Fine prediction method and system for power system station area load

Abstract

The application discloses a kind of fine prediction methods of power system area load, including obtaining the historical data information of each area in target power system;Preliminary industry label of each area is calculated and obtained, and final industry label of each area is summarized and integrated;Typical load curve of each area is obtained using clustering algorithm, and expansion industry label under final industry label is obtained by clustering;The load characteristic change of each expansion industry label in different seasons and the factor influencing load change are analyzed;User portrait of each area is generated and correlation analysis is carried out;Self-adaptive learning and evaluation of the load prediction model of each area are carried out;The load prediction model of each area is examined, and the load fine prediction of corresponding area is carried out using the load prediction model finally passed examination.The application also discloses a kind of systems for realizing the fine prediction method of power system area load.The application has higher reliability and better accuracy.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of electrical automation, and particularly relates to a fine prediction method and system for the load of a power system area. BACKGROUND

[0002] With the development of economy and technology and the improvement of people's living standards, electric energy has become an essential secondary energy in people's production and life, bringing endless convenience to people's production and life. Therefore, guaranteeing the stable and reliable supply of electric energy has become one of the most important tasks of the power system.

[0003] At present, with the rapid development of new power systems, traditional low-voltage distribution areas are also developing towards being safer and more intelligent. In order to ensure the reliability, safety and intelligence of the power system area, fine load prediction of the power system area is particularly important.

[0004] At present, the load prediction scheme of the power system is mostly from the total level of the power system, and there is no fine load prediction for the power system area. This situation makes the reliability and accuracy of the load prediction result of the power system area poor. SUMMARY

[0005] One of the purposes of the present application is to provide a fine prediction method for the load of a power system area with high reliability and good accuracy.

[0006] The second purpose of the present application is to provide a system for realizing the fine prediction method for the load of the power system area.

[0007] The fine prediction method for the load of the power system area provided by the present application comprises the following steps:

[0008] S1. Obtain historical data information of each area in the target power system;

[0009] S2. Calculate the preliminary industry label of each area according to the historical data information obtained in step S1;

[0010] S3. Synthesize the preliminary industry label calculated in step S2 to obtain the final industry label of each area;

[0011] S4. Obtain the typical load curve of each area by using a clustering algorithm according to the historical data information obtained in step S1, and further cluster the typical load curve to obtain an extended industry label under the final industry label;

[0012] S5. Analyze the load characteristic changes of each extended industry label in different seasons and the factors affecting the load changes;

[0013] S6. According to the analysis result obtained in step S5, a user portrait of each transformer area is generated, and a correlation analysis is performed;

[0014] S7. According to the data information obtained in step S6, adaptive learning and evaluation of a load prediction model of each transformer area are performed;

[0015] S8. The load prediction model of each transformer area obtained in step S7 is examined, and the load fine prediction of the corresponding transformer area is performed by using the load prediction model that finally passes the examination.

[0016] According to the historical data information obtained in step S1, the preliminary industry label of each transformer area is calculated, and the specific steps include the following steps:

[0017] For a public transformer area:

[0018] The following formula is used to calculate the electricity consumption proportion of each industry in the transformer area:

[0019]

[0020] In the formula, P i is the electricity consumption proportion of the i-th industry in the transformer area; E i is the electricity consumption of the i-th industry in the transformer area; ∑ j E j is the sum of the electricity consumption of each industry in the transformer area;

[0021] If the electricity consumption proportion P i of the i-th industry in the transformer area is the largest, the preliminary industry label of the transformer area is the industry label of the i-th industry;

[0022] For a transformation transformer area:

[0023] The preliminary industry label of the transformer area is the industry label of the user;

[0024] The industry label is a three-level industry label.

[0025] The preliminary industry label calculated in step S2 is summarized and synthesized to obtain the final industry label of each transformer area, and the specific steps include the following steps:

[0026] The preliminary industry label calculated in step S2 is judged:

[0027] For the preliminary industry label k:

[0028] If the number of transformer areas corresponding to the preliminary industry label k is greater than a set value, the preliminary industry label k is used as the final industry label of the corresponding transformer area;

[0029] If the number of the transformer area corresponding to the preliminary industry label k is not more than the set value, the secondary industry label corresponding to the preliminary industry label k is taken as the final industry label of the corresponding transformer area.

