A baseline load prediction method, device, electronic equipment and storage medium

Abstract

The application provides a baseline load prediction method and device, electronic equipment and storage medium, wherein the method comprises: obtaining historical load data, historical day data and prediction day data determined by indexes; using an improved principal component analysis method based on an entropy value method to determine principal components in the indexes and corresponding principal component comprehensive scores for a matrix composed of the historical day data; calculating meteorological similarities of the prediction day and the historical day according to the historical day data and the prediction day data corresponding to the indexes divided according to the principal component comprehensive scores; determining a historical day set similar to the prediction day according to the meteorological similarities and time similarities of the prediction day and the historical day; and obtaining prediction day load data according to the historical load data, the historical day set, a meteorological matrix and the prediction day data by using a back propagation neural network. Through the application, the problem of low data quality caused by data preprocessing before baseline load prediction in the related art is solved.

Description

Technical Field

[0001] This invention relates to the field of data analysis, and more particularly to a baseline load forecasting method, apparatus, electronic device, and storage medium. Background Technology

[0002] In recent years, the number of newly built public buildings in China has continued to grow, and their energy consumption accounts for 30% to 40% of the total energy consumption of social production and daily life, with peak loads in summer and winter even exceeding 50%. Optimized operation and management related to the load side is one of the key technologies for achieving energy conservation and emission reduction in the power industry. By mobilizing the load resources of public buildings on the demand side to actively participate in grid peak shaving and frequency regulation, the contradiction between power supply and demand can be effectively alleviated and energy conservation and emission reduction can be promoted.

[0003] Traditional baseline load forecasting methods, such as classifying users by type and location, and then establishing regression equations based on the electricity load data of individual users within each category and influencing factors like local temperature and humidity, to calculate regression parameters, have limited accuracy in estimating baseline load. Data preprocessing methods are relatively simple, and when historical load data quality is low due to incompleteness or errors, the preprocessing methods used, such as principal component analysis, suffer from limitations such as inaccurate calculation results and ambiguity in the meaning of principal components due to dimensionality reduction, affecting the accuracy of load forecasting. Therefore, existing technologies suffer from the problem of low data quality obtained from data preprocessing before baseline load forecasting. Summary of the Invention

[0004] The present invention provides a baseline load forecasting method, apparatus, electronic device and storage medium to at least solve the problem of low data quality obtained from data preprocessing before baseline load forecasting in related technologies.

[0005] According to a first aspect of the present invention, a baseline load forecasting method is provided. The method includes: acquiring historical load data and historical daily data and forecast daily data determined by indicators, wherein the indicators include temperature, humidity, air pressure, precipitation, irradiance, and wind speed; using a principal component analysis method based on entropy-based improvement to determine the principal components in the indicators and the corresponding principal component comprehensive scores of the matrix composed of the historical daily data; calculating the meteorological similarity between the forecast day and the historical days based on the historical daily data and forecast daily data corresponding to the indicators after being divided by the principal component comprehensive scores; determining a set of historical days similar to the forecast day based on the meteorological similarity and the temporal similarity between the forecast day and the historical days; and using a backpropagation neural network to obtain the forecast daily load data based on the historical load data, the set of historical days, the meteorological matrix, and the forecast daily data, wherein the meteorological matrix is ​​composed of the historical daily data and forecast daily data corresponding to the divided indicators.

[0006] Optionally, before determining the principal components and corresponding principal component comprehensive scores in the index using a principal component analysis method improved based on entropy value on the matrix composed of the historical daily data, the method further includes: generating a historical daily data matrix based on the historical daily data, wherein the row vectors of the historical daily data matrix represent a sample, the sample including the historical daily data corresponding to the index at the target time, and the column vectors of the historical daily data matrix including the historical daily data of an index at all times; standardizing the historical daily data matrix; pre-impacting missing data based on the average value of the historical daily data in the column vectors of the standardized matrix; determining the samples in the pre-impacted matrix most relevant to the samples containing missing data based on the grey relational algorithm; and calculating the imputation values ​​of the missing data based on the samples in the matrix using the KNN algorithm and the DBSCAN algorithm.

[0007] Optionally, the step of using the KNN and DBSCAN algorithms to calculate the imputation value of missing data based on the samples in the matrix includes: determining the neighborhood radius and threshold of the DBSCAN algorithm based on the data of the samples in the pre-imputation matrix that are most relevant to the samples containing missing data; determining outliers in the data based on the neighborhood radius and threshold; and using the KNN algorithm to calculate the imputation value of missing data on the data in the samples after removing outliers.

[0008] Optionally, the step of determining the principal components and corresponding principal component comprehensive scores in the index using a principal component analysis method improved based on the entropy method on the matrix composed of the historical day data includes: generating a complete historical day data matrix based on the imputed historical day data; normalizing the complete historical day data matrix based on the mean of the column vectors in the complete historical day data matrix; calculating the covariance matrix of the normalized matrix and the eigenvalues ​​and eigenvectors of the covariance matrix; determining the principal components in the index based on the cumulative variance contribution rate of the eigenvalues; calculating the principal component scores based on the eigenvectors and the normalized matrix; and determining the principal component comprehensive score based on the weighted sum of the principal component entropy values ​​and the principal component sample weights, wherein the principal component entropy values ​​and the principal component sample weights are determined based on the principal component scores.

[0009] Optionally, determining the principal component comprehensive score based on the weighted sum of the principal component entropy value and the principal component sample weight, wherein the principal component entropy value and the principal component sample weight are determined based on the principal component score, includes: normalizing the principal components corresponding to the samples in the normalized matrix to generate corresponding principal component sample weights; calculating the entropy value of the principal component based on the principal component sample weights; calculating the weight of the principal component based on the entropy value; and determining the principal component comprehensive score based on the weight of the principal component and the principal component sample weight.

[0010] Optionally, the step of calculating the meteorological similarity between the predicted day and the historical day based on the historical daily data and predicted daily data corresponding to the indicators divided by the principal component comprehensive score includes: dividing the indicators into important indicators and general indicators according to the principal component comprehensive score; obtaining the maximum and minimum values ​​of historical daily data for the predicted day under the important indicators based on the historical daily data and the predicted day data; obtaining the average value of historical daily data for the predicted day under the general indicators based on the historical daily data and the predicted day data; and calculating the meteorological similarity between the historical day and the predicted day using Euclidean distance on the meteorological matrix composed of the average value and the maximum and minimum values.

