Power load prediction method and system

Abstract

The invention discloses a power load prediction method and system, and the method comprises the steps: S1, obtaining the historical operation data of power equipment, and obtaining first application industry information based on the historical operation data of the power equipment; s2, obtaining a pre-screening load subject research record, and obtaining second application industry information based on the pre-screening load subject research record; s3, acquiring an industry correlation value based on the first application industry information and the second application industry information; s4, screening the pre-screened load subject research records based on the industry correlation value and a preset optimal threshold to obtain a target load subject research record; and S5, constructing a power load prediction model according to the target load main body research record, and outputting power equipment load prediction data based on the power load prediction model and the power equipment real-time operation data. The method has the beneficial effects that the power load prediction model is specifically trained for subsequent load prediction, and the accuracy of power system load prediction is improved.

Description

Technical Field

[0001] The present application relates to the technical field of power load forecasting, and in particular to a power load forecasting method and system. Background Art

[0002] Power load forecasting plays a vital role in the operation of smart grids. It has a decisive influence on ensuring the safety, stability and economic operation of the power system. The core of this technology is to use historical load data and various related influencing factors, such as weather conditions, human activity patterns, types of industrial production processes, and time and seasonal characteristics, to predict future power load conditions. Through comprehensive analysis of these data and factors, power load forecasting can help the power grid accurately estimate future power consumption, power demand and power consumption, thereby providing a scientific basis for the dispatch, planning and management of the power system. This kind of forecasting not only helps to avoid shortages or surpluses in power supply, but also effectively reduces the operating costs of the power system, improves energy efficiency, and ultimately achieves optimal allocation and sustainable development of power resources.

[0003] However, in the prior art, all historical data are usually used to construct the power load forecasting model, which will cause the load data of different load entities in all historical data to influence each other. Different load entities may have different load characteristics and change laws. Mixing their data may not accurately capture the unique characteristics of each entity, resulting in the final constructed power load forecasting model being less targeted, and the prediction results of the future load conditions of one of the load entities are less accurate, and cannot accurately fit the load change law of each entity, resulting in deviations in the final prediction results.

[0004] Chinese patent "Electricity Load Forecasting Method, Electricity Load Forecasting Model Training Method and Device", publication number: CN116995673A, publication date: November 3, 2023, specifically discloses: by obtaining time series data related to the power load in the target area, and inputting the time series data into a preset power load forecasting model. Furthermore, multiple time series features of the time series data are extracted through a preset power load forecasting model, and the multiple time series features are fused to output power load forecasting data of various time length types in the target area. In this scheme, a preset power load forecasting model is used to predict the power load. There is no targeted training of the power load forecasting model, and different load subjects have different subject characteristics. The use of a general power load forecasting model has the problem of weak prediction targeting. Furthermore, the accuracy of the load forecast is also low.

[0005] The Chinese patent "Power System Short-term Load Forecasting Method, System, Equipment and Medium", publication number: CN112215442A, publication date: January 12, 2021, specifically discloses: obtaining load data and load impact data of the power system in a preset period before the period to be predicted; obtaining load impact data of the power system in the period to be predicted; according to the load impact data of the power system in the period to be predicted, and the load data and load impact data of the preset period before the period to be predicted, the load forecasting network model is used to predict and obtain the load forecasting result of the power system in the period to be predicted. This scheme uses the load data and load impact data of the preset period before the period to be predicted to construct the prediction model, and there is also the problem of weak prediction pertinence. Summary of the invention

[0006] In view of the problem in the prior art that the historical data of all load entities are used to construct an electric load prediction model, the electric load prediction model has poor pertinence. The present application provides an electric load prediction method and system, which obtains an electric load prediction model by screening target load entity research records that match the corresponding load entities in the historical operation data of electric equipment, and using a neural network model to learn the experience of manually predicting load changes in the target load entity research records; using the established electric load prediction model, the load conditions of the future load entities are predicted according to the electric equipment data, and the electric load prediction model is trained in a targeted manner for subsequent load prediction, thereby improving the accuracy of load prediction of the power system.

[0007] To achieve the above-mentioned technical objectives, a technical solution provided by the present application is a method for power load forecasting, comprising: S1: obtaining historical operation data of power equipment, and obtaining first application industry information based on the historical operation data of power equipment; S2: obtaining pre-screened load subject research records, and obtaining second application industry information based on the pre-screened load subject research records; S3: obtaining industry-related values ​​based on the first application industry information and the second application industry information; S4: screening the pre-screened load subject research records based on the industry-related values ​​and the preset preferred thresholds to obtain target load subject research records; S5: constructing a power load forecasting model based on the target load subject research records, and outputting power equipment load forecasting data based on the power load forecasting model and the real-time operation data of the power equipment.