[0030] The historical data information obtained in step S1 is used to obtain the typical load curve of each transformer area by using a clustering algorithm, and the typical load curve is clustered again to obtain the extended industry label under the final industry label, and the specific steps include the following steps:

[0031] The historical data information obtained in step S1 is used to obtain the typical load curve of each transformer area by using a clustering algorithm;

[0032] The final industry label obtained in step S3 is matched with the typical load curve of each transformer area;

[0033] The typical load curve under each final industry label is clustered again by using a clustering algorithm to obtain the extended industry label under each final industry label.

[0034] The analysis of the load characteristic change of each extended industry label in different seasons and the factors affecting the load change in step S5 includes the following steps:

[0035] The load characteristic change of each extended industry label in different seasons is analyzed:

[0036] The typical transformer area typical load curve in season S a is C a , and the typical transformer area typical load curve in season S b is C b , then the Manhattan distance D ab of C a and C b is n is the number of elements in C a and C b , c ak is the kth element in C a , and c bk is the kth element in C b .

[0037] The load curve in different time periods in season S a is C ai (t), and t is the time period; the load change rate ΔC ai (t) of C ai (t) in each time period is used to represent the load change trend,

[0038] The factors affecting the load change are analyzed:

[0039] The Pearson correlation coefficient r is calculated by the following formula xy Thus, the influence of meteorological factors on different industries is quantified:

[0040]

[0041] where x i is the influence factor data at the collection time; is the sample mean of x i ; l i is the load data; is the sample mean of l i ;

[0042] The meteorological factors include humidity, temperature, rainfall, and wind speed.

[0043] Step S6: Generating the user portrait of each station area according to the analysis result obtained in step S5, and performing correlation analysis, specifically including the following steps:

[0044] Generating the user portrait of each station area in combination with the clustering result obtained in step S4 and the analysis result obtained in step S5;

[0045] Performing correlation analysis: performing meteorological factor correlation analysis for the same period in history; the meteorological factors include humidity, temperature, rainfall, and wind speed.

[0046] Step S7: Performing adaptive learning and evaluation of the load prediction model of each station area according to the data information obtained in step S6, specifically including the following steps:

[0047] Feature importance analysis:

[0048] Performing feature importance analysis on the prediction results of different industry labels, and evaluating the correlation between each feature and the target variable based on mutual information;

[0049] The entropy H(X) of the feature factor X is represented as p(x) is the probability of taking the value x;

[0050] The entropy H(Y) of the station area load prediction Y is represented as p(y) is the probability of taking the value y;

[0051] The joint entropy H(X,Y) of X and Y is represented as p(x,y) is the joint probability distribution of random variables X and Y, i.e., the probability of observing X=x and Y=y simultaneously;

[0052] The mutual information I(X;Y) of X and Y is represented as I(X;Y) = H(X) + H(Y) - H(X,Y);

[0053] The larger the mutual information, the stronger the correlation between X and Y;

[0054] By calculating mutual information, the main influencing factors of load forecasting in each substation under each industry label are evaluated;

[0055] Adaptive Learning and Evaluation of Load Forecasting Models:

[0056] For the substations where the number of samples is less than the set value, the load forecasting models used include the support vector machine model and the kernel support vector machine model;

[0057] Support vector machine model: Given training samples D = {(x1,l1),(x2,l2),...,(x m ,l m )},x m is the mth sample, l m For the label corresponding to the mth sample, construct the model f(x) = wφ(x) + b, where f(x) is the label corresponding to the sample, w is the weight coefficient, φ(x) is the feature space, and b is the bias term;

[0058] Based on the principle of structural risk minimization, the objective function is set as

[0059]

[0060] Where R(x) is the objective function value; ||w|| is the size of the weight coefficient; m is the number of samples; y i is the actual label value of the i-th sample; ε is the allowable deviation;

[0061] Introducing penalty coefficients and slack variables, the objective function is transformed into

[0062]

[0063]

[0064] In the formula is the objective function value; C is the regularization parameter; ξ i is the first slack variable, is the second slack variable; φ(x i ) is the final model function; m is the number of samples;

[0065] Using Lagrange duality theory to solve, we get

[0066]

[0067] In the formula is the first Lagrange coefficient, a i is the second Lagrange coefficient; K(x ix) is a RBF kernel function;

[0068] For the transformer area with volatility greater than a set value, the load prediction model adopted includes a random forest model, a gradient boosting model and a limit random tree model.

