A power load prediction analysis system and method based on big data

Abstract

The application discloses a power load prediction analysis system and method based on big data, relates to the technical field of power load analysis, and comprises a data statistics module, a period marking module, a power load data classification module, a response filling demand model construction module, a data type division module, a scene feature output module, a filling data set generation module and a real-time response filling module. Through classification of missing data according to missing reasons and introduction of three quantitative indexes of output data completion ratio, output data stability ratio and prediction precision value, the influence of different types of missing data on prediction results can be accurately evaluated, the evaluation blind area of traditional technology is avoided, and a scientific basis is provided for subsequent filling decision.

Description

Technical Field

[0001] This invention relates to the field of power load analysis technology, specifically a power load forecasting and analysis system and method based on big data. Background Technology

[0002] In the field of power load forecasting and analysis, traditional technologies rely on complete historical power load data to support the operation of forecasting models (such as ARIMA models and artificial neural networks). However, these technologies have significant shortcomings in practical applications. The power load data acquisition process is susceptible to interference from multiple factors, leading to data gaps. These missing data segments directly affect data integrity. Existing technologies do not finely differentiate between the types and scenarios of missing data, and only use a single method to fill in the data. This fails to adapt to the filling needs under different causes, durations, and time periods of missing data, easily resulting in large deviations between the filled data and the actual load. Furthermore, the lack of a quantitative assessment mechanism for the impact of missing data makes it impossible to determine whether missing data will reduce forecast accuracy. This can lead to blind operations when filling is unnecessary and delayed responses when filling is needed, ultimately resulting in low power load forecast accuracy and poor reliability, making it difficult to meet the actual needs of power dispatching and supply-demand balance. Summary of the Invention

[0003] The purpose of this invention is to provide a power load forecasting and analysis system and method based on big data to solve the problems raised in the prior art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a power load forecasting and analysis method based on big data, the method comprising:

[0005] Step S100: Statistically analyze the historical power load data and predicted events in each monitoring area. Predicted events refer to events that execute power prediction objectives based on power load data within a time period. Divide the historical data into several event periods according to the recording period of the power load data required for the predicted events; identify missing data segments in the event periods, and mark event periods with missing data segments as observation periods and event periods without missing data segments as safety periods; extract the missing power load data for the missing data segments in all observation periods, and perform data identification and classification based on the cause of the missing data.

[0006] Step S200: Obtain the missing duration of the missing data segment record for each type of missing power load data, construct a response filling requirement model, and output the safe power load data type and the missing power data type to be analyzed;

[0007] Step S300: Extract the set of missing time periods and the observation period of the missing power data history to be analyzed, summarize the missing duration and time period type within the observation period, as well as the specific reasons for the missing data, and determine the scenario characteristics of power load monitoring; retrieve the predicted events within the safe period when the scenario characteristics are the same, and analyze and output the filling data set of each type of missing power data to be analyzed.

[0008] Step S400: When the response identifies a missing segment of real-time data, determine the type of power load data, and fill the data set response based on the determination result.

[0009] Furthermore, step S100 includes the following specific steps:

[0010] Extract the minimum node interval duration and the average node interval duration of adjacent power load data acquisition time nodes in each event cycle. When there is a record node interval duration and the minimum node interval duration with a difference greater than the first difference threshold and greater than the average node interval duration, the interval corresponding to the node interval duration is output as the data missing segment.

[0011] Obtain the reasons for missing power load data records within the missing data segment. The reasons for missing data include equipment failure, environmental interference, and human operation. Mark the missing power load data with the same missing reason in the same monitoring area as a group of data, and use the predicted event as the marker index to distinguish the missing power load data recorded in different events under the same group of data.

[0012] The purpose of classification is to determine the data segmentation under different causes of data missing, thereby laying the data analysis foundation for intelligent data completion.

[0013] Furthermore, step S200 includes the following specific steps:

[0014] Step S210: Obtain the average number of records q of missing power load data of each type in the monitoring area within a unit time period, and the total unit time T of the missing power load data of that type in each observation period. Extract the actual storage amount Q of the load data in the corresponding observation period. Calculate the output data completion ratio D of the corresponding type of missing power load data under a prediction event, D=Q / qT. The output data completion ratio reflects the impact of the missing data segment on the overall data acquisition amount. The larger the output data completion ratio, the fewer the missing power load data.

