A power energy consumption prediction method and system based on multi-dimensional data and a medium

Abstract

The application relates to the technical field of power energy consumption prediction, in particular to a power energy consumption prediction method and system based on multidimensional data and a medium, the method comprising the following steps: collecting power consumption data and various influence data of a power grid in real time; determining redundancy imbalance degrees of the various influence data in a time sequence interval; calculating mutual information dependency degrees of the various influence data in the time sequence interval; identifying each mutation point of the power consumption data in the time sequence interval; determining prediction effective contributions of the various influence data in the time sequence interval; combining the redundancy imbalance degrees and the prediction effective contributions to correct regression coefficients in a prediction process of the power consumption data of the power grid by using a PCR principal component regression algorithm. Therefore, the prediction accuracy of power energy consumption is improved.

Description

Technical Field

[0001] This application relates to the field of power energy consumption prediction technology, specifically to a power energy consumption prediction method, system, and medium based on multidimensional data. Background Technology

[0002] With the digital transformation of smart grids and the popularization of big data technology, accurate power consumption forecasting has become a core technology for improving energy utilization, ensuring the safe and stable operation of the power grid, and promoting sustainable economic development. Power consumption forecasting based on multidimensional data is based on technologies such as the Internet of Things, big data, and cloud computing. It acquires massive amounts of data from all aspects of the power system and has rich data dimensions for in-depth mining of power consumption change characteristics, so as to achieve accurate prediction of future power consumption.

[0003] Traditional methods of using Principal Component Regression (PCR) algorithms combined with multidimensional data to predict power consumption can capture the correlation patterns of energy consumption changes by relying on the correlation features between multidimensional data. However, the massive amount of raw multidimensional data also poses threats of sample imbalance and redundant noise. Specifically, the number of samples with normal operating conditions in the multidimensional sample data is much larger than that of low-frequency events, causing the principal component direction to be dominated by the majority class samples, while the unique correlation features existing in the minority class samples are ignored. At the same time, the noise in the raw multidimensional data can directly contaminate the calculation of the regression equation in the PCR algorithm, causing the prediction results to deviate from the actual power consumption data, resulting in the drawback of being unable to accurately predict power consumption. Summary of the Invention

[0004] To address the aforementioned technical problems, the purpose of this application is to provide a method, system, and medium for predicting power consumption based on multidimensional data. The specific technical solution adopted is as follows:

[0005] In a first aspect, embodiments of this application provide a method for predicting power consumption based on multidimensional data, the method comprising the following steps:

[0006] Real-time collection of power grid consumption data and its various impact data;

[0007] Analyze the distribution of all mutation points in the various impact data within the time series interval, and the differences in the distribution of mutation points between the various impact data and the remaining impact data, to determine the redundancy and imbalance of the various impact data within the time series interval.

[0008] By analyzing the impact of abrupt changes in various influencing data on electricity consumption data within a time series interval, and the cross-correlation between various influencing data and electricity consumption data, the mutual information dependence of various influencing data within a time series interval is calculated.

[0009] identify each mutation point of the power consumption data in the time interval; determine the prediction effective contribution of each influence data in the time interval based on the correlation degree in time between the mutation points of each influence data and the mutation points of the power consumption data in the time interval, and the trend similarity of each influence data at the same time as the mutation point of the power consumption data and the mutual information dependence.

[0010] Combine the redundancy imbalance degree and the prediction effective contribution to correct the regression coefficient in the process of predicting the power consumption data of the power grid by using the PCR principal component regression algorithm.

[0011] In one embodiment, the determination of the redundancy imbalance degree of each influence data in the time interval includes:

[0012] Calculate the dispersion degree of all mutation points in each influence data in the time interval, and multiply it by the distribution asymmetry degree of all mutation points in each influence data in the time interval to obtain the sample imbalance of each influence data in the time interval.

[0013] Calculate the mean of all mutation points in any influence data in the time interval, determine the cumulative result of the Js divergence between the mutation points in the any influence data and the mutation points in the remaining all influence data, map the mean to a positive number, and calculate the ratio of the cumulative result to the positive number as the noise redundancy degree of the any influence data in the time interval; wherein the number of types of influence data is greater than 2.

[0014] Determine the redundancy imbalance degree using the sample imbalance and the noise redundancy degree, wherein the redundancy imbalance degree is positively correlated with the sample imbalance and the noise redundancy degree of each influence data in the time interval.

[0015] In one embodiment, the distribution asymmetry degree is the absolute value of the skewness of the time sequence composed of all mutation points in each influence data in the time interval.

[0016] In one embodiment, the redundancy imbalance degree is the normalized result of the product of the sample imbalance and the noise redundancy degree of each influence data in the time interval.

