Electric power material prediction method and system based on bidirectional difference, medium and processor
Through the combination of two-way differential calculation and XGBoost model, the problem of poor prediction results in medium and long-term prediction of power material prediction is solved, accurate prediction and resource conservation are achieved, and the safety and economicality of the power grid are ensured.
Patent Information
- Application Number
- CN202411214069.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-31
- Publication Date
- 2024-12-17
AI Technical Summary
In the prior art, the long-term prediction effect of power material prediction is poor and the computing resources consumes a lot, making it difficult to effectively capture the long-term dependencies in time series data.
The two-way differential calculation method is adopted to achieve accurate long-term prediction through differential preprocessing in annual and monthly directions, combined with the XGBoost model, long-term dependencies in the data are mined, and the application of large models is reduced, and accurate long-term prediction is achieved.
Accurate long-term power consumption predictions are achieved, computing resources are saved, prediction accuracy is improved, manual reporting is reduced, and the safety and economicality of the power grid is guaranteed.
Smart Images
Figure CN119151059A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of electric power material prediction, and in particular to an electric power material prediction method, system, medium and processor based on bidirectional difference. Background Art
[0002] my country's power grid is huge in scale, providing a solid foundation for all aspects of national economy and people's livelihood. However, the huge scale also brings huge challenges to enterprise operations, among which the prediction of material consumption is a basic issue. On the one hand, a low forecast value will lead to shortages and supply interruptions, and the power grid cannot operate normally; on the other hand, when the forecast value is too high, it will cause inventory backlogs, restricting the capital utilization efficiency of enterprises. Therefore, accurately predicting the consumption of power materials is of great significance to ensure the safety and economy of power grid production.
[0003] The consumption of electric power materials is generally counted in fixed time periods such as months and quarters. The statistical data constitutes time series data. Therefore, the time series data prediction method can be used to explore the internal laws of historical data samples and predict future data based on the factors affecting the data. At present, there are mainly statistical methods, machine learning methods and deep learning methods for time series data prediction. (1) The representative of statistical methods is ARIMA. The ARIMA model is an improved version of ARMA. It converts non-stationary time series into stationary time series by difference and then models them. It has good linear fitting ability, but poor long-term prediction effect. (2) In machine learning methods, XGBoost is an efficient and flexible boosting algorithm. It combines multiple weak classifiers into a strong classifier, but it is difficult to capture long-term dependencies in time series data. (3) For deep learning methods, LSTM improves traditional RNN through memory units and gating mechanisms. It can effectively capture long-term dependencies and solve the problems of gradient vanishing and gradient explosion. However, it requires a large amount of data to drive the model, and there are also problems of complex models and long training time. Therefore, in general, the current existing technologies have poor long-term prediction effects and require large computing resources.
[0004] In view of this, a method, system, medium and processor for predicting electric power materials based on bidirectional differential are needed. Summary of the invention
[0005] In view of the poor effect of long-term prediction in the prior art and the large amount of computing resources required, the present invention provides a method, system, medium and processor for predicting power materials based on bidirectional difference. Through the calculation method of bidirectional difference, the long-term dependency relationship in the data is mined to achieve accurate long-term prediction effect, while reducing the use of large models and saving the required computing resources. The specific technical solution is as follows:
[0006] A method for predicting electric power materials based on bidirectional difference, comprising:
[0007] Collect Y years and 12Y historical data in total;
[0008] Preprocess historical data to eliminate the influence of outliers and noise;
[0009] Based on the annual periodicity of power materials and the continuity of adjacent months, the historical data are differentially calculated in the annual and monthly directions;
[0010] Carry out annual residual value forecast and monthly residual value forecast respectively;
[0011] The benchmark value of the month to be tested is calculated by taking the average value of the corresponding month in the recording year of the month to be tested;
[0012] Reconstruct based on the baseline value, annual residual value and monthly residual value to obtain the final material consumption forecast value.
[0013] Furthermore, performing differential calculation on the historical data in the monthly direction includes the following steps:
[0014] To calculate monthly differences, the formula is as follows:
[0015]
[0016] Get the monthly difference set, the formula is as follows:
[0017]
[0018] In the above formula, k represents the sample number; x′ k represents the kth monthly historical data; represents the k-th monthly difference value; D m is a collection of monthly differences.