[0069] Step S8 is described for the load prediction model of each transformer area obtained in step S7, and the load fine prediction of the corresponding transformer area is carried out by using the load prediction model finally passing the examination.

[0070] The accuracy Acc of the load prediction model is calculated by using the following formula:

[0071]

[0072] In the formula, n is the sample number; Q is the load reference value; x i is the actual load value at the i-th moment; x f,i is the load prediction value at the i-th moment;

[0073] If the accuracy Acc is higher than or equal to a set value, the load prediction model passes the examination;

[0074] If the accuracy Acc is lower than the set value, the load prediction model does not pass the examination, and then the adaptive learning of the load prediction model is re-performed until the load prediction model passes the examination.

[0075] The load fine prediction of the corresponding transformer area is carried out by using the load prediction model finally passing the examination.

[0076] The application further provides a system for realizing the fine prediction method of the power system area load, comprising a data acquisition module, a label calculation module, a label correction module, a data clustering module, a load analysis module, a user analysis module, a model learning module and a load prediction module; the data acquisition module, the label calculation module, the label correction module, the data clustering module, the load analysis module, the user analysis module, the model learning module and the load prediction module are sequentially connected; the data acquisition module is used for acquiring historical data information of each area in the target power system and uploading the data information to the label calculation module; the label calculation module is used for calculating preliminary industry labels of each area according to the received data information and the acquired historical data information, and uploading the data information to the label correction module; the label correction module is used for collecting and synthesizing the calculated preliminary industry labels according to the received data information, obtaining final industry labels of each area, and uploading the data information to the data clustering module; the data clustering module is used for acquiring typical load curves of each area by using a clustering algorithm according to the received data information and the acquired historical data information, and performing clustering on the typical load curves to obtain extended industry labels under the final industry labels, and uploading the data information to the load analysis module; the load analysis module is used for analyzing load characteristic changes of each extended industry label in different seasons and factors influencing the load changes according to the received data information, and uploading the data information to the user analysis module; the user analysis module is used for generating user portraits of each area according to the received data information and the obtained analysis results, performing correlation analysis, and uploading the data information to the model learning module; the model learning module is used for performing adaptive learning and evaluation of a load prediction model of each area according to the received data information and the obtained data information, and uploading the data information to the load prediction module; and the load prediction module is used for examining the obtained load prediction model of each area, and performing fine prediction of the load of the corresponding area by using the load prediction model that finally passes the examination.

[0077] The fine prediction method and system of the power system area load provided by the application perform adaptive learning and examination on the area load prediction model based on the labels and portraits of the industries where the areas are located; therefore, the application can not only realize fine prediction of the power system area load, but also has higher reliability and better accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0078] Figure 1 The figure is a method flowchart of the method of the application.

[0079] Figure 2 The figure is a functional module schematic diagram of the system of the application. DETAILED DESCRIPTION

[0080] AsFigure 1 The method flowchart of the method of the application is shown: the fine prediction method of the power system area load disclosed by the application comprises the following steps:

[0081] S1. Obtain historical data information of each area in the target power system;

[0082] S2. Calculate the preliminary industry label of each area according to the historical data information obtained in step S1; specifically comprising the following steps:

[0083] For transformer area:

[0084] The following formula is used to calculate the electricity consumption proportion of each industry in the area:

[0085]

[0086] In the formula, P i is the electricity consumption proportion of the i-th industry in the area; E i is the electricity consumption of the i-th industry in the area; ∑ j E j is the sum of the electricity consumption of each industry in the area;