[0015] Step S220: Based on the time series, the unit duration of each observation period for missing power load data of the same type is used as the data acquisition node. The power load data acquisition map for each predicted event is plotted with the data acquisition node as the x-axis and the power load data collected within the corresponding unit duration as the y-axis. Nodes where the difference in power load data between adjacent acquisition nodes is greater than or equal to the difference threshold are used as segmentation nodes. All data acquisition nodes are traversed to complete time period segmentation, and the time periods contained in adjacent segmentation nodes are used as feature time periods. The peak-valley difference Q recorded in each feature time period is extracted. 差 The peak-valley difference represents the difference between the maximum and minimum values ​​of power load data within a characteristic period. The sum of peak-valley differences Q1 for each corresponding safe period in the same historical monitoring area is extracted, and the average value Q is calculated for the sum of peak-valley differences across all safe periods. 1均 Using the formula: F=(∑Q 差 ) / Q 1均 Calculate the output data stability ratio F, ∑Q, for a predicted event of missing power load data of the corresponding type. 差 This represents the sum of peak-to-valley differences for each characteristic time period in the power load data acquisition graph corresponding to a predicted event. The output data stability ratio reflects the stage that the power load data recorded during the missing time period in the monitored power load data may be in. The peak-to-valley differences of power load data at different nodes in the entire observation period may have significant deviations, thus providing a certain assessment of the importance of the period in which the missing time period is located. The larger the output data stability ratio, the smaller the deviation and the smaller the impact on the stability of the power load data corresponding to the missing time period.

[0016] Step S230: Extract the prediction result value U1 and actual load value U2 of the prediction model for the same type of missing power load data in each prediction event. Use the formula: Y=1-[|U2-U1| / U2] to calculate the prediction accuracy value Y for a prediction event corresponding to the same type of missing power load data. The larger the prediction accuracy value, the smaller the deviation between the predicted value and the actual value, and the smaller the impact of the missing data in the corresponding prediction event on the prediction result.

[0017] Step S240: Using the output data completion ratio and output data stability ratio of missing power load data records of the same type as input variables, and the prediction accuracy value Y under the same prediction event as output variables, generate data group A, A=[(D, F),Y]. Summarize the data group A corresponding to all prediction events of missing power load data records of the same type, and construct the response filling demand function y, y=k1*d D +k2*f F +ε; Input the corresponding data group A to generate the response filling requirement function y for the record reference coefficients k1, k2 and error term ε;

[0018] Step S250: Set the prediction accuracy value threshold Y0, and construct the response filling demand model as follows: Input the completion ratio of the output data obtained in real time and the stability ratio of the output data into the response filling demand function y. If the theoretical prediction accuracy value Y 测 ≥Y0, no response is triggered. If the theoretical prediction accuracy value Y 测 <Y0, a filling demand response is triggered;

[0019] Step S260: Traverse and find the prediction accuracy values of all prediction events of each type of missing power load data record. If the prediction accuracy values of all prediction events are greater than the threshold Y0, mark the corresponding type of missing power load data as the safe power load data type, and extract the maximum missing duration as the critical missing monitoring value; otherwise, it is the missing power data type to be analyzed.

[0020] Further, step S300 includes the following:

[0021] Step S310: The time period type refers to the time period difference divided by the peak-valley difference within the characteristic time period. Calculate the average value Q 平 and the standard deviation Q 标差 of the peak-valley differences of all characteristic time periods recorded in each prediction event of the same type of missing power data to be analyzed. Take [Q 平 -Q 标差 ,Q 平 +Q 标差 as the interval length to divide all characteristic time periods of the complete inspection period, generate the corresponding time period types; and extract the power load data acquisition graph corresponding to the time from the start moment of the inspection period to before the missing as the characteristic acquisition graph;

[0022] Step S320: Use the characteristic acquisition graph, the reason for the missing, the missing duration, and the characteristic time period corresponding to the missing as the scenario monitoring characteristics of the corresponding type of missing power data to be analyzed;

[0023] Step S330: The prediction events within the safe period when the scenario characteristics are the same refer to the prediction events within the safe period where the similarity of the characteristic acquisition graph is greater than the similarity threshold, the reason for the missing is the same, the absolute value of the difference in the missing duration is less than the duration difference threshold, and the characteristic time period corresponding to the missing is the same; and mark them as control events. Extract the power load data of the control events corresponding to the overlapping part with the missing time period as the filling data for each prediction event corresponding to the missing time period; and generate a filling data set of the associated filling data corresponding to different scenario monitoring characteristics for each type of missing power data to be analyzed.