[0019] Furthermore, the monthly residual value prediction includes the following steps:
[0020] D m The monthly differential data with indices from in to i-1 and the index i of the month to be predicted constitute the feature vector C i,n , the formula is as follows:
[0021]
[0022] The sliding window method is used to construct the feature vector C i,n As samples, a training set is constructed, where the number of samples in the training set is N-1-n;
[0023] Build an XGBoost model and train it using the training set;
[0024] Using the samples required for the latest month, the trained XGboost model is used to predict the latest monthly residual value.
[0025] Furthermore, the difference calculation of the historical data in the annual direction includes the following steps:
[0026] Calculate the annual difference vector for the kth month
[0027]
[0028] Construct annual difference data set D y :
[0029]
[0030] In the above formula, k is an integer from 1 to 12; x′ k Represents the historical data of the kth month.
[0031] Furthermore, the annual residual value forecast includes the following steps:
[0032] The mean of the annual difference vector is taken as the annual residual, and the formula is as follows:
[0033]
[0034] r k y is the annual residual of month k, and is the annual difference vector The mean of .
[0035] Furthermore, the reference value of the month to be tested is calculated based on the average of the corresponding months of the month to be tested in the recording year, and the formula is as follows:
[0036]
[0037] In the above formula, r i A is the benchmark value of month i to be tested, x′ i-12 This is historical data for Article i-12.
[0038] Furthermore, the final material consumption forecast value is obtained by reconstructing the base value, annual residual value and monthly residual value, and the formula is as follows:
[0039]
[0040] In the above formula, is the final predicted value of material consumption; r m is the monthly residual forecast value of the month to be tested; r yis the annual residual forecast value of the month to be tested; r A is the reference value of the month to be tested; 1 , 2 , 2 are learnable weight parameters.
[0041] A power material forecasting system based on bidirectional difference, applied to the above-mentioned power material forecasting method based on bidirectional difference, comprises:
[0042] The collection module is used to collect Y years and a total of 12Y historical data;
[0043] A preprocessing module is used to preprocess historical data to eliminate the influence of outliers and noise;
[0044] The difference module is used to perform difference calculations on historical data in the annual and monthly directions based on the annual periodicity of power materials and the continuity of adjacent months;
[0045] A prediction module, which is used to predict annual residual values and monthly residual values respectively;
[0046] A reference value module, which is used to calculate the reference value of the month to be tested based on the average of the corresponding months of the month to be tested in the recording year;
[0047] The reconstruction module is used to reconstruct according to the baseline value, annual residual value and monthly residual value to obtain the final material consumption forecast value.
[0048] A computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the above-mentioned method for predicting electric power materials based on bidirectional difference.
[0049] A processor is used to run a program, wherein the program executes the above-mentioned power material prediction method based on bidirectional difference when running.
[0050] Compared with the prior art, the present invention has the following beneficial effects:
[0051] 1. Through the calculation method of bidirectional difference, the long-term dependencies in the data are mined to achieve accurate long-term prediction effects, while reducing the use of large models and saving the required computing resources.
[0052] 2. Through mining and analyzing the historical data of power material consumption, the power material consumption in the future can be predicted. This aims to solve the inaccuracy of manual reporting of power material demand in current power companies and ensure the safety and economy of power grid production. BRIEF DESCRIPTION OF THE DRAWINGS
[0053] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for the specific embodiments or the description of the prior art. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn according to the actual scale.
[0054] Figure 1 This is a flow chart of the first method for predicting power materials based on bidirectional difference;
[0055] Figure 2 This is a schematic diagram of the monthly difference set structure;
[0056] Figure 3 This is a schematic diagram of the annual difference set structure. DETAILED DESCRIPTION
[0057] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0058] It should be understood that when used in this specification and the appended claims, the terms "include" and "comprises" indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.
[0059] It should also be understood that the terms used in the present specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, the singular forms "a", "an" and "the" are intended to include plural forms unless the context clearly indicates otherwise.
[0060] It should be further understood that the term "and / or" used in the present description and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.
