the deviation rate of each moment point in the daily load curve of each line in the 10kV public special line within the historical time period from each moment point in the initial load curve; When there is a deviation rate of each moment point in the daily load curve from each moment point in the initial load curve greater than the preset deviation rate threshold of each moment point, remove this daily load curve to obtain the typical load curve of each line in the 10kV public special line; The second acquisition module is used to obtain the predicted values of the respective load components by using the load prediction models of the respective load components; The third acquisition module is used to obtain the network supply load prediction model; The fourth acquisition module is used to use the predicted values of the respective load components as the input of the network supply load prediction model to obtain the predicted value of the network supply load.

Images