[0087] If the electricity consumption proportion P i of the i-th industry in the area is the maximum, the preliminary industry label of the area is the industry label of the i-th industry;

[0088] For transition area:

[0089] The preliminary industry label of the area is the industry label of the user;

[0090] The industry label is a three-level industry label;

[0091] S3. The preliminary industry label calculated in step S2 is summarized and integrated to obtain the final industry label of each area; specifically comprising the following steps:

[0092] The preliminary industry label calculated in step S2 is judged:

[0093] For preliminary industry label k:

[0094] If the number of areas corresponding to the preliminary industry label k is greater than a set value, the preliminary industry label k is taken as the final industry label of the corresponding area;

[0095] If the number of areas corresponding to the preliminary industry label k is not greater than a set value, the secondary industry label corresponding to the preliminary industry label k is taken as the final industry label of the corresponding area;

[0096] S4. According to the historical data information obtained in step S1, a clustering algorithm is used to obtain the typical load curve of each transformer area, and the typical load curve is clustered again to obtain the extended industry label under the final industry label; specifically including the following steps:

[0097] According to the historical data information obtained in step S1, a clustering algorithm is used to cluster the historical load data of the transformer area, and the typical load curve of each transformer area is obtained;

[0098] The final industry label obtained in step S3 is matched with the typical load curve of each transformer area;

[0099] The typical load curve under each final industry label is clustered again using a clustering algorithm to obtain the extended industry label under each final industry label;

[0100] S5. Analyze the load characteristic changes of each extended industry label in different seasons, and the factors affecting the load changes; specifically including the following steps:

[0101] Analyze the load characteristic changes of each extended industry label in different seasons:

[0102] Set the typical load curve of the typical transformer area in season S a as C a , and the typical load curve of the typical transformer area in season S b as C b , then calculate the Manhattan distance D ab between C a and C b as n is the number of elements in C a and C b , c ak is the kth element in C a , and c bk is the kth element in C b ;

[0103] Set the load curve of different time periods in season S a as C ai (t), and t is the time period; use the load change rate ΔC ai (t) of C ai (t) in each time period to represent the load change trend,

[0104] Analyze the factors affecting the load changes:

[0105] The Pearson correlation coefficient r xy is calculated using the following formula, thereby quantifying the influence of different industries on meteorological factors:

[0106]

[0107] wherein x i is the influence factor data at the collection time; is the sample average value of x i ; l i is the load data; is the sample average value of l i ;

[0108] The meteorological factors include humidity, temperature, rainfall and wind speed;

[0109] S6. Generating the user portrait of each station area according to the analysis result obtained in step S5, and performing correlation analysis; specifically including the following steps:

[0110] Generating the user portrait of each station area in combination with the clustering result obtained in step S4 and the analysis result obtained in step S5;

[0111] Performing correlation analysis: performing meteorological factor correlation analysis for the same period in history; the meteorological factors include humidity, temperature, rainfall and wind speed;

[0112] S7. Performing adaptive learning and evaluation of the load prediction model of each station area according to the data information obtained in step S6; specifically including the following steps:

[0113] Feature importance analysis:

[0114] Performing feature importance analysis on the prediction results of different industry labels, and evaluating the correlation between each feature and the target variable based on mutual information;

[0115] The entropy H(X) of the feature factor X is represented as p(x) is the probability of taking the value x;

[0116] The entropy H(Y) of the station area load prediction Y is represented as p(y) is the probability of taking the value y;

[0117] The joint entropy H(X,Y) of X and Y is represented as p(x,y) is the joint probability distribution of random variables X and Y, that is, the probability of observing X=x and Y=y at the same time;

[0118] The mutual information I(X;Y) of X and Y is represented as I(X;Y)=H(X)+H(Y)-H(X,Y);

[0119] The greater the mutual information, the stronger the correlation between X and Y; on the contrary, the weaker the correlation; this index can be used to perform correlation analysis on multi-dimensional data, find out the main influence factors of the station area load prediction under each industry label, and provide strong support for subsequent modeling and prediction;

[0120] By calculating mutual information, the main influencing factors of load forecasting in each substation under each industry label are evaluated;