[0061] Embodiment 1
[0062] Based on the analysis of power material consumption, this embodiment found that although power material consumption is highly sudden, it still shows certain rules. The first is monthly continuity: affected by continuous changes in temperature, there is a certain continuity in the consumption of adjacent months. The second is annual similarity: the same month in different years shows similarities. For example, February is close to the Spring Festival, and material consumption is low; in the middle of the year, due to the promotion of various marketing projects, the consumption of power equipment is relatively large. Based on this, the embodiment proposes the following Figure 1 The small sample power material forecasting method based on two-way difference is shown. This method extracts annual differences and monthly differences from a small number of historical samples, uses a linear model to predict the benchmark value of the future year, and then uses the linear model and XGBoost model to predict the residual value between the future sample and the benchmark value. Finally, the benchmark value, annual residual value, and monthly residual value are comprehensively considered to reconstruct the accurate consumption value.
[0063] like Figure 1 The figure shows a flow chart of a method for predicting power materials based on bidirectional difference, including:
[0064] S1: Collect Y years and 12Y historical data in total.
[0065] A total of N=12Y pieces of historical data.
[0066] S2: The consumption of power material data varies between months and is highly volatile. In order to reduce the impact of the above negative factors on the model performance, the historical data is preprocessed to eliminate the influence of outliers and noise.
[0067] Furthermore, the preprocessing includes outlier detection correction and data normalization, and x′ is assumed to be the historical data after outlier correction and normalization.
[0068] After this step, the algorithm is divided into three branches, which respectively perform baseline value, annual residual, and monthly residual forecast calculations.
[0069] S3: Based on the annual periodicity of power materials and the continuity of adjacent months, the historical data are differentially calculated in the annual and monthly directions respectively.
[0070] Further, such as Figure 2 As shown, the historical data is differentially calculated in the monthly direction, D m The monthly difference data is a set consisting of N historical data, and N-1 monthly difference data are generated from N historical data, including the following steps:
[0071] To calculate monthly differences, the formula is as follows:
[0072]
[0073] Get the monthly difference set, the formula is as follows:
[0074]
[0075] In the above formula, k represents the sample number; x′ k represents the kth monthly historical data; represents the k-th monthly difference value; D m is a set of monthly differences, the kth sample is the normalized data x′ k+1 and x′ k The difference; m represents the month, the abbreviation of month. y The superscript y is the abbreviation of year.
[0076] Further, such as Figure 3 As shown, the historical data is differentially calculated in the annual direction. The annual difference is the difference between the data of the same month in different years. For example, between 2024 and 2023, January is subtracted, February is subtracted, and so on until December. Specifically, the following steps are included:
[0077] Calculate the annual difference vector for the kth month
[0078] The subscript change step is 12, corresponding to the same month in different years;
[0079] Construct annual difference data set D y :
[0080]
[0081] In the above formula, k is an integer from 1 to 12, k is the kth month, is a vector; x′ k represents the historical data of the kth month, D y The annual difference vectors are composed of 12 independent months. 1≤k≤12.
[0082] S4: Carry out annual residual value forecast and monthly residual value forecast respectively.
[0083] Furthermore, the monthly difference data D m The length of is N-1, which can support a higher level of residual prediction algorithm. The embodiment adopts the XGBoost-based method for prediction. The monthly residual value prediction includes the following steps:
[0084] D mThe n monthly differential data with indices from in to i-1 and the index i of the month to be predicted constitute the feature vector C i,n , predict the monthly residual for the latest month For labels, the formula is as follows:
[0085]
[0086] The sliding window method is used to construct the feature vector C i,n As samples, a training set is constructed. The number of samples in the training set is N-1-n. For example, assuming that N=50 and n=12, the training set is constructed as follows: 1 m arrive Constitute a training sample, As a label. arrive Constitute the second training sample, is the label. And so on, the last one is arrive Constitute the training samples, is the label, and a total of N-1-n "sample-label" training sets are used for training.
[0087] This embodiment adopts the XGBoost model to capture the monthly dependencies, builds the XGBoost model and trains it using the training set;
[0088] Using the samples required for the latest month, the trained XGboost model is used to predict the latest monthly residual value.
[0089] That is, from D m Extract n historical differential data to predict the next residual data. Taking the above example, when N=50 and n=12, the training process can be expressed as:
[0090]
[0091] The prediction process can be expressed as
[0092] Furthermore, the annual residual value forecast includes the following steps:
[0093] Due to the limitation of the amount of historical data, the historical differential data in the annual direction is relatively small and complex models cannot be used for prediction. Therefore, this embodiment adopts an intuitive and efficient solution, taking the mean of the historical annual differential data as the annual residual, that is, taking the mean of the annual differential vector as the annual residual, and the formula is as follows:
[0094]
[0095] r k y is the annual residual of month k, and is the annual difference vector The mean of .