[0008] Furthermore, the industry-related value is: an industry information similarity value between the first application industry information and the second application industry information.

[0009] Furthermore, S1 also includes: obtaining predicted rhythm based on historical operating data of power equipment; S2 also includes: obtaining research rhythm based on pre-screened load subject research records; S3 also includes: obtaining rhythm multiplexing value based on predicted rhythm and research rhythm; S4 also includes: screening pre-screened load subject research records based on industry-related values, rhythm multiplexing values ​​and preset preferred thresholds.

[0010] Furthermore, S4 also includes: obtaining a preferred value of the pre-screening load subject research record according to the industry-related value and rhythm multiplexing value corresponding to the pre-screening load subject research record; and taking the pre-screening load subject research record whose preferred value is greater than or equal to a preset preferred threshold as the target load subject research record.

[0011] Furthermore, the method of obtaining a rhythm reuse value based on the predicted rhythm and the research rhythm also includes: obtaining a rhythm inclusion relationship between the research rhythm and the predicted rhythm; if the rhythm inclusion relationship is that the research rhythm contains the predicted rhythm, then expanding the research information item on the time axis corresponding to the research rhythm; matching the information attribute of the research information item with the credible information attribute to obtain a credible value of the research information item; determining the time interval corresponding to the time axis of the research rhythm; obtaining a first average value of the credible values ​​of the research information items falling within the time interval; obtaining a second average value of the credible values ​​of the research information items before the time interval on the time axis; assigning a first weight to the first average value and a second weight to the second average value, wherein the first weight is greater than the second weight; multiplying the first average value and the first weight accordingly to obtain a first target value; multiplying the second average value and the second weight accordingly to obtain a second target value; summing the first target value and the second target value corresponding to the research rhythm to obtain a rhythm reuse value.

[0012] Furthermore, it also includes: S6: obtaining real-time load data of power equipment, comparing the real-time load data of power equipment and the predicted load data of power equipment, obtaining the load data difference, and if the load data difference reaches the preset analysis condition, performing difference analysis based on the load data difference and obtaining the load data difference analysis result; S7: based on the load data difference analysis result, determining the model adjustment strategy of the power load prediction model and performing model adjustment according to the model adjustment strategy.

[0013] Furthermore, the S6 also includes: determining a first load data curve corresponding to the real-time load data of the power equipment, and determining a second load data curve corresponding to the predicted load data of the power equipment; aligning the first load data curve and the second load data curve in time to obtain an alignment result; based on the analysis condition trigger judgment template and the alignment result, judging whether the alignment result satisfies any one of the analysis condition trigger conditions in the analysis condition trigger judgment template, and if so, judging that the load data difference reaches the preset analysis condition.

[0014] Furthermore, S6 also includes: obtaining historical warning omission records; attributing warning omissions to historical warning omission records, determining historical warning omission records whose attribution results are warning omissions caused by prediction differences at the time of historical warning omissions, and using them as target warning omission records; determining the historical warning omission time corresponding to the target warning omission record, and using them as the backtracking time; obtaining curve difference characteristics of the historical actual load data curve and the historical forecast load data curve for a preset target time length before the backtracking time; calculating feature similarity values ​​between the curve difference characteristics, and determining target curve difference characteristics that are different from each other based on the calculation results of the feature similarity values; using the target curve difference characteristics as analysis condition trigger conditions, and summarizing them to form an analysis condition trigger determination template.

[0015] Furthermore, the method of judging whether the alignment result satisfies any one of the analysis condition trigger conditions in the analysis condition trigger judgment template based on the analysis condition trigger judgment template includes: parsing the analysis condition trigger judgment template to obtain a search window of the analysis condition trigger condition corresponding to the alignment result; sliding the analysis condition trigger condition in the corresponding search window to obtain a curve difference feature similarity between the sliding analysis condition trigger condition and the alignment result at the sliding position; if the curve difference feature similarity is greater than or equal to a preset curve difference feature similarity threshold, judging that the alignment result at the corresponding sliding position satisfies the sliding analysis condition trigger condition; wherein, sliding the analysis condition trigger condition in the corresponding search window is as follows: sliding the analysis condition trigger condition from the middle position of the corresponding search window to both sides of the search window respectively.

[0016] Another technical solution provided by the present application is a power load forecasting system for implementing the above-mentioned method, including: a power equipment data reading subsystem for reading power equipment operation data; a target load subject research record screening subsystem for screening target load subject research records of the research load subject according to the historical operation data of the power equipment; a power load forecasting model training subsystem for training the power load forecasting model according to the target load subject research records; a future power equipment load data forecasting subsystem for forecasting future power equipment load data based on the power load forecasting model and the real-time operation data of the power equipment.