[0121] Adaptive Learning and Evaluation of Load Forecasting Models:

[0122] For the substations where the number of samples is less than the set value, the load forecasting models used include the support vector machine model and the kernel support vector machine model;

[0123] Support vector machine model: Given training samples D = {(x1,l1),(x2,l2),...,(x m ,l m )},x m is the mth sample, l m For the label corresponding to the mth sample, construct the model f(x) = wφ(x) + b, where f(x) is the label corresponding to the sample, w is the weight coefficient, φ(x) is the feature space, and b is the bias term;

[0124] Based on the principle of structural risk minimization, the objective function is set as

[0125]

[0126] Where R(x) is the objective function value; ||w|| is the size of the weight coefficient; m is the number of samples; y i is the actual label value of the i-th sample; ε is the allowable deviation;

[0127] Introducing penalty coefficients and slack variables, the objective function is transformed into

[0128]

[0129]

[0130] In the formula is the objective function value; C is the regularization parameter; ξ i is the first slack variable, is the second slack variable; φ(x i ) is the final model function; m is the number of samples;

[0131] Using Lagrange duality theory to solve, we get

[0132]

[0133] In the formula is the first Lagrange coefficient, a i is the second Lagrange coefficient; K(x i ,x) is the RBF kernel function;

[0134] For the area with volatility greater than the set value, the load prediction model includes random forest model, gradient boosting model and extreme random tree model;

[0135] The decision tree is a supervised learning algorithm suitable for classification and regression problems, which can extract decision rules according to a series of load-related factors and target values; The input of the decision tree model is the data of the training set, which continuously calculates the size of different values under different influencing factors, and finds the optimal split point with the smallest squared error by the least square method, that is, to solve:

[0136]

[0137] In the formula, c1 is the average value of the target variable in the set R1(j,s); c2 is the average value of the target variable in the set R2(j,s); y i is the target variable value of the ith sample; x i is the ith sample; R1(j,s) is the sample set satisfying x (j) ≤s, and R1(j,s)={x|x (j) ≤s}, x (j) is the jth feature variable of x; R2(j,s) is the sample set satisfying x (j) ≥s, and R2(j,s)={x|x (j) ≥s}; j is the jth feature variable in the data set; s is the split point;

[0138] The calculated values in different regions are obtained, and the above steps are repeated for the sub-regions until the minimum squared error is reached, and finally the regions are divided into several regions to form a regression decision tree; The random forest model is based on the decision tree, which selects the load corresponding to the data set and the load value at the current time through random sampling with replacement; Different decision trees correspond to different data sets, and the final prediction is determined according to the prediction result of each decision tree to reduce the possibility of large error in the prediction of a single decision tree; Random forest is a way to make decisions after getting all the load prediction output values of single decision tree, so it is better than decision tree in accuracy;

[0139] S8. The load prediction model of each area obtained in step S7 is evaluated, and the load prediction model that finally passes the evaluation is used for fine prediction of the load of the corresponding area; Specifically, the following steps are included:

[0140] The accuracy Acc of the load prediction model is calculated by the following formula:

[0141]

[0142] In the formula, n is the number of samples; Q is the load reference value; xi is the actual value of the load at the i-th moment; x f,i is the predicted value of the load at the i-th moment;

[0143] If the accuracy Acc is higher than or equal to the set value, the load prediction model passes the examination;

[0144] If the accuracy Acc is lower than the set value, the load prediction model fails the examination, and the adaptive learning of the load prediction model is restarted from step S1 until the load prediction model passes the examination; at this time, the low accuracy may be caused by factors such as unsuitable algorithm, incorrect industry label, and missing load data; algorithm optimization is carried out for the case of unsuitable algorithm; incorrect label is mainly caused by incorrect portrait result and incorrect initial label, and user portrait algorithm optimization and initial industry label verification are carried out; load data supplement is carried out for the case of missing load data;

[0145] The load fine prediction of the corresponding area is carried out by using the finally passed load prediction model.