[0011] Optionally, determining the set of historical days similar to the predicted day based on the meteorological similarity and the time similarity between the predicted day and historical days includes: determining the time similarity based on the seasonal similarity and date similarity between the predicted day and historical days; calculating the daily similarity between each historical day and the predicted day based on the dot product of the meteorological similarity and the time similarity; sorting the daily similarities and selecting the first preset number of historical days to generate the set of historical days similar to the predicted day.

[0012] According to a second aspect of the present invention, a baseline load forecasting device is also provided, characterized in that the device comprises: an acquisition module, configured to acquire historical load data and historical daily data and forecast daily data determined by indicators, wherein the indicators include temperature, humidity, air pressure, precipitation, irradiance, and wind speed; a first determination module, configured to determine the principal components and corresponding principal component comprehensive scores in the matrix composed of the historical daily data using a principal component analysis method based on entropy improvement; a first calculation module, configured to calculate the meteorological similarity between the forecast day and historical days based on the historical daily data and forecast daily data corresponding to the indicators divided by the principal component comprehensive scores; a second determination module, configured to determine a set of historical days similar to the forecast day based on the meteorological similarity and the temporal similarity between the forecast day and historical days; and a obtaining module, configured to obtain the forecast daily load data using a backpropagation neural network based on the historical load data, the set of historical days, the meteorological matrix, and the forecast daily data, wherein the meteorological matrix is ​​composed of the historical daily data and forecast daily data corresponding to the divided indicators.

[0013] Optionally, the apparatus further includes: a generation module, configured to generate a historical daily data matrix based on the historical daily data, wherein the row vectors of the historical daily data matrix represent a sample, the sample including the historical daily data corresponding to the indicator at the target time, and the column vectors of the historical daily data matrix including the historical daily data of an indicator at all times; a standardization processing module, configured to perform standardization processing on the historical daily data matrix; a pre-impact module, configured to pre-impact missing data based on the average value of the historical daily data in the column vectors of the standardized matrix; a third determination module, configured to determine the sample in the pre-impacted matrix that is most relevant to the sample containing the missing data based on a grey relational algorithm; and a second calculation module, configured to calculate the imputation value of the missing data based on the samples in the matrix using the KNN algorithm and the DBSCAN algorithm.

[0014] Optionally, the second calculation module includes: a first determining unit, configured to determine the neighborhood radius and threshold of the DBSCAN algorithm based on the data of the samples in the pre-impacted matrix most relevant to the samples containing missing data; a second determining unit, configured to determine outliers in the data based on the neighborhood radius and threshold; and a first calculation unit, configured to calculate the imputation value of the missing data in the data of the samples after removing outliers using the KNN algorithm.

[0015] Optionally, the first determining module includes: a first generating unit, configured to generate a complete historical day data matrix based on the filled historical day data; a normalization processing unit, configured to normalize the complete historical day data matrix based on the mean of the column vectors in the complete historical day data matrix; a second calculation unit, configured to calculate the covariance matrix of the normalized matrix and the eigenvalues ​​and eigenvectors of the covariance matrix; a third determining unit, configured to determine the principal components in the index based on the cumulative variance contribution rate of the eigenvalues; a third calculation unit, configured to calculate the principal component score based on the eigenvectors and the normalized matrix; and a fourth determining unit, configured to determine the principal component comprehensive score based on the weighted sum of the principal component entropy value and the principal component sample weight, wherein the principal component entropy value and the principal component sample weight are determined based on the principal component score.

[0016] Optionally, the fourth determining unit includes: a generation submodule, used to normalize the principal components corresponding to the samples in the normalized matrix to generate corresponding principal component sample weights; a first calculation submodule, used to calculate the entropy value of the principal components based on the principal component sample weights; a second calculation submodule, used to calculate the weights of the principal components based on the entropy values; and a determining submodule, used to determine the principal component comprehensive score based on the weights of the principal components and the weights of the principal component sample weights.

[0017] Optionally, the first calculation module includes: a partitioning unit, used to divide the indicators into important indicators and general indicators according to the principal component comprehensive score; a first acquisition unit, used to acquire the maximum and minimum values ​​of historical daily data and historical daily data under important indicators for the historical day and the predicted day, based on the historical daily data and the predicted day data; a second acquisition unit, used to acquire the average value of historical daily data and historical daily data under general indicators for the historical day and the predicted day, based on the historical daily data and the predicted day data; and a fourth calculation unit, used to calculate the meteorological similarity between the historical day and the predicted day using Euclidean distance on the meteorological matrix composed of the average value and the maximum and minimum values.

[0018] Optionally, the second determining module includes: a fifth determining unit, used to determine time similarity based on the seasonal similarity and date similarity between the predicted day and historical days; a fifth calculating unit, used to calculate the daily similarity between each historical day and the predicted day based on the dot product of the meteorological similarity and the time similarity; and a second generating unit, used to sort the daily similarities and select a preset number of historical days to generate a set of historical days similar to the predicted day.

[0019] According to a third aspect of the present invention, an electronic device is also provided, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; wherein the memory is used to store a computer program; and the processor is used to execute the method steps of any of the above embodiments by running the computer program stored in the memory.

[0020] According to a fourth aspect of the present invention, a computer-readable storage medium is also provided, wherein a computer program is stored therein, wherein the computer program is configured to execute the method steps of any of the above embodiments when running.

[0021] In this embodiment of the invention, historical load data, historical daily data, and predicted daily data determined by indicators are acquired. Principal component analysis (PCA) based on an improved entropy method is used to determine the principal components and corresponding comprehensive scores of the indicators in the matrix composed of historical daily data. The meteorological similarity between predicted and historical days is calculated based on the historical and predicted daily data corresponding to the indicators after division by the comprehensive scores of the principal components. A set of historical days similar to the predicted day is determined based on the meteorological similarity and the temporal similarity between the predicted and historical days. A backpropagation neural network is used to obtain the predicted daily load data based on the historical load data, the set of historical days, the meteorological matrix, and the predicted daily data. Traditional preprocessing methods, such as PCA, have limitations such as inaccurate calculation results and ambiguity in the meaning of principal components due to dimensionality reduction, which affect the accuracy of load forecasting and reduce forecast accuracy. This invention uses a principal component analysis method improved based on entropy value method when dividing indicators, and recalculates the weights of each principal component, thereby improving the accuracy of principal component weights and thus improving the quality of data preprocessing before model prediction. This improves the accuracy of meteorological similarity calculation and achieves the technical effect of improving the accuracy of model prediction. It also solves the problem of low data quality obtained from data preprocessing before baseline load prediction in related technologies.