[0096] There are 12 months in a year, and the kth month to be predicted (k is an integer between 1 and 12). For example, to predict March 2024, the annual residual is only the same as the vector The mean of the vector elements is r. 3 y .
[0097] S5: Calculate the benchmark value of the month to be tested based on the average of the corresponding months of the month to be tested in the recording year.
[0098] Furthermore, the reference value of the month to be tested is calculated based on the average of the corresponding months of the month to be tested in the recording year, and the formula is as follows:
[0099]
[0100] In the above formula, r i A is the benchmark value of month i to be tested, x′ i-12 This is historical data for Article i-12.
[0101] For example, if we have historical data from 2020 to 2023 and want to predict September 2024, we can take the average of the data from September 2020 to 2023 as r. 9 A .
[0102] S6: Reconstruct based on the baseline value, annual residual value and monthly residual value to obtain the final material consumption forecast value.
[0103] Furthermore, the final material consumption forecast value is obtained by reconstructing the base value, annual residual value and monthly residual value, and the formula is as follows:
[0104]
[0105] In the above formula, is the final predicted value of material consumption; r m is the monthly residual forecast value of the month to be tested; r y is the annual residual forecast value of the month to be tested; r A is the reference value of the month to be tested; 1 , 2 , 2 is a learnable weight parameter, which is essentially adjusting λ 1 ,λ 2 ,λ3 The value of lambda is chosen so that the value of \hat{y} is closest to the true value. In actual operation, grid search and other methods can be used to find the most appropriate lambda value.
[0106] Through the calculation method of two-way difference, the long-term dependency in the data is mined to achieve accurate long-term prediction effect, while reducing the use of large models and saving the required computing resources. By mining and analyzing the historical data of power material consumption, the power material consumption in the future is predicted. It aims to solve the inaccuracy of manual reporting of power material demand in current power companies and ensure the safety and economy of power grid production.
[0107] Embodiment 2
[0108] A power material forecasting system based on bidirectional difference, applied to the above-mentioned power material forecasting method based on bidirectional difference, comprises:
[0109] The collection module is used to collect Y years and a total of 12Y historical data;
[0110] A preprocessing module is used to preprocess historical data to eliminate the influence of outliers and noise;
[0111] The difference module is used to perform difference calculations on historical data in the annual and monthly directions based on the annual periodicity of power materials and the continuity of adjacent months;
[0112] A prediction module, which is used to predict annual residual values and monthly residual values respectively;
[0113] A reference value module, which is used to calculate the reference value of the month to be tested based on the average of the corresponding months of the month to be tested in the recording year;
[0114] The reconstruction module is used to reconstruct according to the baseline value, annual residual value and monthly residual value to obtain the final material consumption forecast value.
[0115] Embodiment 3
[0116] A computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the above-mentioned method for predicting electric power materials based on bidirectional difference.
[0117] Embodiment 4
[0118] A processor is used to run a program, wherein the program executes the above-mentioned power material prediction method based on bidirectional difference when running.
[0119] The present application provides a method for forecasting electric power materials based on bidirectional difference, including: S1: collecting Y years and a total of 12Y historical data; S2: pre-processing the historical data to eliminate the influence of outliers and noise; S3: based on the annual periodicity of electric power materials and the continuity of adjacent months, performing differential calculations on the historical data in the annual and monthly directions; S4: performing annual residual value prediction and monthly residual value prediction respectively; S5: calculating the benchmark value of the month to be tested based on the mean of the corresponding month of the month to be tested in the recorded year; S6: reconstructing according to the benchmark value, annual residual value and monthly residual value to obtain the final material consumption forecast value. Through the calculation method of bidirectional difference, the long-term dependency relationship in the data is mined to achieve accurate long-term forecasting effect, while reducing the use of large models and saving the required computing resources. By mining and analyzing the historical data of electric power material consumption, the prediction of electric power material consumption in the future is realized, the inaccuracy of manual reporting of electric power material demand in current power enterprises is solved, and the safety and economy of power grid production are ensured.