[0146] For example, Figure 2As shown is a functional module schematic diagram of the system of the application: the system for implementing the fine prediction method of the power system area load disclosed in the application comprises a data acquisition module, a label calculation module, a label correction module, a data clustering module, a load analysis module, a user analysis module, a model learning module and a load prediction module; the data acquisition module, the label calculation module, the label correction module, the data clustering module, the load analysis module, the user analysis module, the model learning module and the load prediction module are sequentially connected; the data acquisition module is used for acquiring historical data information of each area in the target power system, and uploading the data information to the label calculation module; the label calculation module is used for calculating preliminary industry labels of each area according to the received data information and the acquired historical data information, and uploading the data information to the label correction module; the label correction module is used for collecting and synthesizing the calculated preliminary industry labels according to the received data information, obtaining final industry labels of each area, and uploading the data information to the data clustering module; the data clustering module is used for obtaining typical load curves of each area by using a clustering algorithm according to the received data information and the acquired historical data information, and performing clustering on the typical load curves again to obtain extended industry labels under the final industry labels, and uploading the data information to the load analysis module; the load analysis module is used for analyzing load characteristic changes of each extended industry label in different seasons and factors influencing the load changes according to the received data information, and uploading the data information to the user analysis module; the user analysis module is used for generating user portraits of each area and performing correlation analysis according to the received data information and the obtained analysis results, and uploading the data information to the model learning module; the model learning module is used for performing adaptive learning and evaluation of a load prediction model of each area according to the received data information and the obtained data information, and uploading the data information to the load prediction module; and the load prediction module is used for examining the obtained load prediction model of each area according to the received data information, and performing fine load prediction of the corresponding area by using the load prediction model that finally passes the examination.

Claims

1. A method for fine-grained prediction of load in a power system substation, comprising the following steps: S1. Obtain historical data information for each substation in the target power system; S2. Based on the historical data information obtained in step S1, the preliminary industry labels of each area are calculated; specifically, the steps include: For public substation areas: The following formula is used to calculate the proportion of electricity consumption of each industry in the substation area: Where P i is the proportion of electricity consumption of the ith industry in the area; E i is the electricity consumption of the i-th industry in the substation; ∑ i E i It is the sum of electricity consumption of various industries in the substation area; If the electricity consumption of industry i in the substation accounts for P i The largest, then the preliminary industry label of the area is the industry label of the i-th industry; For dedicated transformer areas: The initial industry label of the station area is the industry label of the user; The industry label mentioned is a third-level industry label; S3. Summarize the preliminary industry labels calculated in step S2 to obtain the final industry labels for each area; S4. Based on the historical data information obtained in step S1, a clustering algorithm is used to obtain the typical load curve of each area, and the typical load curve is clustered again to obtain the expanded industry label under the final industry label; S5. Analyze the load characteristics of each expanded industry label in different seasons and the factors that affect load changes; S6. Based on the analysis results obtained in step S5, generate user portraits for each area and perform correlation analysis; S7. Based on the data information obtained in step S6, adaptive learning and evaluation of the load forecasting model for each area; S8. Assess the load forecasting model of each substation obtained in step S7, and use the load forecasting model that finally passes the assessment to perform refined load forecasting for the corresponding substation.

2. The method for fine-grained prediction of power system load according to claim 1 is characterized in that Step S3 summarizes and integrates the preliminary industry labels calculated in step S2 to obtain the final industry labels for each substation, which specifically includes the following steps: Determine the preliminary industry label calculated in step S2: For the preliminary industry label k: If the number of zones corresponding to the preliminary industry label k is greater than the set value, the preliminary industry label k will be used as the final industry label of the corresponding zone; If the number of zones corresponding to the preliminary industry label k is not greater than the set value, the secondary industry label corresponding to the preliminary industry label k is used as the final industry label of the corresponding zone.

3. The method for fine-grained prediction of power system load according to claim 2 is characterized in that Step S4, based on the historical data information obtained in step S1, uses a clustering algorithm to obtain the typical load curve of each substation, and clusters the typical load curve again to obtain the extended industry label under the final industry label, specifically including the following steps: Based on the historical data information obtained in step S1, clustering the historical load data of the substation is performed using a clustering algorithm to obtain a typical load curve for each substation; Match the final industry label obtained in step S3 with the typical load curve of each substation; The typical load curves under each final industry label are clustered again using a clustering algorithm to obtain the extended industry labels under each final industry label.