[0024] The construction of the data set can accurately fill in power load data under different missing scenarios for different types of missing power data, thereby greatly reducing the problem of reduced accuracy in analyzing and predicting power load under missing data, and improving the efficiency of diversified analysis and intelligent response of data filling scenarios.

[0025] Furthermore, step S400 includes the following:

[0026] Obtain the missing reason corresponding to the missing segment of real-time data. If the missing reason is the same as the missing reason corresponding to the safe power load data, obtain the missing duration of the missing segment record of real-time data. If the missing duration is less than or equal to the critical missing monitoring value of the corresponding safe power load data, do not respond to data filling. If the missing duration is greater than the critical missing monitoring value of the corresponding safe power load data, return to steps S200 to S300, construct the response filling requirement model of the type of power load data and output the filling data set.

[0027] When the cause of the missing data is the same as the cause of the missing power data to be analyzed, the real-time scene features of the missing segment record of the real-time data are obtained, the filling data set of the corresponding missing power data record to be analyzed is found, and the scene monitoring features that are the same as the real-time scene features and the corresponding filling data are matched to respond.

[0028] A power load forecasting and analysis system based on big data includes a data statistics module, a periodic marking module, a power load data classification module, a response filling demand model construction module, a data type division module, a scenario feature output module, a filling data set generation module, and a real-time response filling module.

[0029] The data statistics module is used to collect historical power load data and predict events for each monitoring area;

[0030] The period marking module is used to divide historical data into several event periods based on the recording period of the power load data required for predicting events; and to identify missing data segments in the event periods, marking event periods with missing data segments as examination periods and event periods without missing data segments as safety periods;

[0031] The power load data classification module is used to identify and classify data based on the cause of missing data.

[0032] The response filling requirement model building module is used to obtain the missing duration of the missing data segment records for each type of missing power load data and build the response filling requirement model.

[0033] The data type partitioning module is used to output the safe power load data type and the missing power data type to be analyzed;

[0034] The scenario feature output module is used to summarize the missing duration and time period type within the observation period, as well as the specific reasons for the missing, to determine the scenario features of power load monitoring;

[0035] The data set generation module is used to analyze and output data sets for missing power data of various types to be analyzed.

[0036] The real-time response filling module is used to identify missing segments in real-time data, determine the data type of the power load, and match and fill the missing data set based on the determination result.

[0037] Furthermore, the response-filling demand model building module includes an output data completion ratio calculation unit, an output data stability ratio calculation unit, a prediction accuracy value calculation unit, a response-filling demand function analysis unit, and a model building unit;

[0038] The output data completion ratio calculation unit is used to calculate the output data completion ratio of the corresponding type of missing power load data under a prediction event.

[0039] The output data stability ratio calculation unit is used to calculate the output data stability ratio of the corresponding type of missing power load data under a prediction event.

[0040] The prediction accuracy calculation unit is used to extract the prediction results and actual load values ​​of the prediction model for the same type of missing power load data in each prediction event, and calculate the prediction accuracy value.

[0041] The response filling demand function analysis unit is used to construct the response filling demand function by taking the output data completion ratio and output data stability ratio of the same type of missing power load data records as input variables and the prediction accuracy value under the same prediction event as output variables.

[0042] The model building unit is used to set the prediction accuracy threshold and build a response filling requirement model.

[0043] Furthermore, the data type segmentation module includes a safe power load data type marking unit and a missing power data type marking unit to be analyzed;

[0044] The safe power load data type marking unit is used to traverse and search for the prediction accuracy value of all prediction events of each type of missing power load data record. If the prediction accuracy value of all prediction events is greater than the threshold, the corresponding type of missing power load data is marked as a safe power load data type.

[0045] The missing power data type marker unit is used to mark a missing power data type as to be analyzed when the judgment conditions of the safe power load data type marker unit are not met.

[0046] Compared with the prior art, the beneficial effects of the present invention are:

[0047] By classifying missing data according to the cause of the missing data and introducing three quantitative indicators—output data completion ratio, output data stability ratio, and prediction accuracy value—the impact of different types of missing data on the prediction results can be accurately assessed, avoiding the blind spot of the "one-size-fits-all" assessment in traditional techniques and providing a scientific basis for subsequent filling decisions.

[0048] This application constructs a response filling demand function and model based on quantitative indicators, and only triggers filling when the theoretical prediction accuracy value is lower than the threshold. At the same time, all missing data with the accuracy of all predicted events are marked as "safe power load data type", which does not require additional filling, reduces invalid data processing process, and improves system operating efficiency.