[0017] Beneficial effects of this application: 1. Obtain industry information related to the application of power equipment through the historical operation data of power equipment, as the first application industry information of the load subject corresponding to the power equipment, obtain the load subject application industry information in the research record by pre-screening the load subject research record, as the second application industry information, obtain the industry correlation value according to the correlation between the first application industry information and the second application industry information, screen the pre-screened load subject research record based on the industry correlation value and the preset preferred threshold, so that the target load subject research record for constructing the power load forecasting model matches the main characteristics of the power equipment, improves the prediction pertinence of the power load forecasting model, and improves the accuracy of the load forecasting of the power equipment.

[0018] 2. Add the calculation of rhythm reuse value, and select the load subject research records whose time regularity information is similar to the time regularity of the power equipment load and whose application industry information is similar to the industry applied by the power equipment load subject as the target load subject research records, so as to avoid the influence of the load subject with inconsistent time regularity on the final prediction result, improve the pertinence of the prediction, and further improve the accuracy of the power equipment load prediction.

[0019] 3. The curve difference characteristics of the historical actual load data curve and the historical predicted load data curve are obtained through the historical warning omission records, and the target curve difference characteristics are used as the analysis condition trigger condition, so that the analysis condition trigger condition is based on the actual data curve and the predicted data curve with missed warnings, so that the final curve difference (alignment result) matches the curve difference (analysis condition trigger condition), which is more accurate and further improves the accuracy of model correction. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A schematic diagram of a flow chart of a power load forecasting method according to the present application.

[0021] Figure 2 This is a structural schematic diagram of a power load forecasting system for this application. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solutions and advantages of the present application more clear, the present application is further described in detail below in conjunction with the drawings and examples. It should be understood that the specific implementation method described here is only an optimal embodiment of the present application, which is only used to explain the present application and does not limit the scope of protection of the present application. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0023] like Figure 1 As shown, as an embodiment 1 of the present application, a method for predicting power load includes: S1: Acquire historical operation data of electric power equipment, and acquire first application industry information based on the historical operation data of electric power equipment; S2: obtaining a pre-screening load subject research record, and obtaining second application industry information based on the pre-screening load subject research record; S3: obtaining industry-related values ​​based on the first application industry information and the second application industry information; S4: Screening the pre-screened load subject research records based on industry-related values ​​and preset optimization thresholds to obtain target load subject research records; S5: Construct a power load forecasting model based on the target load subject research records, and output power equipment load forecasting data based on the power load forecasting model and the real-time operation data of power equipment.

[0024] In this embodiment, industry information related to the application of power equipment is obtained through the historical operation data of the power equipment, as the first application industry information of the load subject corresponding to the power equipment, and the load subject application industry information in the research record is obtained by pre-screening the load subject research record as the second application industry information. The industry correlation value is obtained according to the correlation between the first application industry information and the second application industry information, and the pre-screened load subject research record is screened based on the industry correlation value and the preset preferred threshold value, so that the target load subject research record for constructing the power load forecasting model matches the main characteristics of the power equipment, thereby improving the prediction pertinence of the power load forecasting model and improving the accuracy of the load forecasting of the power equipment.

[0025] The historical operation data of power equipment refers to the historical operation data of various equipment (such as generators, transformers and transmission lines) collected from the power system. The load subject refers to the end user or equipment that consumes electric energy in the power system. The pre-screening load subject research record at least includes the record of load changes of the research load subject obtained based on the basic information of the load subject and the historical electricity consumption data obtained by big data.

[0026] The power load forecasting model is: an intelligent AI model for predicting future load changes in the power system, such as a neural network model, which can be a BP neural network model. During training, the experience of manually predicting load changes in the research records of the target load subject is learned to obtain the power load forecasting model; By reading the power equipment data, screening the target load subject research records of the load subject corresponding to the power equipment data, and using the neural network model to learn the experience of manually predicting load changes in the target load subject research records, the power load forecasting model is obtained; using the established power load forecasting model, the load conditions of the future load subject are predicted according to the power equipment data, and the power load forecasting model is trained specifically for subsequent load forecasting, thereby improving the accuracy of the power system load forecasting.

[0027] In some embodiments, obtaining historical operation data of the power equipment includes: Based on the Internet of Things technology, the historical operation data of the power equipment is read according to the power equipment data reading node.

[0028] The power equipment data reading node is a communication node connected to the power equipment. By introducing the Internet of Things technology and reading the power equipment data through the power equipment data reading node, the rationality of the power equipment data reading is improved.

[0029] Based on the pre-screening load subject research records, the second application industry information is obtained, including: The second application industry information is obtained according to the application industry of the load subject in the pre-screening load subject research record.

