An electricity consumption prediction method and terminal based on hypergraph images

Abstract

The application discloses a kind of based on hyperimage electricity consumption prediction method and terminal, influence factor sequence in electricity consumption dataset is mapped as voxel coordinate, electricity consumption sequence is mapped as voxel value, and the hyperimage corresponding to multivariate time series is constructed;Based on the time delay estimation algorithm of hyperimage smoothness, the time delay between time series is estimated by minimizing the smoothness measure of hyperimage, and the alignment of electricity consumption dataset is realized;For the electricity consumption dataset after time series alignment, electricity consumption prediction experiment is carried out.Therefore, by the time series alignment processing of electricity consumption dataset, the overall quality of dataset and the robustness of prediction model are significantly improved, and the accuracy and reliability of electricity consumption prediction are effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power consumption prediction, and in particular to a power consumption prediction method based on hypergraph and a terminal. BACKGROUND

[0002] Power consumption prediction, as an important part of energy management, is of great significance to the safe and stable operation of the power system. With the rapid development of China's economy, the demand for electricity is rising, and the contradiction between supply and demand is becoming increasingly prominent. Accurate prediction of electricity demand helps to optimize the allocation of power resources, improve the efficiency of the power system, and reduce operating costs. With the construction of the global energy internet and the widespread access of clean energy, power consumption prediction faces unprecedented challenges and opportunities.

[0003] For a long time, power consumption prediction has been a research hotspot in the power industry. It not only involves the safe and stable operation of the power system, but also is closely related to the operation efficiency of the power market, the formulation of energy policy, and environmental protection, etc. Traditional power consumption prediction methods, such as time series analysis and regression analysis, can meet the demand to some extent, but when faced with uncertain influencing factors such as climate change, market fluctuations, and policy adjustments, their prediction accuracy and adaptability often cannot meet the requirements of modern power systems. In recent years, with the rapid development of big data and deep learning technologies, power consumption prediction has entered a new stage of development. The application of these technologies enables power consumption prediction to more accurately capture the dynamic changes in power consumption and better cope with various complex situations. However, these methods still have some challenges in power consumption prediction, such as the alignment deviation between power consumption and influencing factors on the time axis due to the power consumption collection process, sensor delay, and changes in the environment for obtaining weather information. This situation can seriously damage the stable quantitative relationship between power consumption and influencing factors, affecting the overall reliability of the model and the accuracy of the prediction results. Therefore, there is an urgent need for a power consumption prediction method that can solve the above problems. SUMMARY

[0004] The technical problem to be solved by the present application is to provide a power consumption prediction method based on hypergraph and a terminal, which can improve the actual effect and reliability of power consumption prediction accuracy and prediction performance.

[0005] To solve the above technical problems, the technical scheme adopted by the present application is:

[0006] A power consumption prediction method based on hypergraph, comprising the steps of:

[0007] Collect a multi-element time series power consumption dataset, map the influence factor components of the power consumption dataset to voxel coordinates, map the power consumption components of the power consumption dataset to voxel value vectors, and construct a hyper-image corresponding to the time series according to the voxel coordinates and the voxel value vectors;

[0008] Construct a time delay estimation algorithm based on the smoothness measure of the hyper-image, estimate the time delay between the time series by minimizing the smoothness measure of the hyper-image, and align the power consumption dataset;

[0009] Use the aligned power consumption dataset for power consumption prediction.

[0010] To solve the above technical problems, another technical solution adopted by the present application is:

[0011] A power consumption prediction terminal based on a hyper-image, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements each step of the above-mentioned power consumption prediction method based on a hyper-image.