[0049] The system constructs scene monitoring features based on "feature acquisition map, cause of missing data, duration of missing data, and time period of missing features". It then matches historical data from the same scenario's safety cycle to generate a set of missing data, ensuring that the missing data is highly consistent with the real-time missing scenario. This solves the problem of poor adaptability of traditional single-filling methods, significantly improves the accuracy of missing data filling, and thus ensures the accuracy of power load forecasting results. Attached Figure Description

[0050] Figure 1 This is a schematic diagram of the structure of a power load forecasting and analysis system based on big data according to the present invention. Detailed Implementation

[0051] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] Example: Figure 1 As shown, this invention provides a power load forecasting and analysis method based on big data, the method comprising:

[0053] Step S100: Statistically analyze the historical power load data and predicted events in each monitoring area. Predicted events refer to events that execute power prediction objectives based on power load data within a time period. Divide the historical data into several event periods according to the recording period of the power load data required for the predicted events; identify missing data segments in the event periods, and mark event periods with missing data segments as observation periods and event periods without missing data segments as safety periods; extract the missing power load data for the missing data segments in all observation periods, and perform data identification and classification based on the cause of the missing data.

[0054] Step S200: Obtain the missing duration of the missing data segment record for each type of missing power load data, construct a response filling requirement model, and output the safe power load data type and the missing power data type to be analyzed;

[0055] Step S300: Extract the set of missing time periods and the observation period of the missing power data history to be analyzed, summarize the missing duration and time period type within the observation period, as well as the specific reasons for the missing data, and determine the scenario characteristics of power load monitoring; retrieve the predicted events within the safe period when the scenario characteristics are the same, and analyze and output the filling data set of each type of missing power data to be analyzed.

[0056] Step S400: When the response identifies a missing segment of real-time data, determine the type of power load data, and fill the data set response based on the determination result.

[0057] Step S100 includes the following specific steps:

[0058] Extract the minimum node interval duration and the average node interval duration of adjacent power load data acquisition time nodes in each event cycle. When there is a record node interval duration and the minimum node interval duration with a difference greater than the first difference threshold and greater than the average node interval duration, the interval corresponding to the node interval duration is output as the data missing segment.

[0059] Obtain the reasons for missing power load data records within the missing data segment. The reasons for missing data include equipment failure, environmental interference, and human operation. Mark the missing power load data with the same missing reason in the same monitoring area as a group of data, and use the predicted event as the marker index to distinguish the missing power load data recorded in different events under the same group of data.

[0060] As illustrated in the embodiments: equipment failures include sensor and meter malfunctions; environmental interference includes the effects of extreme weather, such as lightning strikes causing short circuits; human error includes malicious damage to the acquisition unit switches and improper data permission settings, resulting in data being marked as confidential and unable to access the public database due to incorrect permission configuration. In this application, power load data is a general term for the power data to be collected, and the physical quantities contained in different acquisition nodes are the same. The purpose of classification is to determine the data division under different causes of data loss, thereby providing a data analysis basis for intelligent data completion.

[0061] Step S200 includes the following specific steps:

[0062] Step S210: Obtain the average number of records q of missing power load data of each type in the monitoring area within a unit time period, and the total unit time T of the missing power load data of that type in each observation period. Extract the actual storage amount Q of the load data in the corresponding observation period. Calculate the output data completion ratio D of the corresponding type of missing power load data under a prediction event, D=Q / qT. The output data completion ratio reflects the impact of the missing data segment on the overall data acquisition amount. The larger the output data completion ratio, the fewer the missing power load data.

[0063] The same type of missing power load data may contain data from multiple forecast events, so the output data completion ratio is calculated separately for each forecast event of the same type of missing power load data.