[0022] In this embodiment of the invention, a historical day data matrix is ​​generated based on historical day data; the historical day data matrix is ​​standardized and pre-impacted; based on the grey relational algorithm, the samples in the pre-impacted matrix most relevant to the samples containing missing data are determined; the KNN and DBSCAN algorithms are used to calculate the imputation values ​​for the missing data based on the samples in the matrix. By using the grey relational algorithm instead of the traditional distance formula to obtain nearest neighbor values, the reliability of the nearest neighbor values ​​is improved. Combining the KNN algorithm with the DBSCAN algorithm, outliers are detected during clustering, eliminating interference from outliers (noise points), thus improving the data imputation effect and further enhancing the quality of the data obtained after preprocessing. Attached Figure Description

[0023] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0025] Figure 1 This is a schematic diagram of the hardware environment for an optional baseline load forecasting method according to an embodiment of the present invention;

[0026] Figure 2 This is a flowchart illustrating an optional baseline load forecasting method according to an embodiment of the present invention;

[0027] Figure 3 This is a schematic diagram of the overall process of an optional baseline load forecasting method according to an embodiment of the present invention;

[0028] Figure 4 This is a structural block diagram of an optional baseline load forecasting device according to an embodiment of the present invention;

[0029] Figure 5 This is a structural block diagram of an optional electronic device according to an embodiment of the present invention. Detailed Implementation

[0030] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0031] It should be noted that in the description of this invention, the terms "first," "second," etc., are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0032] According to one aspect of the present invention, a baseline load forecasting method is provided. Optionally, in this embodiment, the above-described baseline load forecasting method can be applied to, for example... Figure 1 In the hardware environment shown. For example... Figure 1As shown, terminal 102 may include memory 104, processor 106, and display 108 (optional component). Terminal 102 can communicate with server 112 via network 110. Server 112 can provide services (such as application services) to the terminal or clients installed on the terminal. Database 114 can be set up on or independently of server 112 to provide data storage services to server 112. In addition, server 112 may run a processing engine 116, which can be used to execute the steps performed by server 112.

[0033] Optionally, in this embodiment, the baseline load forecasting method described above can be executed by server 112, by terminal 102, or jointly by server 112 and terminal 102. The execution of the baseline load forecasting method of this embodiment by terminal 102 can also be performed by a client installed on it. Taking the application of the baseline load forecasting method to the central processing unit as an example... Figure 2 This is a flowchart illustrating an optional baseline load forecasting method according to an embodiment of the present invention, as shown below. Figure 2 As shown, the process of this method may include the following steps:

[0034] Step S201: Obtain historical load data and historical daily data and forecast daily data determined by indicators, including meteorological environmental indicators such as temperature, humidity, air pressure, precipitation, irradiance, and wind speed. Optionally, first obtain historical daily data and forecast daily data related to temperature, humidity, air pressure, precipitation, irradiance, and wind speed indicators, as well as historical load data. Historical daily data includes temperature, humidity, air pressure, precipitation, irradiance, and wind speed data at certain times on each day of past years. Forecast daily data consists of known or predicted temperature, humidity, air pressure, precipitation, irradiance, and wind speed data for the date to be predicted for the load forecast.

[0035] Step S202: For the matrix composed of historical daily data, a principal component analysis method based on entropy value improvement is used to determine the principal components in the index and the corresponding comprehensive scores of the principal components. Optionally, the historical daily data is represented in matrix form, and the principal components in the index are determined using the principal component analysis method based on entropy value improvement, and the corresponding comprehensive scores of the principal components are calculated according to the principal component analysis method based on entropy value improvement. In calculating the comprehensive scores of the principal components, the principal component analysis method based on entropy value improvement uses the entropy value corresponding to each principal component to weight the principal component scores, thereby improving the accuracy of the principal component weights.

[0036] Step S203: Calculate the meteorological similarity between the predicted day and the historical day based on the historical and predicted day data corresponding to the indicators segmented by the principal component comprehensive score. Optionally, the indicators are segmented according to the principal component comprehensive score calculated in step S202. A new matrix is ​​constructed using the segmented indicators as the basis for dividing the column vectors of the new matrix. Then, the meteorological similarity between the predicted day and the historical day is calculated based on the new matrix, which is the similarity between the predicted day and the historical day under the meteorological environment obtained by comprehensively considering the segmented indicators.

[0037] Step S204: Determine the set of historical days similar to the predicted day based on meteorological similarity and the time similarity between the predicted day and historical days. Optionally, determine the set of historical days most similar to the predicted day based on the dot product of the time similarity between historical days and the predicted day and the meteorological similarity calculated in step S203, that is, select multiple historical dates most similar to the predicted day.

[0038] Step S205: A backpropagation neural network is used to obtain the predicted daily load data based on historical load data, historical day set, meteorological matrix, and predicted daily data. The meteorological matrix consists of historical day data and predicted daily data corresponding to the divided indicators. Optionally, based on the historical day set obtained in step S204, corresponding data are selected from the historical load data and meteorological matrix to train the backpropagation neural network. Through training, the backpropagation neural network can fit the relationship curve between meteorological data and load, and thus the trained model is used to obtain the predicted daily load forecast result based on the predicted daily data.

[0039] In this embodiment of the invention, historical load data and historical and predicted daily data determined by indicators are acquired. A principal component analysis method based on entropy is used to determine the principal components and corresponding principal component comprehensive scores in the matrix composed of historical daily data. The meteorological similarity between predicted and historical days is calculated based on the historical and predicted daily data corresponding to the indicators after classification by principal component comprehensive scores. A set of historical days similar to the predicted day is determined based on the meteorological similarity and the temporal similarity between the predicted and historical days. A backpropagation neural network is used to obtain the predicted day load data based on the historical load data, the set of historical days, the meteorological matrix, and the predicted day data. Because an improved principal component analysis method is used when classifying indicators, the quality of the data is improved, and the accuracy of the principal component weights is further improved, thereby improving the accuracy of meteorological similarity calculation. This achieves the technical effect of improving the accuracy of model prediction and solves the problem of low data quality obtained from data preprocessing before baseline load prediction in related technologies.

[0040] As an optional embodiment, before using a principal component analysis method based on entropy to determine the principal components and corresponding principal component comprehensive scores in the matrix composed of historical daily data, the method further includes: generating a historical daily data matrix, wherein the row vectors of the historical daily data matrix represent a sample, a sample includes the historical daily data corresponding to the indicator at the target time, and the column vectors of the historical daily data matrix include the historical daily data of an indicator at all times; standardizing the historical daily data matrix; pre-impacting missing data based on the average value of the historical daily data in the column vectors of the standardized matrix; determining the samples in the pre-impacted matrix that are most relevant to the samples containing missing data based on the grey relational algorithm; and calculating the imputation values ​​of the missing data based on the samples in the matrix using the KNN algorithm and the DBSCAN algorithm.