[0012] The present application has the following advantages: the influence factor sequence in the power consumption dataset is mapped to voxel coordinates, and the power consumption sequence is mapped to voxel values, and a hyper-image corresponding to the multi-element time series is constructed; a time delay estimation algorithm based on the smoothness of the hyper-image estimates the time delay between the time series by minimizing the smoothness measure of the hyper-image, and aligns the power consumption dataset; the power consumption dataset after time series alignment is used for power consumption prediction experiments. Therefore, by aligning the time series of the power consumption dataset, the overall quality of the dataset and the robustness of the prediction model are significantly improved, and the accuracy and reliability of the power consumption prediction are effectively improved. BRIEF DESCRIPTION OF DRAWINGS

[0013] Figure 1 A flowchart of a power consumption prediction method based on a hyper-image according to an embodiment of the present application;

[0014] Figure 2 A schematic diagram of constructing a hyper-image according to an embodiment of the present application;

[0015] Figure 3 A schematic diagram of a power consumption prediction process according to an embodiment of the present application;

[0016] Figure 4 A graph showing the change of power consumption and various influence factors in the dataset according to an embodiment of the present application;

[0017] Figure 5 The power consumption prediction results of each prediction model before and after the alignment of the dataset according to an embodiment of the present application;

[0018] Figure 6 FIG. 1 is a schematic diagram of an electricity consumption prediction terminal based on hyper-images according to an embodiment of the present application.

[0019] Label explanation:

[0020] 1. An electricity consumption prediction terminal based on hyper-images; 2. a memory; 3. a processor. DETAILED DESCRIPTION

[0021] To explain the technical content, purposes and effects of the present application in detail, the following will be described in combination with embodiments and the accompanying drawings.

[0022] Please refer to Figure 1 The embodiment of the present application provides an electricity consumption prediction method based on hyper-images, including the following steps:

[0023] Collecting a multi-element time series electricity consumption dataset, mapping the influence factor components of the electricity consumption dataset to voxel coordinates, mapping the electricity consumption components of the electricity consumption dataset to voxel value vectors, and constructing hyper-images corresponding to the time series according to the voxel coordinates and the voxel value vectors;

[0024] Constructing a time delay estimation algorithm based on the smoothness measure of the hyper-image, estimating the time delay between the time series by minimizing the smoothness measure of the hyper-image, and aligning the electricity consumption dataset;

[0025] Using the aligned electricity consumption dataset to perform electricity consumption prediction.

[0026] From the above description, the beneficial effects of the present application are as follows: the influence factor sequence in the electricity consumption dataset is mapped to voxel coordinates, and the electricity consumption sequence is mapped to voxel values, and hyper-images corresponding to the multi-element time series are constructed; a time delay estimation algorithm based on the smoothness of the hyper-image is used to estimate the time delay between the time series by minimizing the smoothness measure of the hyper-image, and the alignment of the electricity consumption dataset is realized; the time series aligned electricity consumption dataset is used for electricity consumption prediction experiment. Therefore, by aligning the time series of the electricity consumption dataset, the overall quality of the dataset and the robustness of the prediction model are significantly improved, and the accuracy and reliability of the electricity consumption prediction are effectively improved.

[0027] Further, the time delay estimation algorithm based on the smoothness measure of the hyper-image includes the following:

[0028] Constructing a hyper-image response graph based on a local neighborhood covariance matrix according to the hyper-image;

[0029] Measuring the smoothness of the hyper-image according to the F-norm of the hyper-image response graph.

[0030] As can be known from the above description, the hyper image response graph based on the local neighborhood covariance matrix is constructed on the basis of the hyper image, which is used for the smoothness strategy of the hyper image. In this way, the time sequence alignment of the data set based on the smoothness is facilitated.

[0031] Further, the hyper image response graph based on the local neighborhood covariance matrix is constructed according to the hyper image, comprising:

[0032] A neighborhood window is taken as the center of each voxel in the hyper image, the neighborhood window is moved along a direction, the overall change of the voxel value at the same relative position before and after the movement is counted, and the deviation accumulation value of the voxel value is obtained;

[0033] The local neighborhood covariance matrix of the voxel is determined according to the deviation accumulation value of the voxel value, and the trace of the local neighborhood covariance matrix is calculated to obtain the hyper image response graph.

[0034] As can be known from the above description, the hyper image response graph is a hyper image with the same resolution as the hyper image, and the voxel value is the trace of the local neighborhood covariance matrix. That is, there is a local neighborhood covariance matrix for each voxel in the original hyper image, then the trace of each local neighborhood covariance matrix is calculated, and the result is recorded in the voxel position corresponding to the hyper image response graph. In order to facilitate the calculation of the smoothness of the hyper image.