[0064] Step S220: Based on the time series, the unit duration of each observation period for missing power load data of the same type is used as the data acquisition node. The power load data acquisition map for each predicted event is plotted with the data acquisition node as the x-axis and the power load data collected within the corresponding unit duration as the y-axis. Nodes where the difference in power load data between adjacent acquisition nodes is greater than or equal to the difference threshold are used as segmentation nodes. All data acquisition nodes are traversed to complete time period segmentation, and the time periods contained in adjacent segmentation nodes are used as feature time periods. The peak-valley difference Q recorded in each feature time period is extracted. 差 The peak-valley difference represents the difference between the maximum and minimum values ​​of power load data within a characteristic period. The sum of peak-valley differences Q1 for each corresponding safe period in the same historical monitoring area is extracted, and the average value Q is calculated for the sum of peak-valley differences across all safe periods. 1均 The peak-to-valley differences within the safety period are also extracted based on the segmented characteristic time periods, and the segmentation method for these characteristic time periods is the same as that for the observation period; using the formula: F=(∑Q 差 ) / Q 1均 Calculate the output data stability ratio F, ∑Q, for a predicted event of missing power load data of the corresponding type. 差 This represents the sum of peak-to-valley differences for each characteristic time period in the power load data acquisition graph corresponding to a predicted event. The output data stability ratio reflects the stage that the power load data recorded during the missing time period in the monitored power load data may be in. The peak-to-valley differences of power load data at different nodes in the entire observation period may have significant deviations, thus providing a certain assessment of the importance of the period in which the missing time period is located. The larger the output data stability ratio, the smaller the deviation and the smaller the impact on the stability of the power load data corresponding to the missing time period.

[0065] The data required for drawing the power load data acquisition diagram in this application is numerical power consumption data;

[0066] Step S230: Extract the prediction result value U1 and actual load value U2 of the prediction model for the same type of missing power load data in each prediction event. In this application, the prediction model used for the same missing power load data is the same. The prediction model can include ARIMA model, artificial neural network model, etc.; use the formula: Y=1-[|U2-U1| / U2] to calculate the prediction accuracy value Y for a prediction event corresponding to the same type of missing power load data; the larger the prediction accuracy value, the smaller the deviation between the predicted value and the actual value, and the smaller the impact of the missing data in the corresponding prediction event on the prediction result;

[0067] Step S240: Take the output data completion ratio and output data stability ratio of the missing power load data records of the same type as input variables, and the prediction accuracy value Y corresponding to the same prediction event as the output variable to generate a data set A, A = [(D, F), Y]. Summarize the data sets A corresponding to all prediction events of the missing power load data records of the same type, and construct a response filling demand function y, y = k1*d D + k2*f F + ε; Input the corresponding data set A to generate a response filling demand function y that records the reference coefficients k1, k2, and the error term ε;

[0068] Step S250: Set a threshold Y0 for the prediction accuracy value, and construct a response filling demand model as follows: Input the output data completion ratio and output data stability ratio obtained in real time into the response filling demand function y. If the theoretical prediction accuracy value Y 测 ≥ Y0, no response is triggered. If the theoretical prediction accuracy value Y 测 < Y0, a filling demand response is triggered;

[0069] Step S260: Traverse and find the prediction accuracy values of all prediction events of each type of missing power load data record. If there are prediction accuracy values of all prediction events greater than the threshold Y0, mark the corresponding type of missing power load data as a safe power load data type, and extract the maximum missing duration as the critical missing monitoring value; otherwise, it is a type of missing power data to be analyzed.

[0070] Step S300 includes the following:

[0071] Step S310: The time period type refers to the time period difference divided by the peak-valley difference within the characteristic time period. Calculate the average value Q 平 and standard deviation Q 标差 of the peak-valley differences of all characteristic time periods recorded in each prediction event of the same type of missing power data to be analyzed. Use [Q 平 - Q 标差 , Q 平 + Q 标差 as the interval length to divide all characteristic time periods of the complete inspection period, generate the corresponding time period type; and extract the power load data acquisition graph corresponding to the time from the start time of the inspection period to before the missing as the characteristic acquisition graph;

[0072] Step S320: Use the characteristic acquisition graph, the reason for the missing, the missing duration, and the characteristic time period corresponding to the missing as the scenario monitoring characteristics of the corresponding type of missing power data to be analyzed;

[0073] Step S330: Predicted events within the safety period when scene features are the same refer to predicted events within the corresponding safety period when the similarity of the feature acquisition maps is greater than the similarity threshold, the missing reasons are the same, the absolute value of the difference in missing duration is less than the duration difference threshold, and the corresponding feature time periods are the same; these are marked as control events, and the power load data of the control events corresponding to the overlapping parts with the missing time periods are extracted as filling data for the predicted events corresponding to each missing time period; and filling data sets of the associated filling data under different scene monitoring features corresponding to the missing power data to be analyzed for each type are generated.

[0074] The construction of the data set can accurately fill in power load data under different missing scenarios for different types of missing power data, thereby greatly reducing the problem of reduced accuracy in analyzing and predicting power load under missing data, and improving the efficiency of diversified analysis and intelligent response of data filling scenarios.