[0120] Those of ordinary skill in the art will appreciate that the units of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition of each example has been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.
[0121] In the embodiments provided by the present invention, it should be understood that the division of units is only a logical function division, and there may be other division methods in actual implementation, for example, multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored, etc.
[0122] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.
[0123] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-0nlyMemory), random access memory (RAM, RandomAccessMemory), mobile hard disk, magnetic disk or optical disk, etc., which can store program code.
[0124] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and specification of the present invention.
Claims
1. A method for predicting power materials based on bidirectional difference, characterized in that: include: Collect Y years and 12Y historical data in total; Preprocess historical data to eliminate the influence of outliers and noise; Based on the annual periodicity of power materials and the continuity of adjacent months, the historical data are differentially calculated in the annual and monthly directions; Carry out annual residual value forecast and monthly residual value forecast respectively; The benchmark value of the month to be tested is calculated by taking the average value of the corresponding month in the recording year of the month to be tested; Reconstruct the final material consumption forecast based on the baseline value, annual residual value and monthly residual value.
2. The power material prediction method based on bidirectional difference according to claim 1 is characterized in that: The monthly differential calculation of historical data includes the following steps: To calculate monthly differences, the formula is as follows: Get the monthly difference set, the formula is as follows: In the above formula, k represents the sample number; x′ k represents the kth monthly historical data; represents the k-th monthly difference value; D m is a collection of monthly differences.
3. The power material prediction method based on bidirectional difference according to claim 2 is characterized in that: The monthly residual value prediction includes the following steps: D m The monthly differential data with indices from in to i-1 and the index i of the month to be predicted constitute the feature vector C i,n , the formula is as follows: The sliding window method is used to construct the feature vector C i,n As samples, a training set is constructed, where the number of samples in the training set is N-1-n; Build an XGBoost model and train it using the training set; Using the samples required for the latest month, the trained XGboost model is used to predict the latest monthly residual value.
4. The power material prediction method based on bidirectional difference according to claim 1 is characterized in that: The difference calculation of the historical data in the annual direction includes the following steps: Calculate the annual difference vector for the kth month Construct annual difference data set D y : In the above formula, k is an integer from 1 to 12; x′ k Represents the historical data of the kth month.
5. The power material prediction method based on bidirectional difference according to claim 4 is characterized in that: The annual residual value prediction includes the following steps: The mean of the annual difference vector is taken as the annual residual, and the formula is as follows: r k y =mean{d k y}; r k y is the annual residual of month k, and is the annual difference vector The mean of .
6. The power material prediction method based on bidirectional difference according to claim 1 is characterized in that: The base value of the month to be tested is calculated by taking the average of the corresponding months of the month to be tested in the recording year, and the formula is as follows: In the above formula, r i A is the benchmark value of month i to be tested, x i ' -12 For the i-12 Historical data.
7. The power material prediction method based on bidirectional difference according to claim 1 is characterized in that: The final material consumption forecast value is obtained by reconstructing the base value, annual residual value and monthly residual value. The formula is as follows: In the above formula, is the final predicted value of material consumption; r m is the monthly residual forecast value of the month to be tested; r y is the annual residual forecast value of the month to be tested; r A is the benchmark value of the month to be tested; λ1, λ2, λ2 are learnable weight parameters.
8. A power material forecasting system based on bidirectional difference, characterized in that: The power material prediction method based on bidirectional difference applied to any one of claims 1 to 7 comprises: The collection module is used to collect Y years and a total of 12Y historical data; A preprocessing module is used to preprocess historical data to eliminate the influence of outliers and noise; The difference module is used to perform difference calculations on historical data in the annual and monthly directions based on the annual periodicity of power materials and the continuity of adjacent months; A prediction module, which is used to predict annual residual values and monthly residual values respectively; A reference value module, which is used to calculate the reference value of the month to be tested based on the average of the corresponding months of the month to be tested in the recording year; The reconstruction module is used to reconstruct according to the baseline value, annual residual value and monthly residual value to obtain the final material consumption forecast value.
9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the power material prediction method based on bidirectional difference as described in any one of claims 1 to 7.
10. A processor, characterized in that: The processor is used to run a program, wherein the program, when running, executes the power material prediction method based on bidirectional difference as described in any one of claims 1 to 7.