[0035] Further, the smoothness of the hyper image is measured according to the F-norm of the hyper image response graph, comprising:

[0036]

[0037] In the formula, The smoothness of the hyper image is measured, H d The hyper image response graph is represented, The voxel coordinate corresponding to the influence factor of the i-th moment of the hyper image is represented, and d represents the time delay vector.

[0038] As can be known from the above description, by calculating the smoothness, after minimizing the smoothness measure of the hyper image, the time delay between the power consumption and the influence factor is obtained, so as to eliminate the alignment deviation between the multiple time series, improve the data quality in the data set, and improve the accuracy of the power consumption prediction.

[0039] Further, the time delay estimation algorithm is constructed based on the smoothness measure of the hyper image, comprising:

[0040] The components of the time delay vector are estimated one by one based on the smoothness measure of the hyper image through serial search:

[0041]

[0042] In the formula, denotes an estimated value of a time delay component of the mth influencing factor, d m denotes a time delay component of the mth influencing factor, M denotes the number of influencing factors;

[0043] The serial search algorithm for time delay vector estimation searches and updates each component of the time delay vector.

[0044] As can be seen from the above description, the serial search algorithm for time delay vector estimation searches and updates each component of the time delay vector, which can effectively improve the estimation efficiency of the time delay vector.

[0045] Further, the power consumption prediction using the aligned power consumption dataset comprises:

[0046] The aligned power consumption dataset is divided into a first training set, a first validation set and a first test set, and the first training set and the first validation set are used for model training;

[0047] The power consumption prediction model is generated according to the trained model and the first test set;

[0048] The first prediction result is calculated by the power consumption prediction model.

[0049] As can be seen from the above description, the data quality of the power consumption dataset can be improved by aligning the power consumption dataset, and the prediction performance of the power consumption prediction model can be improved based on this.

[0050] Further, the power consumption prediction using the aligned power consumption dataset comprises:

[0051] The power consumption prediction using the unaligned power consumption dataset is performed to obtain a second prediction result;

[0052] The power consumption prediction accuracy of the aligned power consumption dataset is determined by comparing the first prediction result and the second prediction result.

[0053] As can be seen from the above description, the power consumption prediction experiments are performed before and after the time series alignment of the power consumption dataset, and the prediction effect of the prediction model before and after the time series alignment is compared to verify the effectiveness of the alignment operation in improving the prediction effect and accuracy.

[0054] Further, the influencing factor component of the power consumption dataset is mapped to a voxel coordinate, and the power consumption component of the power consumption dataset is mapped to a voxel value vector, comprising:

[0055] The influencing factor component of the power consumption dataset is normalized to the size range required for constructing a hyper image, and the power consumption component of the power consumption dataset is normalized to 0-255;

[0056] projecting the influence factor component as a voxel coordinate of the hyper image, and projecting the electricity consumption component corresponding to the voxel coordinate as a voxel value vector of the hyper image.

[0057] From the above description, it can be known that the projection of the voxel coordinate and the voxel value vector is facilitated by the component normalization processing.

[0058] Further, projecting the electricity consumption component corresponding to the voxel coordinate as a voxel value vector of the hyper image comprises:

[0059] The voxel value vector is assigned to the corresponding voxel coordinate by combining the interpolation method.

[0060] From the above description, it can be known that after the normalization of the influence factor component, there is a non-integer, and therefore the interpolation method is needed to map the corresponding electricity consumption vector to the corresponding voxel coordinate, so as to improve the data mapping accuracy.

[0061] Please refer to Figure 6 Another embodiment of the present application provides a power consumption prediction terminal based on a hyper image, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements each step of the above-mentioned power consumption prediction method based on a hyper image when executing the computer program.

[0062] The above-mentioned power consumption prediction method and terminal based on a hyper image of the present application are suitable for improving the actual effect and reliability of power consumption prediction accuracy and prediction performance, which will be described below through specific embodiments:

[0063] Please refer to Figures 1 to 5 An embodiment of the present application is:

[0064] A power consumption prediction method based on a hyper image, characterized in that it comprises the following steps:

[0065] S1, collecting a multivariate time series electricity consumption dataset, mapping the influence factor component of the electricity consumption dataset as a voxel coordinate, mapping the electricity consumption component of the electricity consumption dataset as a voxel value vector, and constructing a hyper image corresponding to the time series according to the voxel coordinate and the voxel value vector.