[0075] Step S400 includes the following:

[0076] Obtain the missing reason corresponding to the missing segment of real-time data. If the missing reason is the same as the missing reason corresponding to the safe power load data, obtain the missing duration of the missing segment record of real-time data. If the missing duration is less than or equal to the critical missing monitoring value of the corresponding safe power load data, do not respond to data filling. If the missing duration is greater than the critical missing monitoring value of the corresponding safe power load data, return to steps S200 to S300, construct the response filling requirement model of the type of power load data and output the filling data set.

[0077] When the cause of the missing data is the same as the cause of the missing power data to be analyzed, the real-time scene features of the missing segment record of the real-time data are obtained, the filling data set of the corresponding missing power data record to be analyzed is found, and the scene monitoring features that are the same as the real-time scene features and the corresponding filling data are matched to respond.

[0078] A power load forecasting and analysis system based on big data includes a data statistics module, a periodic marking module, a power load data classification module, a response filling demand model construction module, a data type division module, a scenario feature output module, a filling data set generation module, and a real-time response filling module.

[0079] The data statistics module is used to collect historical power load data and predict events for each monitoring area;

[0080] The period marking module is used to divide historical data into several event periods based on the recording period of the power load data required for predicting events; and to identify missing data segments in the event periods, marking event periods with missing data segments as examination periods and event periods without missing data segments as safety periods;

[0081] The power load data classification module is used to identify and classify data based on the cause of missing data.

[0082] The response filling requirement model building module is used to obtain the missing duration of the missing data segment records for each type of missing power load data and build the response filling requirement model.

[0083] The data type partitioning module is used to output the safe power load data type and the missing power data type to be analyzed;

[0084] The scenario feature output module is used to summarize the missing duration and time period type within the observation period, as well as the specific reasons for the missing, to determine the scenario features of power load monitoring;

[0085] The data set generation module is used to analyze and output data sets for missing power data of various types to be analyzed.

[0086] The real-time response filling module is used to identify missing segments in real-time data, determine the data type of the power load, and match and fill the missing data set based on the determination result.

[0087] The response-filling demand model building module includes an output data completion ratio calculation unit, an output data stability ratio calculation unit, a prediction accuracy value calculation unit, a response-filling demand function analysis unit, and a model building unit;

[0088] The output data completion ratio calculation unit is used to calculate the output data completion ratio of the corresponding type of missing power load data under a prediction event.

[0089] The output data stability ratio calculation unit is used to calculate the output data stability ratio of the corresponding type of missing power load data under a prediction event.

[0090] The prediction accuracy calculation unit is used to extract the prediction results and actual load values ​​of the prediction model for the same type of missing power load data in each prediction event, and calculate the prediction accuracy value.

[0091] The response filling demand function analysis unit is used to construct the response filling demand function by taking the output data completion ratio and output data stability ratio of the same type of missing power load data records as input variables and the prediction accuracy value under the same prediction event as output variables.

[0092] The model building unit is used to set the prediction accuracy threshold and build a response filling requirement model.

[0093] The data type classification module includes a safe power load data type marking unit and a missing power data type marking unit to be analyzed;

[0094] The safe power load data type marking unit is used to traverse and search for the prediction accuracy value of all prediction events of each type of missing power load data record. If the prediction accuracy value of all prediction events is greater than the threshold, the corresponding type of missing power load data is marked as a safe power load data type.

[0095] The missing power data type marker unit is used to mark a missing power data type as to be analyzed when the judgment conditions of the safe power load data type marker unit are not met.