[0041] Optionally, before using the improved principal component analysis method, missing data can be imputed to make the historical data more complete. Specifically, the historical daily data can be represented in matrix form, as shown below:

[0042]

[0043] Wherein, X is the historical daily data matrix, which is an m×n dimensional matrix. The row vector of the historical daily data matrix X represents a sample, which includes the historical daily data of each indicator at the target time, namely temperature, humidity, air pressure, precipitation, irradiance and wind speed. The column vector of the historical daily data matrix X is the historical daily data of an indicator at all times.

[0044] The historical daily data matrix is ​​standardized by calculating the mean and standard deviation of each sample. The difference between each value in the historical daily data matrix X and its mean is then divided by the standard deviation to obtain the standardized matrix. Missing data is then pre-impounded based on the mean of the historical daily data in the column vectors of the standardized matrix. Furthermore, to avoid coincidences in the imputation results, three missing data patterns—completely random missing, random missing, and non-random missing—are constructed based on the relationship between the proportion of missing data in the matrix column vectors and a threshold. The matrices corresponding to these three missing data patterns are denoted as T1, T2, and T3, respectively. Pre-impounding is performed on T1, T2, and T3 using different calculation methods, such as using the squared mean or weighted average, to obtain the pre-impounded matrices V1, V2, and V3. The matrices T1, T2, and T3 corresponding to the three missing data patterns, or the pre-impounded matrices V1, V2, and V3, can be visualized to explore the correlations.

[0045] Then, based on the grey relational analysis algorithm, for the missing data in the pre-impacted matrix, other samples most relevant to the sample containing the missing data are determined. Specifically, let the target value, i.e., the sample containing the missing data, be x. ir Calculate x ir Distance to other samples in the pre-impacted matrix. As a reference sequence, a reference sequence is a vector used for reference. It can be constructed using the optimal (or worst) values ​​of each indicator, or other reference values ​​can be selected according to the evaluation purpose. As the sequence being compared, let it be denoted as {x} i The comparison sequence is a sample in the pre-filled matrix, from which the most relevant sample is selected.

[0046] Calculate the correlation coefficient ξ between the reference series and the compared series. i (k), based on the correlation coefficient ξ i (k) Identify the samples in the pre-impacted matrix that are most relevant to the samples containing missing data, with the correlation coefficient ξ. i A larger (k) value indicates a stronger correlation. Let ξ i (k) The sample with the maximum value is the most relevant sample in the pre-impacted matrix, and the correlation coefficient ξ is... i The formula for calculating (k) is as follows:

[0047]

[0048] in, For the reference sequence {x * The corresponding element in} For the sequence being compared {x i The corresponding element in}, ρ∈[0,1] is an adjustable parameter, usually ρ=0.5.

[0049] Then, the KNN and DBSCAN algorithms are used based on the correlation coefficient ξ. i (k) Calculate the imputation value for the missing data in the pre-imputation matrix that is most relevant to the sample containing the missing data.

[0050] As an optional embodiment, the KNN algorithm and the DBSCAN algorithm are used to calculate the imputation value of missing data based on the samples in the matrix, including: determining the neighborhood radius and threshold of the DBSCAN algorithm based on the data of the samples in the pre-imputation matrix that are most relevant to the samples containing missing data; determining outliers in the data based on the neighborhood radius and threshold; and using the KNN algorithm to calculate the imputation value of missing data on the data in the samples after removing outliers.

[0051] Optionally, before using the KNN algorithm to calculate missing values ​​based on the most relevant samples, the DBSCAN algorithm is used to remove outliers (noise) from the most relevant samples. This involves determining the neighborhood radius and threshold of the most relevant samples, then determining the sample boundaries based on the neighborhood radius and threshold, and treating data outside the boundaries as outliers. The correlation coefficient ξ will then be used to... i (k) The outliers in the determined samples are removed using the methods described above. Finally, the KNN algorithm is used to calculate the imputation values ​​for the missing data in the samples after outlier removal according to the following formula:

[0052]

[0053] In the formula, x0 is the fill value, ω kr Let ω be the weight of the K nearest neighbor target value. For ease of calculation, let ω kr =1 / k, x kr Based on the correlation coefficient ξ i (k) is a determined sample.

[0054] In this embodiment of the invention, a historical day data matrix is ​​generated based on historical day data; the historical day data matrix is ​​standardized and pre-impacted; based on the grey relational algorithm, the samples in the pre-impacted matrix most relevant to the samples containing missing data are determined; the KNN and DBSCAN algorithms are used to calculate the imputation values ​​for the missing data based on the samples in the matrix. By using the grey relational algorithm instead of the traditional distance formula to obtain nearest neighbor values, the reliability of the nearest neighbor values ​​is improved. Combining the KNN algorithm with the DBSCAN algorithm, outliers are detected during clustering, eliminating interference from outliers (noise points), thus improving the data imputation effect and further enhancing the quality of the data obtained after preprocessing.

[0055] As an optional embodiment, a principal component analysis method based on entropy value improvement is used to determine the principal components and corresponding principal component comprehensive scores in the matrix composed of historical daily data. This includes: generating a complete historical daily data matrix based on the imputed historical daily data; normalizing the complete historical daily data matrix based on the mean of the column vectors in the complete historical daily data matrix; calculating the covariance matrix of the normalized matrix and its eigenvalues ​​and eigenvectors; determining the principal components in the index based on the cumulative variance contribution rate of the eigenvalues; calculating the principal component scores based on the eigenvectors and the normalized matrix; and determining the principal component comprehensive score based on the weighted sum of the principal component entropy values ​​and the principal component sample weights, wherein the principal component entropy values ​​and principal component sample weights are determined based on the principal component scores.

[0056] Optionally, a complete historical day data matrix is ​​generated based on the imputed historical day data, and then the complete historical day data matrix is ​​normalized according to the following formula:

[0057]

[0058] In the formula, X ij Let x be the normalized value of the j-th index for the i-th sample. ij For the imputed historical daily data corresponding to the j-th indicator of the i-th sample, Let be the sample mean of the j-th indicator, n be the sample size, and p be the number of indicators.

[0059] Calculate the covariance matrix of the normalized matrix using the following formula:

[0060]

[0061] In the formula, s kj Let be the covariance between the k-th index and the j-th index. Let be the normalized sample mean of the k-th indicator. Here, under the condition of data sample normalization, the covariance matrix is ​​the correlation coefficient matrix. The Jacobian determinant method is used to obtain the eigenvalues ​​λ of the j-th indicator in the correlation coefficient matrix. j (j = 1, 2, ..., p) and eigenvectors, sorting the eigenvalues ​​in descending order (λ1 ≥ λ2 ≥ ... ≥ λ). p >0).