[0066] In the embodiment, suitable influence factor components in the power consumption data set are mapped to voxel coordinates, and the power consumption components are mapped to the voxel value vectors of the hyper image, a hyper image corresponding to the multivariate time series is constructed, and the spatial hierarchy information of the time series is recorded in the hyper image. The multivariate time series is composed of the power consumption sequence and the influence factor sequence thereof. The construction process of the hyper image adopts the component projection and interpolation method. The selection of suitable influence factors in the power consumption data set needs to consider the correlation between each influence factor and the power consumption. The information gain of each influence factor on the power consumption is calculated, and suitable influence factors are selected according to the size of the information gain.

[0067] Specifically, first of all, due to the delay of the sensor or the change of the environment, there is an inevitable alignment deviation on the time axis between the power consumption and the influence factors. At this time, the power consumption prediction task can be represented as follows:

[0068]

[0069] In the formula, N represents the number of model training samples; y i represents the true measurement value of the power consumption; x m,i represents the historical data of the mth influence factor, M represents the number of influence factors, represents the parameter of the time delay embedding model, fe(·) represents that the time delay embedding model is usually approximated by a deep learning model, and represents the deviation measure of y i and its estimated value

[0070] y j represents the measurement value of the power consumption at time j, L0 represents the prediction length, and i, j I represent time indexes; x m,j represents the measurement value of the mth influence factor at time j, L m represents the historical data length, d = [d1, …, d M ] T , d Z represents a time delay vector for parameterizing the alignment deviation between the power consumption and the influence factors, and d m (m 1, 2, …, M) represents the time delay component of the mth influence factor.

[0071] The key to solving the time series alignment deviation problem lies in accurate time delay estimation. The common time delay estimation method is to construct a time delay measure based on the time sequence information of the time series, but this kind of method ignores the spatial hierarchy information of the time series, that is, similar influence factors correspond to similar prediction targets. Therefore, the present application constructs a hyper image based on the spatial hierarchy information of the time series, constructs a time delay measure based on the smoothness of the hyper image, and realizes the alignment of the multivariate time series. ​​

[0072] The influence factor component in the multivariate time series is mapped to the voxel coordinate, and the electricity consumption component is mapped to the voxel value vector, to construct a multi-channel hyper image MH corresponding to the multivariate time series: That is,

[0073]

[0074] In the formula, and h i respectively represent the voxel coordinate corresponding to the influence factor at time i of the hyper image and the voxel value vector at time i.

[0075] And,

[0076]

[0077] In the formula, and h i respectively represent the mapping values of the influence factor component and the electricity consumption component; represents the set of the range of each influence factor after normalization, which is a non-negative integer;

[0078] From the above formula, it can be seen that the electricity consumption and the mth influence factor are normalized to the range of [0, 255] and [0, W m ], that is, the hyper image is an 8-bit special image with a resolution of , that is, the resolution of the hyper image is W1×W2×…×W m . It should be noted that since the voxel coordinate after normalization will have a non-integer, it is necessary to allocate the corresponding target value vector h i to the appropriate voxel position through an interpolation method. Among them, the voxel with coordinates is called a voxel wherein the dimension of the voxel value vector of the hyper image is equal to the prediction length, and after normalization of the influence factor component, there will be a non-integer, so an interpolation method is needed to map the corresponding electricity consumption vector to the corresponding voxel coordinate. The interpolation method can select a forward interpolation method or a nearest neighbor interpolation method.

[0079] In this embodiment, the prediction of electricity consumption is realized by constructing a time delay embedding model, and a deep learning model is used to approximate the time delay embedding model. Therefore, the voxel value vector of the hyper image directly corresponds to the electricity consumption component in the multivariate time series, specifically the output value vector of the prediction model. Due to the existence of this corresponding relationship, the dimension of the voxel value vector of the hyper image is consistent with the prediction length. To simplify the calculation and facilitate the fast search of the time delay, the voxel value vector of the hyper image is simplified, and the dimension of the voxel value vector of the hyper image is reduced to 1, that is, the voxel value vector degenerates into a scalar.