[0030] In this embodiment, the first application industry information is: the industry field information of the load subject application in the power equipment data, such as: commercial, industrial and civil, etc. The second application industry information is: the industry field information of the load subject application in the pre-screened load subject research record. At this time, the industry-related value is: Industry information similarity value between the first application industry information and the second application industry information.

[0031] The target load subject research record is obtained by selecting the load subject research record whose industry information similarity value is higher than the preset preferred threshold. Only when the matching value of the load subject research record and the corresponding power equipment reaches the preset requirement, it will be used as the training value of the power load prediction model, ensuring that the power load prediction model is targeted at the current power equipment, avoiding the load subject data of other industries from affecting the prediction results, and improving the accuracy of the load prediction of the power equipment.

[0032] As a second embodiment of the present application, a method for predicting power load further includes: S1: Acquire historical operation data of power equipment, and acquire first application industry information and predicted rhythm based on the historical operation data of power equipment; S2: Obtain the pre-screening load subject research record, and obtain the second application industry information and research rhythm based on the pre-screening load subject research record; S3: obtaining industry-related values ​​based on the first application industry information and the second application industry information, and obtaining rhythm reuse values ​​based on the predicted rhythm and the research rhythm; S4: Screening the pre-screened load subject research records based on the industry-related value, the rhythm reuse value and the preset optimization threshold to obtain the target load subject research records; S5: Construct a power load forecasting model based on the target load subject research records, and output power equipment load forecasting data based on the power load forecasting model and the real-time operation data of power equipment.

[0033] In this embodiment, the calculation of the rhythm reuse value is added. By checking whether the rhythm reuse value meets the preset preferred threshold, the load subject research records whose time regularity information is similar to the time regularity of the power equipment load and whose application industry information is similar to the industry applied by the power equipment load subject are screened as the target load subject research records, thereby avoiding the influence of the load subject with inconsistent time regularity on the final prediction result, improving the targetedness of the prediction, and further improving the accuracy of the power equipment load prediction.

[0034] Among them, the research rhythm is: the time regularity information of the load studied in the pre-screened load subject research record. The predicted rhythm is: the predicted time regularity information of the load of the load subject predicted based on the power equipment data. The rhythm reuse value is: the quantitative representation of the reference degree of the pre-screened load subject research record inferred based on the rhythm relationship.

[0035] Step S4 also includes: Obtaining the preferred value of the pre-screening load subject research record according to the industry-related value and rhythm reuse value corresponding to the pre-screening load subject research record; The pre-screening load subject research record whose preferred value is greater than or equal to the preset preferred threshold is used as the target load subject research record.

[0036] The preset priority threshold can be preset according to actual conditions.

[0037] Quantify the similarity between the research records of the pre-screened load subject and the current power equipment load in terms of time rules and the matching degree of their application industry information, and screen the research records similar to the target load subject.

[0038] Among them, obtaining the preferred value of the pre-screening load subject research record according to the industry-related value and rhythm reuse value corresponding to the pre-screening load subject research record also includes: Perform data normalization on the industry-related values ​​and rhythm multiplexing values ​​corresponding to the pre-screening load subject study records; The weighted sum of the industry-related values ​​and rhythm reuse values ​​after data standardization is performed to obtain the preferred values ​​of the pre-screening load subject research records.

[0039] When training the power load prediction model, the training values ​​of the load subject research records obtained by big data are different. The load subject research records with high value are screened for subsequent training, which is more targeted. Therefore, the present application obtains pre-screened load subject research records through big data, determines the first application industry information of the load subject in the power equipment data, and at the same time, obtains the second application industry information of the load subject in the pre-screened load subject research records, calculates the industry-related values ​​of the first application industry information and the second application industry information; parses the pre-screened load subject research records, obtains the research rhythm, and also obtains the predicted rhythm of the load subject; calculates the rhythm reuse value of the research rhythm and the predicted rhythm; cumulatively calculates the industry-related values ​​and rhythm reuse values ​​corresponding to the pre-screened load subject research records, obtains the preferred value, and screens the pre-screened load subject research records whose preferred value is greater than or equal to the preset preferred value threshold as the target load subject research records, thereby improving the availability of the target load subject research records.