[0017] In one embodiment, the calculation of the mutual information dependence of each influence data in the time interval includes:

[0018] Take the time range between each mutation point and the next mutation point in each influence data in the time interval as the short-term influence interval of each mutation point, and calculate the range value of the power consumption data in the short-term influence interval of each mutation point.

[0019] acquire the split threshold of all mutation points in the influence data in the time interval, calculate the difference between the mean of the range values corresponding to all mutation points greater than or equal to the split threshold and the mean of the range values corresponding to all mutation points less than the split threshold;

[0020] take the difference as the index of the exponential function with a natural constant as the base, calculate the mutual information entropy between the influence data and the power consumption data in the time interval, multiply the result of the calculation of the exponential function, and obtain the mutual information dependency of the influence data.

[0021] In one embodiment, the determination of the prediction effective contribution of the influence data in the time interval comprises:

[0022] calculate the metric distance between the time corresponding to all mutation points of the influence data and the time corresponding to all mutation points of the power consumption data in the time interval, and perform trend decomposition on the power consumption mutation points and the influence data at the same time in the time interval to acquire the trend item and the periodic item;

[0023] calculate the similarity between the trend item of the power consumption mutation point and the trend item of the influence data at the same time in the time interval, denoted as the first similarity, calculate the similarity between the periodic item of the power consumption mutation point and the periodic item of the influence data at the same time in the time interval, denoted as the second similarity, determine the sum of the first similarity and the second similarity, map the metric distance to a positive value, and calculate the ratio of the sum and the positive value as the trend-period correlation degree of the influence data in the time interval;

[0024] acquire the prediction effective contribution by using the trend-period correlation degree and the mutual information dependency.

[0025] In one embodiment, the prediction effective contribution is the normalized value of the product of the trend-period correlation degree and the mutual information dependency.

[0026] In one embodiment, the correction utilizes the regression coefficient in the process of predicting the power consumption data of the power grid by the PCR principal component regression algorithm.

[0027] calculate the ratio of the prediction effective contribution and the redundancy imbalance degree of the influence data in the time interval as the prediction weight adjustment factor of the influence data in the time interval, calculate the mean of the prediction weight adjustment factors of the influence data in all time intervals in a preset historical period, and multiply the original regression coefficient corresponding to the influence data to obtain the corrected regression coefficient of the influence data, wherein the length of the preset historical period is greater than the length of the time interval.

[0028] In a second aspect, the embodiments of the present application also provide a power energy consumption prediction medium based on multi-dimensional data, wherein the medium stores a computer program, and the computer program is executed by a processor to implement the steps of the method according to any one of the preceding aspects.

[0029] In a third aspect, the embodiments of the present application also provide a power energy consumption prediction system based on multi-dimensional data, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, and the processor implements the steps of the method according to any one of the preceding aspects when executing the computer program.

[0030] The present application has at least the following beneficial effects:

[0031] The present application can more accurately measure the redundancy and imbalance of different influence factors by analyzing the distribution difference of mutation points in each influence data, determine the redundancy and imbalance of each influence data, help to identify the contribution degree of influence data to power consumption prediction, thereby reducing the interference of redundant data, focusing more on the influence data that has a significant impact on the change of electricity consumption in power consumption prediction, avoiding overfitting phenomenon, and improving the generalization ability of predicting power consumption; by investigating the influence degree of mutation points on electricity consumption data and the mutual correlation between each influence data and electricity consumption data, the synergistic effect of each factor on power consumption can be more comprehensively understood, and the calculation of mutual information dependence helps to find the implicit relationship of influencing electricity consumption, ensuring that these implicit factors can be considered in the prediction process, thereby effectively improving the prediction accuracy; by analyzing the time correlation between each influence data and the mutation points of electricity consumption data, the dynamic changes in time can be better captured, the relative position of each mutation point on the time axis and its influence can be identified, the adaptability of the prediction algorithm to time series data is enhanced, the prediction process is more accurate; in combination with the redundancy and imbalance degree and the effective contribution to prediction, the calculation of the regression coefficient in the PCR principal component regression algorithm is corrected, thereby optimizing the performance of the algorithm, enhancing the response ability of the prediction algorithm to each influence data, making the prediction algorithm more accurately capture the multi-dimensional factors affecting power consumption, automatically adjusting the weight in the prediction process, improving the stability of electricity consumption prediction, reducing the error, more accurately reflecting the data quality of multi-dimensional data samples and the effective correlation of electricity consumption data prediction, reducing the risk of ignoring the effective features of a small number of samples in a large amount of data samples, and making the power consumption prediction result closer to the real power grid energy consumption data structure. BRIEF DESCRIPTION OF DRAWINGS

[0032] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiment or prior art description. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0033] Figure 1 A step flow chart of a power energy consumption prediction method based on multi-dimensional data provided by an embodiment of the present application. DETAILED DESCRIPTION

[0034] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined invention purpose, the following will combine the drawings and preferred embodiments to specifically describe the specific embodiments, structure, features and effects of the power energy consumption prediction method, system and medium based on multi-dimensional data according to the present application. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0035] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0036] The specific scheme of the power energy consumption prediction method, system and medium based on multi-dimensional data provided by the present application will be specifically described below in combination with the drawings.