[0080] At the same time, {y i} of the low-dimensional projection and {y i+1} can effectively estimate the time delay between and {y i}, so that the measure of the time delay vector can be constructed by the local smoothness of the single-channel hyperimage SH: satisfies:

[0081]

[0082] S2, based on the smoothness measure of the hyperimage, a time delay estimation algorithm is constructed, and the alignment of the time series of the power consumption data set is estimated by minimizing the smoothness measure of the hyperimage.

[0083] wherein, after the hyperimage is constructed, a neighborhood window is taken as the center of each voxel in the hyperimage, the window is moved in a certain direction, the total change of the voxel value at the same relative position before and after the movement is counted, and so on, to obtain the local neighborhood covariance matrix of each voxel, and the trace of the local neighborhood covariance matrix of each voxel of the hyperimage is calculated to obtain the hyperimage response graph; the F-norm of the hyperimage response graph is calculated, and the F-norm is used as a standard for measuring the smoothness of the hyperimage, so as to realize the measurement of the smoothness of the hyperimage. Wherein, the hyperimage response graph is a hyperimage with the same resolution as the hyperimage, and the voxel value is the trace of the local neighborhood covariance matrix. That is, there is a local neighborhood covariance matrix for each voxel in the original hyperimage, then the trace of each local neighborhood covariance matrix is calculated, and the result is recorded in the corresponding voxel position of the hyperimage response graph.

[0084] The spatial hierarchical information of the time series is manifested on the hyperimage as the hyperimage has the characteristics of maximum smoothness and minimum smoothness measure when the time series is aligned; by using the above characteristics, a time delay estimation algorithm based on the smoothness measure of the hyperimage is realized, that is, the alignment of the multivariate time series is realized by minimizing the smoothness measure of the hyperimage to estimate the time delay between the time series.

[0085] wherein, the time delay estimation algorithm based on the smoothness measure of the hyperimage is a serial time delay estimation based on the smoothness measure of the hyperimage, and specifically includes:

[0086] In searching the time delay of a certain influencing factor component, fixed values are randomly assigned to the time delay components of the rest of the influencing factor sequence, and then a hypergraph image corresponding to the time sequence is constructed, the smoothness thereof is measured and recorded; when the time delay of the influencing factor component is the real time delay, the smoothness measurement of the hypergraph image obtains the minimum value; the same operation is performed for the time delay search of the rest of the influencing factor components, and finally the accurate time delay between all the influencing factor components and the electricity consumption component is obtained. According to the time delay results of each influencing factor and the electricity consumption, the time sequence alignment of the electricity consumption data set is performed.

[0087] Taking two influencing factor sequences A and B as an example: in searching the time delay between sequence A and the electricity consumption, fixed values are assigned to the time delay components of sequence B, then the time delay component k of sequence A is continuously adjusted within a certain time delay range A , a hypergraph image under different time delays is constructed, the smoothness thereof is measured and recorded, and the time delay component k A that makes the smoothness measurement minimum is obtained; after k A is obtained, the alignment deviation between sequence A and the electricity consumption is eliminated, then the time delay component k B of sequence B is continuously searched according to the above steps, and finally k A and k B are obtained.

[0088] It should be noted that after the time sequence alignment of the electricity consumption data set is performed, a proper method should be selected for verification, which can be determined by the specific circumstances.

[0089] In this embodiment, the time delay estimation based on the smoothness measurement of the hypergraph image mainly includes the construction of the smoothness measurement of the hypergraph image and the design of the time delay estimation algorithm, and the specific steps are as follows:

[0090] (1) The smoothness measurement of the hypergraph image is constructed, specifically including:

[0091] In order to measure the smoothness of the hypergraph image, a hypergraph smoothness response map based on the local neighborhood covariance matrix is proposed.