[0040] In some embodiments, obtaining a rhythm multiplexing value based on the predicted rhythm and the studied rhythm further includes: Obtain the rhythm inclusion relationship between the studied rhythm and the predicted rhythm; If the rhythm inclusion relationship is that the research rhythm includes the predicted rhythm, the research information item is expanded on the time axis corresponding to the research rhythm; the information attribute of the research information item is matched with the credible information attribute to obtain the credible value of the research information item; Determine the time interval of the research rhythm corresponding to the time axis; Obtaining a first average of the credibility values ​​of the research information items falling within the time interval; Obtaining a second average value of the credibility values ​​of the research information items before the time interval on the time axis; Assigning a first weight to the first average value, and assigning a second weight to the second average value, wherein the first weight is greater than the second weight; Multiply the first average value and the first weight correspondingly to obtain a first target value; Multiply the second average value and the second weight correspondingly to obtain a second target value; The first target value and the second target value corresponding to the studied rhythm are summed to obtain the rhythm multiplexing value.

[0041] Among them, the rhythm inclusion relationship includes: the research rhythm includes the predicted rhythm, the predicted rhythm is partially included in the research rhythm, and the predicted rhythm and the research rhythm are unrelated. The research information item is: the research content within the research rhythm. The information attribute is: the content characteristics of the research content within the research rhythm, such as: what research method is used, what research standard is adopted; the credible information attribute is: the credibility index of the research information item, such as: the normative index of the research method; the credible value indicates the credibility of the content of the research information item. The time interval is: the local time range of the research rhythm corresponding to the time axis.

[0042] Since the research information items falling within the time interval are directly used as references, and the credibility values ​​of the research information items before the time interval indirectly affect the credibility of the research information items, the first weight is given a greater weight.

[0043] In this embodiment, the rhythm inclusion relationship is determined as the time axis corresponding to the research rhythm in which the research rhythm includes the predicted rhythm, and the research information items in the corresponding research rhythm are correspondingly expanded on the time axis; the information attributes of the research information items are matched with the credible information attributes to determine the credible value of the research information items; the time interval corresponding to the time axis of the research rhythm is determined, and the credible value of the research information items falling within the time interval directly represents the reference quality of the corresponding information, and the first average value is obtained; the second average value of the credible value of the research information items before the time interval on the time axis affects the research information items falling within the time interval, so a larger value is given to the first weight; the first average value and the first weight are correspondingly multiplied to obtain the first target value, the second average value and the second weight are correspondingly multiplied to obtain the second target value, and the first target value and the second target value corresponding to the research rhythm are summed to obtain the rhythm reuse value. By comparing the inclusion relationship between the research rhythm and the predicted rhythm, the research records that are most similar to the time law of the target load subject can be accurately found, reducing the prediction error caused by the inconsistency of the time law. At the same time, when calculating the rhythm reuse value, the credibility of the research information item is taken into account, and different weights are assigned to the research information items within the time interval and before the time interval, respectively, to ensure that more credible research information has a greater impact on the prediction results, thereby improving the accuracy of the prediction. Consideration of multiple factors makes the prediction model more robust to various changing factors, and can effectively avoid interference from abnormal data, ensuring the stability and reliability of the power load prediction model.

[0044] As a third embodiment of the present application, a method for predicting power load further includes: S6: Acquire real-time load data of the power equipment, compare the real-time load data of the power equipment with the predicted load data of the power equipment, obtain the load data difference, and if the load data difference reaches a preset analysis condition, perform difference analysis based on the load data difference and obtain the load data difference analysis result; S7: Based on the load data difference analysis results, determine the model adjustment strategy of the power load forecasting model and perform model adjustment according to the model adjustment strategy.

[0045] In this embodiment, the load data difference is obtained based on the difference between the real-time load data of the power equipment and the load forecast data of the power equipment, and it is judged whether the load data difference is too large according to the preset analysis conditions, and then it is judged whether the output of the power load forecast model deviates from the actual value. If so, the power load forecast model is adjusted accordingly. This avoids the deviation of the power load forecast caused by the change of electricity consumption habits, ensures that the output result of the power load forecast model closely follows the actual load situation in actual application, and ensures the accuracy of the power equipment load forecast.

[0046] The preset analysis conditions at least include: a difference threshold, a duration threshold and a frequency threshold. At this time, step S6 also includes: If the load data difference does not meet the preset analysis conditions, the number of load data difference occurrences is recorded.

[0047] When the load data difference does not exceed the difference threshold, the number of load data difference occurrences in the current time series is 1, and whether the load data difference meets the preset analysis conditions is further judged based on whether the duration of the occurrence of 1 is greater than the duration threshold and whether the frequency of the occurrence of 1 in a certain time is greater than the frequency threshold.

[0048] The result of load data difference analysis is: the information that characterizes the deficiencies of the power load forecasting model obtained after the difference analysis is completed. Based on the result of load data difference analysis, the model adjustment strategy of the power load forecasting model is determined and adjusted accordingly. Among them, the model adjustment strategy is: a plan for optimizing and adjusting the power load forecasting model based on the result of load data difference analysis, such as: what model parameters to modify, what improved algorithm to introduce, etc.