[0062] The variance contribution rate G(λ) is calculated using the following formula, which represents the proportion of a specific eigenvalue to the total number of eigenvalues. j ):

[0063]

[0064] Then, the principal components in the index are determined based on the cumulative variance contribution rate of the eigenvalues. The cumulative variance contribution rate is calculated using the following formula:

[0065]

[0066] In the formula, G(m) is the cumulative variance contribution rate of the first m eigenvalues. When G(m) > 85%, m is called the number of principal components, and the first m eigenvalues ​​are the principal components in the index. It should be noted that the variance contribution rate G(λ) j The cumulative variance contribution rate (G(m)) represents the proportion of information contained in that dimension, i.e., the indicator. When the cumulative variance contribution rate G(m) > 85%, it indicates that the indicator contains sufficient information. Furthermore, the principal component loadings (l) can be calculated using the following formula: cj Wherein, principal component loadings represent the correlation coefficients between the original variables and the principal components in principal component analysis:

[0067]

[0068] In the formula, l cj Let a be the loading coefficient of the j-th index of the c-th principal component. cj λ is the j-th element of the eigenvector corresponding to the c-th principal component. c is the eigenvalue corresponding to the c-th principal component.

[0069] Principal component scores F are calculated based on the eigenvectors of the covariance matrix and the normalized matrix. c The calculation formula is as follows:

[0070] F c =a c1 X1+a c2 X2 + ... + a cp X p (c = 1, 2, ..., m)

[0071] In the formula, X p These are the column vectors in the normalized matrix. Finally, the principal component comprehensive score is determined by a weighted sum of the principal component entropy values ​​and the proportions of principal component samples. Both the principal component entropy values ​​and the principal component sample proportions are based on the principal component score F. c Sure.

[0072] As an optional embodiment, the principal component comprehensive score is determined based on the weighted sum of the principal component entropy value and the principal component sample weight. The principal component entropy value and principal component sample weight are determined based on the principal component score, including: normalizing the principal components corresponding to the samples in the normalized matrix to generate the corresponding principal component sample weights; calculating the entropy value of the principal component based on the principal component sample weights; calculating the weight of the principal component based on the entropy value; and determining the principal component comprehensive score based on the weight of the principal component and the principal component sample weight.

[0073] In traditional principal component analysis, the overall principal component score is calculated using the following formula:

[0074]

[0075] The principal component analysis method based on the entropy method proposed in this invention normalizes each principal component according to the following formula:

[0076]

[0077] In the formula, F is the normalized score of the c-th principal component of the i-th sample. ic F is the score of the c-th principal component of the i-th sample. c(max) For the maximum score of the c-th principal component, F c(min) The minimum score of the c-th principal component.

[0078] The following formula is used to calculate the proportion P of the principal component sample corresponding to the c-th principal component of the i-th sample, based on the normalized score of the calculated principal components. ic :

[0079]

[0080] Then, use the following formula based on the principal component sample weight P. ic Calculate the entropy value E of the c-th principal component. c :

[0081]

[0082] The entropy value E of the c-th principal component is used as follows: c Calculate the weights ω of the corresponding principal components. c :

[0083]

[0084] Finally, based on the weights ω of the principal components... c Principal component sample weight P ic The weighted sum of the principal components determines the overall principal component score F', and the calculation formula is shown below:

[0085]

[0086] In this embodiment, the influence of principal component entropy is considered when calculating the principal component composite score. The weights of each principal component are recalculated based on the degree of variation of each principal component, i.e., the dispersion of the observed values, according to traditional principal component analysis. Traditional principal component analysis only considers the variance contribution rate of the principal components for weighted composite calculation. When the principal components are independent of each other, the information content of the calculation result may decrease rather than increase. This embodiment improves principal component analysis by using the entropy method, achieving the goal of highlighting local system information and being less affected by subjective factors. Using the improved principal component composite score as the index division parameter improves the accuracy of the division and has better practicality and versatility.

[0087] As an optional embodiment, the meteorological similarity between the predicted day and the historical day is calculated based on the historical daily data and predicted daily data corresponding to the indicators after being divided by the principal component comprehensive score. This includes: dividing the indicators into important indicators and general indicators according to the principal component comprehensive score; obtaining the maximum and minimum values ​​of historical daily data for the predicted day under the important indicators based on the historical daily data and the predicted day data; obtaining the average value of historical daily data for the predicted day under the general indicators based on the historical daily data and the predicted day data; and calculating the meteorological similarity between the historical day and the predicted day using Euclidean distance on the meteorological matrix composed of the average value and the maximum and minimum values.

[0088] Optionally, the principal component composite scores are sorted, and the indicators corresponding to the composite scores of the top preset number of principal components are taken as important indicators, while the others are taken as general indicators. In this embodiment, the two parameters with higher composite scores are temperature and humidity, so temperature and humidity are taken as important indicators to be considered when forecasting load, while wind speed, air pressure, precipitation, and irradiance are general indicators. The maximum and minimum values ​​of the important indicators, i.e., temperature and humidity, and the average values ​​of the general indicators, i.e., wind speed, air pressure, precipitation, and irradiance, are obtained from the historical daily data to form a meteorological matrix. The row vector of the meteorological matrix can be represented as W = [T ph ,T el H ph H el [Wind, MP, DP, R], where each term in W represents the maximum and minimum temperature, maximum and minimum humidity, average wind speed, average air pressure, average precipitation, and average irradiance, respectively. The resulting meteorological matrix is ​​standardized according to the following formula:

[0089]

[0090] In the formula, x jmax x represents the column vector of the meteorological matrix ij The maximum value of x jmin x represents the column vector of the meteorological matrix ij The minimum value, where the calculation of the above maximum and minimum values ​​does not include the forecast day data, x ij This represents the value in the i-th row and j-th column of the meteorological matrix, i.e., the value of the j-th indicator in W corresponding to the i-th day. For the standardized meteorological matrix, the meteorological similarity between historical days and predicted days is calculated using the following formula:

[0091]

[0092] As an optional embodiment, a set of historical days similar to the predicted day is determined based on meteorological similarity and the time similarity between the predicted day and historical days, including: determining the time similarity based on the seasonal similarity and date similarity between the predicted day and historical days; calculating the daily similarity between each historical day and the predicted day based on the dot product of meteorological similarity and time similarity; sorting the daily similarity and selecting the first preset number of historical days to generate a set of historical days similar to the predicted day.