[0092] Taking a neighborhood window centered on each voxel in the hypergraph image, moving the window along a certain direction, and counting the overall change of the voxel values at the same relative position before and after the movement, that is:

[0093]

[0094] In the formula, E(Δg) represents the deviation accumulation of the voxel value, Δg = [Δg1, …, Δg M ] represents the offset of the window, represents the window function, and in this embodiment, the Gaussian window function is taken. In order to simplify the formula, the Taylor expansion of in the calculation formula of the deviation accumulation of the voxel value is carried out and brought back into the formula, and finally the formula is obtained as follows:

[0095]

[0096] In the formula, yes The partial derivative, i.e., voxels in the super-image Gradient in each direction, voxels The local neighborhood covariance matrix has the following specific form:

[0097]

[0098] Define the response map H of the superimage SH. d :

[0099]

[0100] In the formula, tr(·) represents the trace of the matrix; λ m Representation matrix eigenvalues, λ m ≥0, m=1,…,M; i.e., response graph H d For a superimage with the same resolution as the superimage SH, and whose voxels... The value is The traces.

[0101] In this embodiment, it is based on the response graph H. d The F-norm measures the smoothness of the hyperimage. Let... To represent the smoothness measure of a hyperimage, we have:

[0102]

[0103] (2) Design of the time delay estimation algorithm, specifically including:

[0104] When the influencing factor sequence and the electricity consumption sequence are aligned, the corresponding hyperimage of the time series is the smoothest. Therefore, the time delay vector between the time series can be estimated by minimizing the smoothness metric of the hyperimage.

[0105]

[0106] When estimating the time delay vector d, it is necessary to simultaneously search for each time delay component d in the time delay space. m Finding the optimal value is computationally complex. Therefore, in this embodiment, a component dt of the time delay vector d is randomly selected. m Then, assign fixed values ​​to the remaining components of d randomly. mThe smoothness metric obtains a minimum value with the real time delay. Therefore, in order to effectively improve the estimation efficiency of the time delay vector, the embodiment adopts a serial search strategy to estimate the components of the time delay vector d one by one, that is, a serial search algorithm for time delay vector d estimation, as follows:

[0107]

[0108] The serial search algorithm for time delay vector d estimation is used to search and update each component of the time delay vector d, so as to realize the estimation of the time delay vector d. In the search, d m Generally, it starts from 0 and takes different values in a suitable time delay search range.

[0109] S3, using the aligned power consumption dataset to perform power consumption prediction.

[0110] Specifically, a plurality of prediction models are selected, and the parameters of each model are optimized, and the input and output parameters of the model are configured; in addition, the historical data window size and the prediction step of the training sample are determined, and the sampling frequency is adjusted.

[0111] As shown in Figure 3 , the dataset is divided into three parts of training, validation and test sets. The full influencing factors and power consumption sequence in the dataset are used to perform power consumption prediction experiments; by comparing the prediction effects before and after the time series alignment, the actual effect and reliability of the proposed time series alignment method on the prediction performance and accuracy are evaluated and verified.

[0112] In the embodiment, by comparing the prediction effects before and after the time series alignment, the improvement degree of the alignment method on the prediction performance can be quantified directly. Specifically, if the prediction error after alignment is significantly reduced (such as the reduction of MSE, MAE and the increase of R2), it indicates that the proposed alignment method effectively eliminates the time delay deviation in the original data and improves the quality of the dataset. This improvement in data quality can directly improve the time series alignment relationship between the influencing factors and the power consumption, so that the prediction model can more accurately capture the internal rules between variables, and finally improve the accuracy of power consumption prediction and the overall prediction ability of the model.

[0113] In the embodiment, the purpose of selecting multiple prediction models is to compare the power consumption prediction performance of different prediction models to ensure the effectiveness and reliability of the experimental results; the prediction conditions of the dataset before and after the time series alignment should be the same; it should be noted that the historical data window size, prediction length and the like of the model are not unique and can be adjusted according to the specific circumstances.

[0114] The embodiment will be described in combination with actual application as follows:

[0115] The research data set of this embodiment is from the electricity consumption data of Fujian Province and the local meteorological factor data, with a duration from March 1, 2019 to March 31, 2023. Among them, the data set includes electricity consumption, daily average temperature, daily average humidity, daily average wind speed and daily average apparent temperature, with a sampling interval of one day. Figure 4 The variation images of the electricity consumption and each influencing factor in the data set are drawn.