[0049] When the power load forecasting model is used for power equipment forecasting, its accuracy will change as the forecasting environment changes (for example, the power load is affected by natural factors (indoor and outdoor temperature, humidity, wind speed, etc.)). Therefore, the present invention compares the actual load data and the predicted load data to obtain the load data difference; since the power load data is prone to mutation, once the difference is found, it is analyzed, which occupies computing resources. Therefore, before performing the difference analysis, it is determined whether the load data difference meets the analysis conditions. If it meets the conditions, the difference analysis is performed to obtain the load data difference analysis results; based on the load data difference analysis results, the model adjustment strategy of the power load forecasting model is determined and adjusted accordingly to improve the rationality and timeliness of the adjustment of the power load forecasting model.

[0050] In some other cases, step S6 further includes: Determine a first load data curve corresponding to the real-time load data of the electric power equipment, and determine a second load data curve corresponding to the predicted load data of the electric power equipment; Temporally aligning the first load data curve and the second load data curve to obtain an alignment result; Based on the analysis condition trigger determination template and the alignment result, it is determined whether the alignment result satisfies any analysis condition trigger condition in the analysis condition trigger determination template. If so, it is determined that the load data difference reaches the preset analysis condition.

[0051] The first load data curve is a curve drawn based on the actual measured power load data, such as the actual load change curve within a certain time period; the second load data curve is a curve drawn based on the predicted load data within a certain time period calculated by the load prediction model. The trigger condition for the analysis condition is that the image describing the time-aligned first load data curve and the second load data curve meets the image description feature of the analysis condition, such as 10 consecutive first load data curve points are higher than the second load data curve points.

[0052] Align the first load data curve of the actual load data and the second load data curve of the predicted load data, and judge whether the alignment result meets the analysis condition triggering condition according to the introduced analysis condition triggering template and alignment result. When it meets the condition, judge to conduct subsequent data analysis, which improves the accuracy of load data difference analysis triggering and improves the utilization rate of analysis computing resources. Judge whether the load data difference meets the preset analysis condition according to the alignment result, so as to ensure that the comparison is carried out at the same time, avoid the difference judgment error caused by time deviation, and further improve the accuracy of model correction.

[0053] The analysis condition trigger determination template stores multiple analysis condition trigger conditions. In this case, step S6 also includes: Obtain historical warning omission records; Perform warning omission attribution on historical warning omission records, determine the historical warning omission records whose attribution results are the warning omissions caused by the prediction differences at the time of historical warning omission, and use them as target warning omission records; Determine the historical warning omission time corresponding to the target warning omission record and use it as the backward time; Obtaining curve difference characteristics of a historical actual load data curve and a historical forecast load data curve of a preset target time length before the backward time; Calculating feature similarity values ​​between curve difference features, and determining mutually different target curve difference features according to the calculation results of the feature similarity values; The target curve difference characteristics are used as the analysis condition trigger conditions, and are summarized to form an analysis condition trigger judgment template.

[0054] In this embodiment, the historical warning omission record is: the record of abnormal electric load data that the warning system failed to accurately predict in the past power load forecast. The preset target time length is preset manually, for example: 15 minutes; the curve difference feature is: at which time point there is a difference between the historical actual load data curve and the historical forecast load data curve, how much larger or smaller. The feature similarity value is: the quantitative value of the similarity between the extracted curve difference features corresponding to different target warning omission records; the different determination conditions are: the feature similarity value is less than 95.

[0055] The historical warning omission records obtained are attributed to the warning omissions, and the attribution results are determined to be the target warning omission records caused by the warning difference (for example: the actual power load of device A is lower than the standard amount, and the predicted power load of device A meets the standard amount, then the warning will not be triggered before the power load of device A is actually measured). According to the historical warning omission time (backward time) recorded in the target warning omission record, the curve difference characteristics of the historical actual load data curve and the historical predicted load data curve are obtained; the feature similarity value between the curve difference characteristics is calculated, and according to the calculation results of the feature similarity value, the target curve difference characteristics that are different from each other are determined and summarized to form an analysis condition trigger judgment template, and the analysis condition trigger conditions are extracted from the records that cause the warning omissions, thereby improving the suitability of the acquisition of the analysis condition trigger conditions. The curve difference characteristics of the historical actual load data curve and the historical predicted load data curve are obtained through the historical warning omission records, and the target curve difference characteristics are used as the analysis condition trigger condition, so that the analysis condition trigger condition is based on the actual data curve and the predicted data curve with missed warnings, so that the final curve difference (alignment result) matches the curve difference (analysis condition trigger condition), which is more accurate and further improves the accuracy of model correction.