[0093] Optionally, time similarity is determined based on the seasonal and date similarity between the predicted date and historical dates, i.e., judging the seasonal similarity and the magnitude of the date span between historical dates and the predicted date. This is calculated using the following formula:

[0094]

[0095]

[0096] In the formula, σ(t) represents the time similarity. Indicating seasonal similarity, The similarity is represented by β1 and β2, which are the seasonal similarity decay coefficients and date span decay coefficients between historical and predicted dates, respectively. In this embodiment, β1 = 0.96 and β2 = 0.95. θ1 and θ2 are adjustment coefficients corresponding to the season and date, which can be set as needed. For ease of calculation, the calculation is based on 360 days per year and 90 days per season. If t > 360, then t = t - 360. In t1(t), if... So

[0097] The daily similarity between each historical day and the predicted day is calculated using the following formula:

[0098]

[0099] In the formula, σ(t) represents the time similarity. Indicating meteorological similarity, sim i For daily similarity, the similarity of days is sorted according to the calculation results, and a set of historical days similar to the predicted day is generated by selecting the first preset number of historical days.

[0100] In this embodiment, by introducing a similar day selection method, the climate, time, season and other variables of the day to be predicted are used as references. Historical days with similar climate, time and season are selected as inputs to the neural network model to train the model, thereby improving the accuracy and precision of load prediction. In the face of special or extreme weather, the prediction curve can be effectively fitted, making the load prediction more in line with the actual situation.

[0101] As an optional embodiment, Figure 3 This is a schematic diagram of the overall process of an optional baseline load forecasting method according to an embodiment of the present invention, as shown below. Figure 3As shown, historical load data and meteorological data are collected, specifically historical data for each day under meteorological indicators such as temperature, humidity, air pressure, precipitation, irradiance, and wind speed. Then, the historical data is traversed to remove outliers and impute them. Specifically, based on a grey relational algorithm, the data most relevant to the imputed values ​​are determined, and then KNN and DBSCAN algorithms are used to remove outliers from the relevant data to impute the collected meteorological data. Next, a time factor is calculated to construct a daily feature vector, i.e., the meteorological factor, that can represent the meteorological characteristics of the day. The time factor and meteorological factor are two important influencing factors in baseline load forecasting. The meteorological factor and time factor are used to rank the data for the corresponding weather categories based on similarity, thereby obtaining a set of similar days. The preprocessed data is then fed into a BP neural network load forecasting model to output the forecast results.

[0102] According to another aspect of the present invention, a baseline load forecasting apparatus is also provided, such as... Figure 4 As shown, the device may include: an acquisition module 401, used to acquire historical load data and historical daily data and predicted daily data determined by indicators, wherein the indicators include temperature, humidity, air pressure, precipitation, irradiance, and wind speed; a first determination module 402, used to determine the principal components and corresponding principal component comprehensive scores in the matrix composed of the historical daily data using a principal component analysis method improved based on the entropy method; a first calculation module 403, used to calculate the meteorological similarity between the predicted day and the historical day based on the historical daily data and predicted daily data corresponding to the indicators divided by the principal component comprehensive scores; a second determination module 404, used to determine a set of historical days similar to the predicted day based on the meteorological similarity and the time similarity between the predicted day and the historical day; and a result module 405, used to obtain the predicted day load data using a backpropagation neural network based on the historical load data, the set of historical days, the meteorological matrix, and the predicted day data, wherein the meteorological matrix is ​​composed of the historical daily data and predicted day data corresponding to the divided indicators.

[0103] It should be noted that the acquisition module 401 in this embodiment can be used to execute the above step S201, the first determination module 402 in this embodiment can be used to execute the above step S202, the first calculation module 403 in this embodiment can be used to execute the above step S203, the second determination module 404 in this embodiment can be used to execute the above step S204, and the obtaining module 405 in this embodiment can be used to execute the above step S205.

[0104] Through the above modules, an improved principal component analysis method was used when dividing the indicators, which improved the quality of the data and further improved the accuracy of the principal component weights. This improved the accuracy of meteorological similarity calculation and achieved the technical effect of improving the accuracy of model prediction. It also solved the problem of low data quality obtained from data preprocessing before baseline load prediction in related technologies.

[0105] As an optional embodiment, the apparatus further includes: a generation module for generating a historical daily data matrix based on historical daily data, wherein the row vectors of the historical daily data matrix represent a sample, a sample includes historical daily data corresponding to an indicator at a target time, and the column vectors of the historical daily data matrix include historical daily data of an indicator at all times; a standardization processing module for standardizing the historical daily data matrix; a pre-impact module for pre-impacting missing data based on the average value of historical daily data in the column vectors of the standardized matrix; a third determination module for determining, based on a grey relational algorithm, the sample in the pre-impacted matrix most relevant to the sample containing missing data; and a second calculation module for calculating the imputation value of the missing data based on the samples in the matrix using the KNN algorithm and the DBSCAN algorithm.

[0106] As an optional embodiment, the second calculation module includes: a first determining unit, used to determine the neighborhood radius and threshold of the DBSCAN algorithm based on the data of the samples in the pre-impacted matrix most relevant to the samples containing missing data; a second determining unit, used to determine outliers in the data based on the neighborhood radius and threshold; and a first calculation unit, used to calculate the imputation value of the missing data in the data of the samples after removing outliers using the KNN algorithm.

[0107] As an optional embodiment, the first determining module includes: a first generating unit, used to generate a complete historical day data matrix based on the imputed historical day data; a normalization processing unit, used to normalize the complete historical day data matrix based on the mean of the column vectors in the complete historical day data matrix; a second calculation unit, used to calculate the covariance matrix of the normalized matrix and the eigenvalues ​​and eigenvectors of the covariance matrix; a third determining unit, used to determine the principal components in the index based on the cumulative variance contribution rate of the eigenvalues; a third calculation unit, used to calculate the principal component scores based on the eigenvectors and the normalized matrix; and a fourth determining unit, used to determine the principal component comprehensive score based on the weighted sum of the principal component entropy value and the principal component sample weight, wherein the principal component entropy value and the principal component sample weight are determined based on the principal component scores.

[0108] As an optional embodiment, the fourth determining unit includes: a generation submodule, used to normalize the principal components corresponding to the samples in the normalized matrix and generate the corresponding principal component sample weights; a first calculation submodule, used to calculate the entropy value of the principal components based on the principal component sample weights; a second calculation submodule, used to calculate the weights of the principal components based on the entropy values; and a determining submodule, used to determine the principal component comprehensive score based on the weights of the principal components and the principal component sample weights.