[0037] Please refer to Figure 1 which shows a step flow chart of a power energy consumption prediction method based on multi-dimensional data provided by an embodiment of the present application. The method comprises the following steps:

[0038] S1, real-time collection of power consumption data and its various influence data of the power grid, and pre-processing.

[0039] This embodiment takes the power grid of a transformer area as an example to predict the power energy consumption. The power data and environmental data of the transformer area power grid are obtained in real time through the smart meters and various sensors deployed in the transformer area power grid, wherein the power data includes power supply, power consumption and power factor data, and the environmental data includes temperature, irradiance and wind speed data. The power data and environmental data are both collected synchronously, and the collection frequency is set to 1Hz. The implementer can set it according to the actual situation, and this embodiment does not limit it. The difference between the power supply and the power consumption of the transformer area power grid at each moment is taken as the transformer area loss power. The loss power, power supply, power factor, temperature, irradiance and wind speed data are all recorded as the influence data of the power consumption.

[0040] To prevent the abnormal values in the multi-dimensional original data collected in the electric energy consumption prediction process from interfering and the different data dimensions from affecting the subsequent analysis, the power consumption data of the transformer area power grid and each item of influence data are respectively taken as inputs, 3 criteria are used for denoising, and the missing value filling method is used for filling the missing values at the noise point removal positions. The Z-Scroe standardization is used for unification, and the preprocessing of the power consumption data and each item of influence data is completed. Among them, the 3 criteria, the median filling method and the Z-Scroe standardization all belong to the known technology, and the specific process will not be described again.

[0041] S2, the distribution state of all mutation points in each item of influence data in the time sequence interval is analyzed, and the distribution difference between the mutation points in each item of influence data and the mutation points in the remaining items of influence data is determined. The redundancy imbalance degree of each item of influence data in the time sequence interval is determined.

[0042] In the electric power energy consumption prediction process based on multi-dimensional data, the power consumption directly reflects the demand of the terminal user, and is the core basis for the power generation plan, load management and market transaction of the power system. The accurate prediction of the power consumption of the transformer area power grid can provide a basis for meeting the scheduling and planning of the smart power system. When the PCR principal component regression is used for power consumption prediction, the sample imbalance and redundant noise between each item of influence data will cause the deviation of the principal component direction and the important feature contribution of the minority class to be ignored, and finally cause the decline of the power consumption prediction accuracy.

[0043] Specifically, in the electric power energy consumption prediction process of the multi-dimensional data of the transformer area power grid, when the sample imbalance and redundant noise in each item of influence data in the multi-dimensional sample data are more serious, the influence of the sample imbalance and redundant noise on the electric energy consumption data prediction is more serious, the mutation condition of each item of influence data is more unstable, and the distribution skew of each item of influence data caused by the dominant influence of the majority class sample is more serious. At the same time, the data abnormal change amplitude in each item of influence data is low, and the change correlation between each item of influence data in different dimensions is low due to the redundant noise. At this time, the participation weight of the influence data with serious sample imbalance and redundant noise in the prediction process should be reduced.

[0044] Based on the above analysis, each 24h is divided as a time interval, and each time interval is taken as an example. Each item of influence data in the time interval is sequentially grouped into an influence data sequence, and a mutation point detection algorithm based on Bayes is used to obtain all mutation points in each item of influence data sequence. The sequence composed of all mutation points in each item of influence data sequence is recorded as a mutation sequence of each item of influence data. The length of the time interval can be set by the implementer according to the actual situation, and the mutation point detection algorithm based on Bayes is a known technology, and the implementer can select other available mutation point detection algorithms.

[0045] The dispersion degree of the mutation sequence of each item of influence data is calculated, and the dispersion degree is multiplied by the distribution asymmetry degree of the mutation sequence of each item of influence data to obtain a multiplication result as the sample imbalance of each item of influence data in the time interval. The dispersion degree can be calculated by variance, standard deviation, coefficient of variation, etc. In this embodiment, variance is used as the calculation method of the dispersion degree. The distribution asymmetry degree of the mutation sequence is calculated by the absolute value of the skewness of the mutation sequence. The implementer can select other available calculation methods of the distribution asymmetry degree of the mutation sequence, and this embodiment does not limit it.