[0056] Based on the analysis condition triggering judgment template and the alignment result, judging whether the alignment result satisfies any one of the analysis condition triggering judgment templates includes: Parse the analysis condition trigger judgment template to obtain the search window of the analysis condition trigger condition corresponding to the alignment result; Slide the analysis condition trigger condition in the corresponding search window to obtain the curve difference feature similarity between the sliding analysis condition trigger condition and the alignment result of the sliding position; If the curve difference feature similarity is greater than or equal to a preset curve difference feature similarity threshold, it is determined that the alignment result of the corresponding sliding position meets the sliding analysis condition trigger condition; The step of sliding the analysis condition trigger condition in the corresponding search window is as follows: sliding the analysis condition trigger condition from the middle position of the corresponding search window to both sides of the search window respectively.

[0057] The search window is: the judgment range of whether the analysis condition trigger condition in the alignment result meets the corresponding trigger condition. The curve difference feature similarity is: the value obtained by comparing the target curve difference feature corresponding to the analysis condition trigger condition and the curve difference feature of the alignment result of the sliding position to indicate the similarity between the two. The preset curve difference feature similarity threshold is manually set in advance.

[0058] In this embodiment, the analysis condition trigger condition is located in a judgment range (search window) corresponding to the alignment result for judging whether the corresponding trigger condition is met; the analysis condition trigger condition is slid from the middle position of the corresponding search window to both sides of the search window respectively, so that the matching efficiency of the analysis condition trigger condition can be improved, and the curve difference feature similarity of the sliding analysis condition trigger condition and the alignment result of the sliding position is obtained. When the curve difference feature similarity is greater than or equal to the preset curve difference feature similarity threshold, it means that the alignment result of the corresponding sliding position meets the sliding analysis condition trigger condition, and the difference analysis can be performed based on the load data after the alignment result of the sliding position, thereby improving the analysis efficiency of the load difference analysis.

[0059] like Figure 2 As shown, as a fourth embodiment of the present application, a power load forecasting system includes: The power equipment data reading subsystem 1 is used to read the power equipment operation data; The target load subject research record screening subsystem 2 is used to screen the target load subject research records of the research load subject according to the historical operation data of the power equipment; The power load prediction model training subsystem 3 is used to train the power load prediction model according to the target load subject research record; the future power equipment load data prediction subsystem 4 is used to predict the future power equipment load data based on the power load prediction model and the real-time operation data of the power equipment.

[0060] The power equipment data reading subsystem 1 collects and integrates real-time data from various power equipment, including but not limited to current, voltage, power factor, equipment temperature, operating status, etc., to obtain power equipment operating data, which includes real-time power equipment operating data and historical power equipment operating data.

[0061] The target load subject research record screening subsystem 2 screens the target load subject research records in the research data according to the load subject industry in the historical operation data of the power equipment through industry matching, reduces data noise, and improves the pertinence and accuracy of the analysis.

[0062] The power load forecasting model training subsystem 3 obtains the power load forecasting model through machine learning training based on the research records of the screened target load subject to capture the law of load changes.

[0063] The future power equipment load data prediction subsystem 4 uses the trained power load prediction model and the latest power equipment operation data to predict the power load situation in the future in real time or regularly, so as to facilitate the optimal allocation of power resources, improve resource utilization, and reduce the risks caused by the imbalance between supply and demand.

[0064] The specific implementation described above is a preferred implementation of an electric power load forecasting method and system of the present application, and is not intended to limit the specific implementation scope of the present application. The scope of the present application includes but is not limited to the specific implementation embodiment. All equivalent changes made in accordance with the shape and structure of the present application are within the protection scope of the present application.

Claims

1. A method for predicting power load, characterized in that: include: S1: Acquire historical operation data of electric power equipment, and acquire first application industry information based on the historical operation data of electric power equipment; S2: obtaining a pre-screening load subject research record, and obtaining second application industry information based on the pre-screening load subject research record; S3: Acquire industry-related values ​​based on the first application industry information and the second application industry information; S4: Screening the pre-screened load subject research records based on industry-related values ​​and preset optimization thresholds to obtain target load subject research records; S5: Construct a power load forecasting model based on the target load subject research records, and output power equipment load forecasting data based on the power load forecasting model and the real-time operation data of power equipment.

2. A method for predicting power load according to claim 1, characterized in that: The industry-related values ​​are: Industry information similarity value between the first application industry information and the second application industry information.

3. A method for predicting power load according to claim 1, characterized in that: The step S1 also includes: obtaining a predicted rhythm based on historical operation data of the power equipment; The S2 also includes: obtaining a study rhythm based on the pre-screening load subject study record; The S3 also includes: obtaining a rhythm multiplexing value based on the predicted rhythm and the studied rhythm; The S4 also includes: screening the pre-screening load subject research records based on industry-related values, rhythm multiplexing values ​​and preset preferred thresholds.