[0109] As an optional embodiment, the first calculation module includes: a partitioning unit, used to divide the indicators into important indicators and general indicators according to the principal component comprehensive score; a first acquisition unit, used to acquire the maximum and minimum values ​​of historical daily data and predicted daily data under important indicators based on historical daily data and predicted daily data; a second acquisition unit, used to acquire the average value of historical daily data and predicted daily data under general indicators based on historical daily data and predicted daily data; and a fourth calculation unit, used to calculate the meteorological similarity between historical days and predicted days using Euclidean distance on the meteorological matrix composed of the average value and the maximum and minimum values.

[0110] As an optional embodiment, the second determining module includes: a fifth determining unit, used to determine the time similarity based on the seasonal similarity and date similarity between the predicted day and historical days; a fifth calculating unit, used to calculate the daily similarity between each historical day and the predicted day based on the dot product of meteorological similarity and time similarity; and a second generating unit, used to sort the daily similarities and select a preset number of historical days to generate a set of historical days similar to the predicted day.

[0111] It should be noted that the examples and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the content disclosed in the above embodiments. It should also be noted that the above modules, as part of a device, can operate in environments such as... Figure 1 The hardware environment shown can be implemented through software or hardware, and the hardware environment includes the network environment.

[0112] According to another aspect of the present invention, an electronic device for implementing the baseline load forecasting method described above is also provided. The electronic device may be a server, a terminal, or a combination thereof.

[0113] Figure 5 This is a structural block diagram of an optional electronic device according to an embodiment of the present invention, such as... Figure 5As shown, the system includes a processor 501, a communication interface 502, a memory 503, and a communication bus 504. The processor 501, communication interface 502, and memory 503 communicate with each other via the communication bus 504. The memory 503 stores computer programs. When the processor 501 executes the computer program stored in the memory 503, it performs the following steps:

[0114] Historical load data, as well as historical and forecast daily data determined by indicators including temperature, humidity, air pressure, precipitation, irradiance, and wind speed, are acquired. Principal component analysis (PCA) based on an improved entropy method is used to determine the principal components and corresponding comprehensive scores of the indicators within the matrix composed of historical daily data. Meteorological similarity between forecast and historical days is calculated based on the historical and forecast daily data corresponding to the indicators after division by the comprehensive scores of the principal components. A set of historical days similar to the forecast day is determined based on the meteorological similarity and the temporal similarity between the forecast and historical days. A backpropagation neural network is used to obtain the forecast daily load data from the historical load data, the set of historical days, the meteorological matrix, and the forecast daily data. The meteorological matrix is ​​composed of the historical and forecast daily data corresponding to the divided indicators.

[0115] As an example, such as Figure 5 As shown, the memory 503 may include, but is not limited to, the acquisition module 401, the first determination module 402, the first calculation module 403, the second determination module 404, and the obtaining module 405 from the baseline load forecasting device. Furthermore, it may include, but is not limited to, other module units from the baseline load forecasting device, which will not be elaborated in this example. The processor may be a general-purpose processor, including, but not limited to, CPUs (Central Processing Units), NPs (Network Processors), etc. In addition, the electronic device also includes a display for displaying the baseline load forecasting results.

[0116] Those skilled in the art will understand that Figure 5 The structure shown is for illustrative purposes only. The device that implements the above baseline load forecasting method can be a terminal device, such as a smartphone (e.g., an Android phone, an iOS phone), a tablet computer, or other terminal devices. Figure 5 This does not limit the structure of the aforementioned electronic devices. For example, the terminal device may also include components that are more... Figure 5 The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 5The different configurations shown are illustrated. Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. This program can be stored in a computer-readable storage medium, which may include: a flash drive, ROM, RAM, a magnetic disk, or an optical disk, etc.

[0117] According to another aspect of the present invention, a storage medium is also provided. Optionally, in this embodiment, the storage medium can be used to execute program code for a baseline load forecasting method. Optionally, in this embodiment, the storage medium can be located on at least one of a plurality of network devices in the network shown in the above embodiment. Optionally, in this embodiment, the storage medium is configured to store program code for performing the following steps:

[0118] Historical load data, as well as historical and forecast daily data determined by indicators including temperature, humidity, air pressure, precipitation, irradiance, and wind speed, are acquired. Principal component analysis (PCA) based on an improved entropy method is used to determine the principal components and corresponding comprehensive scores of the indicators within the matrix composed of historical daily data. Meteorological similarity between forecast and historical days is calculated based on the historical and forecast daily data corresponding to the indicators after division by the comprehensive scores of the principal components. A set of historical days similar to the forecast day is determined based on the meteorological similarity and the temporal similarity between the forecast and historical days. A backpropagation neural network is used to obtain the forecast daily load data from the historical load data, the set of historical days, the meteorological matrix, and the forecast daily data. The meteorological matrix is ​​composed of the historical and forecast daily data corresponding to the divided indicators.

[0119] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments, and will not be repeated here. Optionally, in this embodiment, the above storage medium may include, but is not limited to, various media capable of storing program code such as USB flash drives, ROM, RAM, portable hard drives, magnetic disks, or optical disks.