4. The method for fine-grained prediction of power system load according to claim 3 is characterized in that The analysis of the load characteristic changes of each expanded industry tag in different seasons and the factors affecting the load changes in step S5 specifically includes the following steps: Analyze the load characteristics of each expansion industry label in different seasons: Set Season S a The typical load curve of a typical area is C a , Season S b The typical load curve of a typical area is C b , then calculate C a and C b Manhattan distance D ab for n is C a and C b The number of elements in c ak C a The kth element in c bk C b The kth element in ; Set Season S a The load curves at different time periods are C ai (t), t is the time period; using C ai (t) Load change rate ΔC in each period ai (t) represents the load change trend, Analyze the factors affecting load changes: The Pearson correlation coefficient r is calculated using the following formula xy , thereby quantifying the impact of meteorological factors on different industries: Where x i To collect the influencing factor data at the moment; is x i The sample mean of i is the load data; l i The sample mean of The meteorological factors include humidity, temperature, rainfall and wind speed.

5. The method for fine-grained prediction of power system load according to claim 4 is characterized in that Step S6 generates user profiles for each substation based on the analysis results obtained in step S5 and performs correlation analysis, specifically including the following steps: Combine the clustering results obtained in step S4 and the analysis results obtained in step S5 to generate user profiles for each substation; Conduct correlation analysis: Conduct correlation analysis on meteorological factors for the same historical period; the meteorological factors include humidity, temperature, rainfall and wind speed.

6. The method for fine-grained prediction of power system load according to claim 5 is characterized in that Step S8 is to assess the load forecasting model of each substation obtained in step S7, and use the load forecasting model that finally passes the assessment to perform refined load forecasting for the corresponding substation, which specifically includes the following steps: The following formula is used to calculate the accuracy Acc of the load forecasting model: Where n is the number of samples; Q is the load reference value; x i is the actual load value at the i-th moment; x f,i is the load forecast value at the i-th moment; If the accuracy Acc is higher than or equal to the set value, the load forecasting model passes the assessment; If the accuracy Acc is lower than the set value, the load forecasting model fails the assessment, and the process returns to step S1 to re-perform the adaptive learning of the load forecasting model until the load forecasting model passes the assessment; The load forecasting model that finally passes the assessment is used to make detailed load forecasts for the corresponding substations.

7. A system for implementing the refined prediction method for power system substation load according to any one of claims 1 to 6, characterized in that It includes a data acquisition module, a label calculation module, a label correction module, a data clustering module, a load analysis module, a user analysis module, a model learning module and a load forecasting module; the data acquisition module, the label calculation module, the label correction module, the data clustering module, the load analysis module, the user analysis module, the model learning module and the load forecasting module are connected in series in sequence; the data acquisition module is used to obtain the historical data information of each substation in the target power system and upload the data information to the label calculation module; the label calculation module is used to calculate the preliminary industry label of each substation based on the received data information and the acquired historical data information, and upload the data information to the label correction module; The label correction module is used to summarize and integrate the preliminary industry labels calculated based on the received data information, obtain the final industry labels of each substation, and upload the data information to the data clustering module; The data clustering module is used to obtain the typical load curve of each substation based on the received data information and the acquired historical data information using a clustering algorithm, and cluster the typical load curve again to obtain the extended industry label under the final industry label, and upload the data information to the load analysis module; The load analysis module is used to analyze the load characteristic changes of each expanded industry tag in different seasons and the factors affecting the load changes based on the received data information, and upload the data information to the user analysis module; The user analysis module is used to generate user profiles for each substation based on the received data information and the obtained analysis results, and perform correlation analysis, and upload the data information to the model learning module; The model learning module is used to perform adaptive learning and evaluation of the load forecasting model of each substation based on the received data information and upload the data information to the load forecasting module; The load forecasting module is used to assess the load forecasting models of each substation based on the received data information, and use the load forecasting model that finally passes the assessment to perform refined load forecasting for the corresponding substation.

Images