4. A method for predicting power load according to claim 3, characterized in that: The S4 further comprises: Obtaining the preferred value of the pre-screening load subject research record according to the industry-related value and rhythm reuse value corresponding to the pre-screening load subject research record; The pre-screening load subject research record whose preferred value is greater than or equal to the preset preferred threshold is used as the target load subject research record.

5. A method for predicting power load according to claim 3, characterized in that: The obtaining of rhythm multiplexing values ​​based on predicted rhythms and studied rhythms also includes: Obtain the rhythm inclusion relationship between the studied rhythm and the predicted rhythm; If the rhythm inclusion relationship is that the research rhythm includes the predicted rhythm, the research information item is expanded on the time axis corresponding to the research rhythm; the information attribute of the research information item is matched with the credible information attribute to obtain the credible value of the research information item; Determine the time interval of the research rhythm corresponding to the time axis; Obtaining a first average of the credibility values ​​of the research information items falling within the time interval; Obtaining a second average value of the credibility values ​​of the research information items before the time interval on the time axis; Assigning a first weight to the first average value, and assigning a second weight to the second average value, wherein the first weight is greater than the second weight; Multiply the first average value and the first weight correspondingly to obtain a first target value; Multiply the second average value and the second weight correspondingly to obtain a second target value; The first target value and the second target value corresponding to the studied rhythm are summed to obtain the rhythm multiplexing value.

6. A method for predicting power load according to claim 1, characterized in that: Also includes: S6: Acquire real-time load data of the power equipment, compare the real-time load data of the power equipment with the predicted load data of the power equipment, obtain the load data difference, and if the load data difference reaches a preset analysis condition, perform difference analysis based on the load data difference and obtain the load data difference analysis result; S7: Based on the load data difference analysis results, determine the model adjustment strategy of the power load forecasting model and perform model adjustment according to the model adjustment strategy.

7. A method for predicting power load according to claim 6, characterized in that: The S6 further includes: Determine a first load data curve corresponding to the real-time load data of the electric power equipment, and determine a second load data curve corresponding to the predicted load data of the electric power equipment; Temporally aligning the first load data curve and the second load data curve to obtain an alignment result; Based on the analysis condition trigger determination template and the alignment result, it is determined whether the alignment result satisfies any analysis condition trigger condition in the analysis condition trigger determination template. If so, it is determined that the load data difference reaches the preset analysis condition.

8. A method for predicting power load according to claim 7, characterized in that: The S6 further includes: Obtain historical warning omission records; Perform warning omission attribution on historical warning omission records, determine the historical warning omission records whose attribution results are the warning omissions caused by the prediction differences at the time of historical warning omission, and use them as target warning omission records; Determine the historical warning omission time corresponding to the target warning omission record and use it as the backward time; Obtaining curve difference characteristics of a historical actual load data curve and a historical forecast load data curve of a preset target time length before the backward time; Calculating feature similarity values ​​between curve difference features, and determining mutually different target curve difference features according to the calculation results of the feature similarity values; The target curve difference characteristics are used as the analysis condition trigger conditions, and are summarized to form an analysis condition trigger judgment template.

9. A method for predicting power load according to claim 6, characterized in that: The step of judging whether the alignment result satisfies any one of the analysis condition triggering conditions in the analysis condition triggering judgment template based on the analysis condition triggering judgment template and the alignment result includes: Parse the analysis condition trigger judgment template to obtain the search window of the analysis condition trigger condition corresponding to the alignment result; Slide the analysis condition trigger condition in the corresponding search window to obtain the curve difference feature similarity between the sliding analysis condition trigger condition and the alignment result of the sliding position; If the curve difference feature similarity is greater than or equal to a preset curve difference feature similarity threshold, it is determined that the alignment result of the corresponding sliding position meets the sliding analysis condition trigger condition; The step of sliding the analysis condition trigger condition in the corresponding search window is as follows: sliding the analysis condition trigger condition from the middle position of the corresponding search window to both sides of the search window respectively.

10. A power load forecasting system, used to implement the method according to any one of claims 1 to 9, characterized in that: include: The power equipment data reading subsystem is used to read the power equipment operation data; The target load subject research record screening subsystem is used to screen the target load subject research records of the research load subject according to the historical operation data of the power equipment; The power load forecasting model training subsystem is used to train the power load forecasting model according to the target load subject research records; the future power equipment load data prediction subsystem is used to predict the future power equipment load data based on the power load forecasting model and the real-time operation data of the power equipment.

Images