[0096] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A power load forecasting and analysis method based on big data, characterized in that: The method includes: Step S100: Statistically analyze the historical power load data and predicted events in each monitoring area. The predicted event refers to an event that executes a power prediction objective based on the power load data within a time period. Divide the historical data into several event periods according to the recording period of the power load data required for the predicted event; identify missing data segments in the event periods, and mark the event periods with missing data segments as observation periods and the event periods without missing data segments as safety periods; extract the missing power load data for the missing data segments in all observation periods, and perform data identification and classification based on the cause of the missing data. Step S200: Obtain the missing duration of the missing data segment record for each type of missing power load data, construct a response filling requirement model, and output the safe power load data type and the missing power data type to be analyzed; The system iterates through all predicted events for each type of missing power load data record and finds the prediction accuracy value. If the prediction accuracy value of all predicted events is greater than the threshold, the corresponding type of missing power load data is marked as a safe power load data type, and the maximum missing duration is extracted as the critical missing monitoring value; otherwise, it is a missing power load data type to be analyzed. Step S300: Extract the set of missing time periods and the observation period of the missing power data history to be analyzed, summarize the missing duration and time period type within the observation period, as well as the specific reasons for the missing data, and determine the scenario characteristics of power load monitoring; retrieve the predicted events within the safe period when the scenario characteristics are the same, and analyze and output the filling data set of each type of missing power data to be analyzed. Step S400: When the response identifies a missing segment of real-time data, determine the type of power load data, and fill the data set response based on the determination result; Step S400 includes the following: Obtain the missing reason corresponding to the missing segment of real-time data. If the missing reason is the same as the missing reason corresponding to the safe power load data, obtain the missing duration of the missing segment record of real-time data. If the missing duration is less than or equal to the critical missing monitoring value of the corresponding safe power load data, do not respond to data filling. If the missing duration is greater than the critical missing monitoring value of the corresponding safe power load data, return to steps S200 to S300, construct the response filling requirement model of the power load data of the aforementioned type, and output the filling data set. When the cause of the missing data is the same as the cause of the missing power data to be analyzed, the real-time scene features of the missing segment record of the real-time data are obtained, the filling data set of the corresponding missing power data record to be analyzed is found, and the scene monitoring features that are the same as the real-time scene features and the corresponding filling data are matched to respond.

2. The power load forecasting and analysis method based on big data according to claim 1, characterized in that: Step S100 includes the following specific steps: Extract the minimum node interval duration and the average node interval duration of adjacent power load data acquisition time nodes in each event cycle. When there is a record node interval duration and the minimum node interval duration with a difference greater than the first difference threshold and greater than the average node interval duration, the interval corresponding to the node interval duration is output as the data missing segment. Obtain the reasons for missing power load data records within the missing data segment, including equipment failure, environmental interference, and human operation; mark missing power load data with the same missing reason recorded in the same monitoring area as a group of data, and use the predicted event as the marker index to distinguish missing power load data recorded in different events under the same group of data.

3. The power load forecasting and analysis method based on big data according to claim 1, characterized in that: Step S200 includes the following specific steps: Step S210: Obtain the average data recording amount q of each type of missing power load data in the monitoring area within a unit time period, and the total unit time T of the missing power load data of that type in each observation period, extract the actual storage amount Q of the load data in the corresponding observation period, and calculate the output data completion ratio D of the corresponding type of missing power load data under a prediction event, D=Q / qT; Step S220: Based on the time series, the unit duration of the records in each observation period of the same type of missing power load data is taken as the data acquisition node. The power load data acquisition map of each prediction event is drawn with the data acquisition node as the horizontal axis and the power load data collected within the corresponding unit duration of the acquisition node as the vertical axis. The nodes with the difference between the power load data recorded by adjacent acquisition nodes that is greater than or equal to the difference threshold are taken as the segmentation nodes. All data acquisition nodes are traversed to complete the time period segmentation. The time periods contained in adjacent segmentation nodes are taken as the feature time periods. extract Peak-to-valley difference Q recorded in each characteristic time period 差 The peak-valley difference represents the difference between the maximum and minimum values ​​of power load data within a characteristic period. The sum of peak-valley differences Q1 for each corresponding safe period in the same historical monitoring area is extracted, and the average Q of the sum of peak-valley differences for all safe periods is calculated. 1均 Using the formula: F=(∑Q 差 ) / Q 1均 Calculate the output data stability ratio F, ∑Q, for a predicted event of missing power load data of the corresponding type. 差 This represents the sum of peak-valley differences for each characteristic time period in the power load data acquisition graph corresponding to a predicted event; Step S230: Extract the prediction result value U1 and actual load value U2 of the prediction model for the same type of missing power load data in each prediction event, and use the formula: Y=1-[|U2-U1| / U2] to calculate the prediction accuracy value Y for the same type of missing power load data in a prediction event; Step S240: Using the output data completion ratio and output data stability ratio of missing power load data records of the same type as input variables, and the prediction accuracy value Y under the same prediction event as output variables, generate data group A, A=[(D, F),Y]. Summarize the data group A corresponding to all prediction events of missing power load data records of the same type, and construct the response filling demand function y, y=k1*d D +k2*f F +ε; Input the corresponding data group A to generate the response filling requirement function y for the record reference coefficients k1, k2 and error term ε; Step S250: Set the prediction accuracy value threshold Y0, and construct the response filling demand model as follows: Input the output data completion ratio and the output data stability ratio obtained in real time into the response filling demand function y. If the theoretical prediction accuracy value Y 测 ≥ Y0, no response is triggered. If the theoretical prediction accuracy value Y 测 < Y0, trigger the filling demand response.