[0120] The sequence numbers of the above embodiments of the present invention are merely for descriptive purposes and do not represent the superiority or inferiority of the embodiments. In the above embodiments of the present invention, the descriptions of each embodiment have their own emphasis; parts not described in detail in a certain embodiment can be referred to in the relevant descriptions of other embodiments. The above descriptions are merely preferred embodiments of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A baseline load forecasting method, characterized in that, The method includes: Acquire historical load data as well as historical daily data and predicted daily data determined by indicators, wherein the indicators include temperature, humidity, air pressure, precipitation, irradiance and wind speed; A historical daily data matrix is ​​generated based on the historical daily data, wherein the row vector of the historical daily data matrix represents a sample, the sample includes the historical daily data corresponding to the indicator at the target time, and the column vector of the historical daily data matrix includes the historical daily data of an indicator at all times. The historical daily data matrix is ​​standardized. Missing data are pre-filled based on the average value of historical daily data in the column vectors of the standardized matrix; Based on the grey relational algorithm, the sample in the pre-impacted matrix that is most relevant to the sample containing missing data is determined. The KNN and DBSCAN algorithms are used to calculate the imputation values ​​for missing data based on the samples in the matrix. The principal components and their corresponding comprehensive scores are determined by using a principal component analysis method based on entropy to improve the principal component method on the matrix composed of the historical daily data. The meteorological similarity between the predicted day and the historical day is calculated based on the historical daily data and predicted daily data corresponding to the indicators divided by the principal component comprehensive score. Based on the meteorological similarity and the temporal similarity between the predicted date and historical dates, a set of historical dates similar to the predicted date is determined; A backpropagation neural network is used to obtain the predicted daily load data based on the historical load data, the historical day set, the meteorological matrix, and the predicted daily data. The meteorological matrix is ​​composed of the historical daily data and the predicted daily data corresponding to the divided indicators. The determination of the principal components and corresponding principal component comprehensive scores in the index using a principal component analysis method based on an improved entropy method on the matrix composed of the historical daily data includes: Generate a complete historical day data matrix based on the filled-in historical day data; The complete historical day data matrix is ​​normalized based on the mean of the column vectors in the matrix. Calculate the covariance matrix of the normalized matrix, as well as the eigenvalues ​​and eigenvectors of the covariance matrix; The principal components in the index are determined based on the cumulative variance contribution rate of the eigenvalues. Calculate the principal component score based on the feature vector and the normalized matrix; The principal component comprehensive score is determined by a weighted sum of the principal component entropy value and the principal component sample weight, wherein the principal component entropy value and the principal component sample weight are determined based on the principal component score; The step of calculating the meteorological similarity between the predicted day and the historical day based on the historical daily data and predicted daily data corresponding to the indicators divided by the principal component comprehensive score includes: Based on the principal component composite score, the indicators are divided into important indicators and general indicators; Based on the historical daily data and the predicted daily data, obtain the maximum and minimum values ​​of the historical daily data for the historical days and the predicted days under important indicators; Based on the historical day data and the predicted day data, obtain the average value of the historical day data and the predicted day data under general indicators; The meteorological similarity between the historical days and the predicted days is calculated using Euclidean distance on the meteorological matrix composed of the average value, the maximum value, and the minimum value.

2. The baseline load forecasting method according to claim 1, characterized in that, The step of calculating the imputation values ​​for missing data based on samples in the matrix using the KNN and DBSCAN algorithms includes: The neighborhood radius and threshold of the DBSCAN algorithm are determined based on the data of the samples in the pre-impregnated matrix that are most relevant to the samples containing missing data. Outliers in the data are determined based on the neighborhood radius and the threshold. The KNN algorithm is used to calculate the imputation values ​​for missing data in the samples after outlier removal.

3. The baseline load forecasting method according to claim 1, characterized in that, The step of determining the principal component comprehensive score based on the weighted sum of principal component entropy values ​​and principal component sample weights, wherein the principal component entropy values ​​and principal component sample weights are determined based on the principal component scores, includes: The principal components corresponding to the samples in the normalized matrix are normalized to generate the corresponding principal component sample weights. Calculate the entropy value of the principal component based on the proportion of the principal component samples; The weights of the principal components are calculated based on the entropy values. The principal component comprehensive score is determined by the weighted sum of the principal component weights and the proportions of the principal component samples.

4. The baseline load forecasting method according to claim 1, characterized in that, The step of determining the set of historical days similar to the predicted day based on the meteorological similarity and the time similarity between the predicted day and historical days includes: Time similarity is determined based on the seasonal and date similarity between the predicted date and historical dates; The daily similarity between each historical day and the predicted day is calculated based on the dot product of the meteorological similarity and the temporal similarity. The daily similarity is sorted, and a set of historical days similar to the predicted day is generated by selecting the first preset number of historical days.

5. A baseline load forecasting device, characterized in that, The device includes: The acquisition module is used to acquire historical load data and historical daily data and predicted daily data determined by indicators, wherein the indicators include temperature, humidity, air pressure, precipitation, irradiance and wind speed. The first determining module is used to determine the principal components and corresponding principal component comprehensive scores in the index by using a principal component analysis method based on entropy value improvement on the matrix composed of the historical daily data. The first calculation module is used to calculate the meteorological similarity between the predicted day and the historical day based on the historical daily data and predicted daily data corresponding to the indicators divided by the principal component comprehensive score; The second determining module is used to determine a set of historical days similar to the predicted day based on the meteorological similarity and the time similarity between the predicted day and historical days; The module is used to obtain predicted daily load data by using a backpropagation neural network based on the historical load data, the historical day set, the meteorological matrix, and the predicted day data. The meteorological matrix is ​​composed of the historical day data and predicted day data corresponding to the divided indicators. Also includes: The system comprises the following modules: a generation module for generating a historical daily data matrix, where each row vector represents a sample, and each sample includes the historical daily data corresponding to the indicator at the target time; and a column vector for including the historical daily data of an indicator at all times. A standardization module is used to standardize the historical daily data matrix. A pre-impact module is used to pre-impact missing data based on the average value of the historical daily data in the column vectors of the standardized matrix. A third determination module is used to determine the most relevant samples in the pre-impacted matrix to the samples containing missing data, based on a grey relational algorithm. A second calculation module is used to calculate the imputation values ​​for the missing data based on the samples in the matrix using the KNN and DBSCAN algorithms. The first determining module includes: a first generation unit, used to generate a complete historical day data matrix based on the imputed historical day data; a normalization processing unit, used to normalize the complete historical day data matrix based on the mean of the column vectors in the complete historical day data matrix; a second calculation unit, used to calculate the covariance matrix of the normalized matrix, as well as the eigenvalues ​​and eigenvectors of the covariance matrix; a third determining unit, used to determine the principal components in the index based on the cumulative variance contribution rate of the eigenvalues; a third calculation unit, used to calculate the principal component scores based on the eigenvectors and the normalized matrix; and a fourth determining unit, used to determine the principal component comprehensive score based on the weighted sum of the principal component entropy value and the principal component sample weight, wherein the principal component entropy value and the principal component sample weight are determined based on the principal component score. The step of calculating the meteorological similarity between the predicted day and the historical day based on the historical daily data and predicted daily data corresponding to the indicators divided by the principal component comprehensive score includes: Based on the principal component composite score, the indicators are divided into important indicators and general indicators; Based on the historical daily data and the predicted daily data, obtain the maximum and minimum values ​​of the historical daily data for the historical days and the predicted days under important indicators; Based on the historical day data and the predicted day data, obtain the average value of the historical day data and the predicted day data under general indicators; The meteorological similarity between the historical days and the predicted days is calculated using Euclidean distance on the meteorological matrix composed of the average value, the maximum value, and the minimum value.

6. An electronic device comprising a processor, a communication interface, a memory, and a communication bus, wherein, The processor, the communication interface, and the memory communicate with each other via the communication bus, characterized in that... The memory is used to store computer programs; The processor is configured to perform the method steps of any one of claims 1 to 4 by running the computer program stored in the memory.

7. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, wherein the computer program, when executed by a processor, implements the steps of the method described in any one of claims 1 to 4.

Images