[0116] After obtaining the electricity consumption data set, data preprocessing is first performed. Then, the time series alignment is performed on the data set by using the time delay estimation algorithm based on the hypergraph image smoothness measurement. The data preprocessing includes two key steps: identifying and removing obvious outliers in the data set and processing missing items in the data set. Table 1 shows the time delay between each influencing factor and electricity consumption obtained by the time delay estimation algorithm proposed in this embodiment.

[0117] Table 1 Time delay of influencing factors on electricity consumption

[0118] Daily average temperature Daily average humidity Daily average wind speed Daily average apparent temperature Time lag 18 4 3 17

[0119] After obtaining the time delay of each influencing factor and electricity consumption, the time series alignment is performed on the electricity consumption data set according to the specific time delay of each influencing factor.

[0120] The same operation is performed on the electricity consumption data set before and after the time series alignment, and the data set is divided into training, validation and test sets. The last three months of data are used for testing, and the remaining data is divided into 80% and 20% for training and validation of the model, respectively. At the same time, the historical data window size of the prediction model input data is set to 20 days, the sampling frequency is 1 day, and the prediction step is 5 days.

[0121] Among them, the selected prediction models include Transformer, PatchTST, TimesNet, iTransformer, MICN and Crossformer.

[0122] In this embodiment, the mean squared error (MSE), mean absolute error (MAE), normalized root mean square error (NRMSE) and R-squared (R 2 ) are selected as error indicators to evaluate the electricity consumption prediction effect of the prediction model before and after the data set alignment. Among them, the smaller the first three indicators, the higher the prediction accuracy, and the larger the R 2 , the better the model fitting effect. At the same time, in order to intuitively show the prediction effect, Figure 5The prediction results of different models on the test set before and after the alignment of the power consumption dataset are shown.

[0123] The evaluation results are shown in Table 2, where the bold font in MSE, MAE and NRMSE is a smaller value, and the bold font in R 2 The results in Table 2 show that, except for the MAE error indicator of the Transformer, the top three error indicators of all models on the power consumption dataset after time series alignment are smaller than those before time series alignment; after time series alignment, the R 2 value is higher than the error indicator R 2 before alignment. This shows that after the time series alignment operation, the error of the power consumption prediction model is smaller, the fitting effect of the model is better, and the prediction accuracy is higher.

[0124] Error indicators of each prediction model before and after alignment of the dataset

[0125]

[0126] The power consumption prediction method based on hypergraph image proposed in this embodiment mainly includes mapping the power consumption component in the power consumption dataset to a voxel value vector and mapping the influencing factor component to a voxel coordinate to construct a hypergraph corresponding to the time series. Then, a smoothness measure of the hypergraph is constructed, and a time delay measure is further constructed based on the smoothness measure to obtain an accurate time delay estimation between the influencing factor and the power consumption, and then the alignment of the power consumption dataset is realized. On this basis, multiple prediction models are used to realize the prediction of the power consumption. Through the alignment operation of the time series, the data quality of the dataset is effectively improved, and the prediction accuracy of the power consumption prediction model is improved.

[0127] Please refer to Figure 6 Embodiment two of the present application is:

[0128] A power consumption prediction terminal 1 based on a hypergraph, comprising a memory 2, a processor 3, and a computer program stored in the memory 2 and executable on the processor 3, wherein the processor 3 implements each step of the power consumption prediction method based on the hypergraph of embodiment one when executing the computer program.

[0129] In summary, the application provides a kind of electricity consumption prediction method and terminal based on hypergraph, belongs to electricity consumption prediction field. Including: the influence factor sequence in electricity consumption dataset is mapped to voxel coordinate, electricity consumption sequence is mapped to voxel value, and hypergraph corresponding to multivariate time series is constructed;On the basis of hypergraph, the hypergraph response graph based on local neighborhood covariance matrix is constructed, which is used for the smoothness measurement of hypergraph;Design time delay estimation algorithm based on hypergraph smoothness, estimate the time delay between time series by minimizing the smoothness measurement of hypergraph, realize the alignment of electricity consumption dataset;Respectively, before and after time series alignment, electricity consumption dataset, electricity consumption prediction experiment is carried out, and the prediction of electricity consumption is realized.The application improves the overall quality of the dataset and the robustness of the prediction model by aligning the time series of the electricity consumption dataset, effectively improves the accuracy and reliability of the electricity consumption prediction.