4. The power load forecasting and analysis method based on big data according to claim 3, characterized in that: Step S300 includes the following: Step S310: The time period type refers to the time period difference divided by the peak-valley difference within the characteristic time period. Calculate the average peak-valley difference Q of all characteristic time periods recorded in each predicted event of the same type of missing power data to be analyzed. 平 Sum of standard deviation Q 标差 , with [Q 平 -Q 标差 Q 平 +Q 标差 The interval length is used to divide all characteristic time periods of the complete examination period and generate the corresponding time period type; The power load data collection map from the start of the observation period to the point before the data loss was extracted was used as the feature collection map. Step S320: Use the feature acquisition map, the cause of the missing data, the duration of the missing data, and the feature time period corresponding to the missing data as the scene monitoring features of the power missing data to be analyzed for the corresponding type; Step S330: The predicted events within the safe period when the scene features are the same refer to the predicted events within the safe period when the similarity of the feature acquisition maps is greater than the similarity threshold and the missing reasons are the same, the absolute value of the difference in missing duration is less than the duration difference threshold, and the feature time periods corresponding to the missing are the same. The data are marked as control events, and the power load data of the control events corresponding to the parts that overlap with the missing time periods are extracted as filling data for the predicted events corresponding to each missing time period. A set of filling data is generated for the corresponding monitoring features of different scenarios for each type of missing power data to be analyzed.

5. A power load forecasting and analysis system based on big data, such as using the power load forecasting and analysis method based on big data according to any one of claims 1-4, characterized in that: The system includes a data statistics module, a periodic marking module, a power load data classification module, a response filling demand model construction module, a data type division module, a scenario feature output module, a filling data set generation module, and a real-time response filling module. The data statistics module is used to collect historical power load data and predict events for each monitoring area; The cycle marking module is used to divide historical data into several event cycles based on the recording cycle of the power load data required for predicting events; and to identify data missing segments in the event cycles, marking event cycles with data missing segments as examination cycles and event cycles without data missing segments as safety cycles; The power load data classification module is used to identify and classify data based on the cause of missing data. The response filling requirement model construction module is used to obtain the missing duration of the missing data segment records for each type of missing power load data, and construct the response filling requirement model. The data type segmentation module is used to output the safe power load data type and the missing power data type to be analyzed. The scenario feature output module is used to summarize the missing duration and time period type within the observation period, as well as the specific reasons for the missing, to determine the scenario features of power load monitoring. The data set filling generation module is used to analyze and output data sets for filling missing power data of various types to be analyzed. The real-time response filling module is used to determine the data type of the power load when a missing segment of real-time data is identified, and to match and fill the missing data set based on the determination result.

6. The power load forecasting and analysis system based on big data according to claim 5, characterized in that: The response filling demand model construction module includes an output data completion ratio calculation unit, an output data stability ratio calculation unit, a prediction accuracy value calculation unit, a response filling demand function analysis unit, and a model construction unit; The output data completion ratio calculation unit is used to calculate the output data completion ratio of the corresponding type of missing power load data under a prediction event; The output data stability ratio calculation unit is used to calculate the output data stability ratio of the corresponding type of missing power load data under a prediction event; The prediction accuracy calculation unit is used to extract the prediction result value of the prediction model and the actual load value of the missing power load data of the same type in each prediction event, and calculate the prediction accuracy value. The response filling demand function analysis unit is used to construct the response filling demand function by taking the output data completion ratio and output data stability ratio of the same type of missing power load data records as input variables and the prediction accuracy value under the same prediction event as output variables. The model building unit is used to set a threshold for the prediction accuracy value and build a response filling demand model.

7. The power load forecasting and analysis system based on big data according to claim 6, characterized in that: The data type classification module includes a safe power load data type marking unit and a missing power data type marking unit to be analyzed. The safe power load data type marking unit is used to traverse and search for the prediction accuracy value of all prediction events of each type of missing power load data record. If the prediction accuracy value of all prediction events is greater than the threshold, the corresponding type of missing power load data is marked as a safe power load data type. The missing power data type marking unit is used to mark a missing power data type as to be analyzed when the judgment conditions of the safe power load data type marking unit are not met.

Images