[0130] The above is only an embodiment of the application, and does not limit the patent scope of the application, and any equivalent transformation or direct or indirect application in related technical fields using the content of the specification and drawings is also included in the patent protection scope of the application.

Claims

1. A method for predicting electricity consumption based on hyperimages, characterized in that, Including the following steps: Collect a multivariate time series of electricity consumption dataset, map the influencing factor components of the electricity consumption dataset to voxel coordinates, map the electricity consumption components of the electricity consumption dataset to voxel value vectors, and construct a hyperimage corresponding to the time series based on the voxel coordinates and the voxel value vectors. A time delay estimation algorithm is constructed based on the smoothness metric of the superimage. The algorithm estimates the time delay between time series by minimizing the smoothness metric of the superimage, and then aligns the electricity consumption dataset. Use the aligned electricity consumption dataset to predict electricity consumption; A time delay estimation algorithm is constructed based on the smoothness metric of the super-image, which includes the following steps: Construct a hyperimage response map based on the local neighborhood covariance matrix based on the hyperimage; The smoothness of the superimage is measured by the F-norm of the superimage response map; Constructing a hyperimage response map based on the local neighborhood covariance matrix according to the hyperimage includes: Take a neighborhood window centered on each voxel in the superimage, move the neighborhood window in one direction, and count the overall change of voxel values ​​at the same relative position within the window before and after the movement to obtain the cumulative deviation value of the voxel values. The local neighborhood covariance matrix of the voxel is determined based on the accumulated deviation value of the voxel value, and the super-image response map is obtained by calculating the trace of the local neighborhood covariance matrix. The smoothness of the superimage is measured based on the F-norm of the superimage response map, including: In the formula, H represents a measure of the smoothness of a super-image. d Represents the hyperimage response map. d represents the voxel coordinates corresponding to the influencing factors at time i of the hyperimage, and d represents the time delay vector.

2. The method for predicting electricity consumption based on hyperimages according to claim 1, characterized in that, A time delay estimation algorithm is constructed based on the smoothness metric of the super-image, including: Based on the smoothness metric of the hyperimage, the components of the time delay vector are estimated one by one through a serial search: In the formula, Indicates the first m Estimates of the time delay components of each influencing factor. d m Indicates the first m The time delay component of each influencing factor, M Indicates the number of influencing factors; A serial search algorithm for time delay vector estimation is used to search for and update each component of the time delay vector.

3. The method for predicting electricity consumption based on hyperimages according to claim 1, characterized in that, Electricity consumption prediction is performed using the aligned electricity consumption dataset, including: The aligned electricity consumption dataset is divided into a first training set, a first validation set, and a first test set. The first training set and the first validation set are used to train the model. A power consumption prediction model is generated based on the trained model and the first test set. The first prediction result is calculated using the electricity consumption prediction model.

4. The method for predicting electricity consumption based on hyperimages according to claim 3, characterized in that, Electricity consumption prediction is performed using the aligned electricity consumption dataset, followed by: The electricity consumption dataset before alignment is used to predict electricity consumption, and a second prediction result is obtained. The accuracy of electricity consumption prediction for the aligned electricity consumption dataset is determined by comparing the first prediction result and the second prediction result.

5. The method for predicting electricity consumption based on hyperimages according to claim 1, characterized in that, Mapping the influencing factor components of the electricity consumption dataset to voxel coordinates, and mapping the electricity consumption components of the electricity consumption dataset to voxel value vectors, includes: The influencing factor components of the electricity consumption dataset are normalized to the size range required to construct the hyperimage, and the electricity consumption components of the electricity consumption dataset are normalized to 0~255. The influencing factor components are projected into voxel coordinates of the superimage, and the electricity consumption component corresponding to the voxel coordinates is projected into a voxel value vector of the superimage.

6. The method for predicting electricity consumption based on hyperimages according to claim 1, characterized in that, Projecting the electricity consumption component corresponding to the voxel coordinates into a voxel value vector of the hyperimage includes: The voxel value vector is assigned to the corresponding voxel coordinates by combining interpolation.

7. A power consumption prediction terminal based on hyperimages, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements each step of the power consumption prediction method based on super-images according to any one of claims 1 to 6.

Images