[0046] Secondly, taking the i-th item of influence data as an example, the mean of the mutation sequence of the i-th item of influence data is calculated, the cumulative result of the Js divergence of the mutation sequence of the i-th item of influence data and the mutation sequence of all the remaining items of influence data is calculated, the mean is mapped to a positive number, and the ratio of the cumulative result to the positive number is calculated as the noise redundancy degree of the i-th item of influence data in the time interval. The purpose of mapping the mean to a positive number is to avoid the denominator being 0, which affects the calculation of the ratio. In this embodiment, the mean is mapped to a positive number by calculating the sum of the mean and a preset value greater than 0, and the sum is taken as the positive number. In this embodiment, the preset value greater than 0 is set to 0.01, and the implementer can set it according to the actual situation. The implementer can select other available mapping methods, and this embodiment does not limit it.

[0047] Further, the redundancy imbalance degree of each item of influence data in the time interval is calculated, which is used to represent the sample imbalance and the severity of redundant noise in each dimension of sample data in the power consumption prediction process. The specific expression is:

[0048] ; In the formula, is the redundancy imbalance degree of the i-th item of influence data in the time interval, is the sample imbalance of the i-th item of influence data in the time interval, is the noise redundancy degree of the i-th item of influence data in the time interval, and norm() is a normalization function, so that The range of its value is within the range of (0, 1).

[0049] It should be understood that redundancy imbalance reflects the sample imbalance and noise redundancy of each influencing data within a time series interval during the power consumption prediction process; sample imbalance reflects the instability of abrupt changes and the intensity of data distribution skewness of each influencing data within a time series interval during the power consumption prediction process; while noise redundancy characterizes the limited range of abnormal changes in each influencing data within a time series interval during the power consumption prediction process, as well as the correlation between changes in influencing data under different dimensions; in the power consumption prediction process based on multidimensional data, the more severe the sample imbalance and redundant noise in any influencing data in the multidimensional sample data, the higher the instability of abrupt changes in the influencing data, and the more obvious the data distribution skewness, that is, the larger the calculated sample imbalance; at the same time, the smaller the magnitude of abnormal changes in the influencing data caused by redundant noise, the lower the correlation between changes in influencing data under different dimensions, that is, the larger the calculated noise redundancy, resulting in a larger redundancy imbalance of the influencing data.

[0050] S3 calculates the mutual information dependency of various influencing data within the time series interval by considering the impact of abrupt changes in various influencing data on electricity consumption data and the cross-correlation between various influencing data and electricity consumption data.

[0051] In the process of predicting power consumption based on multidimensional data, when the regression coefficients in the PCR principal component regression algorithm are improved only by addressing the redundancy and imbalance of various influencing data that are correlated with the power consumption data, the contribution weight analysis between the various influencing data and the power consumption data is lacking. The predictive correlation between the power consumption data and different influencing data is ignored. Regression prediction based solely on the principal components extracted from PCR principal component regression will lead to the prediction data failing to meet the interaction effect between the various influencing data and the power consumption, resulting in inaccurate prediction results.

[0052] Specifically, in the process of power consumption prediction based on multidimensional data, when any influencing data of the power grid in a distribution area has a stronger prediction weight in the PCR principal component prediction regression of the electricity consumption data, the threshold effect of the influencing data is more obvious. That is, after the influencing data reaches a certain threshold, a slight change in the influencing data will cause a greater change in the electricity consumption data of the power grid in the distribution area compared to before the influencing data reaches the threshold. Moreover, the higher the degree of mutual information dependence between the influencing data and the electricity consumption data, the more the weight of the influencing data in the prediction process should be increased.

[0053] Based on the above analysis, taking the i-th impact data as an example, the time range between each mutation point in the i-th impact data in the time interval and the next mutation point is taken as the short-term impact interval of each mutation point, and the range value of the electricity consumption data in the short-term impact interval of each mutation point is calculated. The threshold segmentation algorithm of Otsu is used to obtain the segmentation threshold of the mutation sequence of the i-th impact data, the mean value of the range value of the electricity consumption data in the short-term impact interval of all mutation points of the i-th impact data greater than or equal to the segmentation threshold is calculated, and is recorded as the first mean value. The mean value of the range value of the electricity consumption data in the short-term impact interval of all mutation points of the i-th impact data less than the segmentation threshold is calculated, and is recorded as the second mean value. The threshold segmentation algorithm of Otsu is a known technology, and the implementer can select other feasible threshold segmentation algorithms, and this embodiment does not limit this.

[0054] The difference between the first mean value and the second mean value is calculated, and the difference is taken as the index of the exponential function with the natural constant as the base number, and the calculation result of the exponential function is taken as the threshold effect intensity of the i-th impact data on the electricity consumption; the greater the threshold effect intensity, the higher the corresponding change amplitude of the i-th impact data mutation condition reaching the segmentation threshold compared with the change amplitude before reaching the segmentation threshold. The mutual information entropy between the i-th impact data sequence and the electricity consumption data sequence in the time interval is calculated, and is multiplied by the threshold effect intensity of the i-th impact data on the electricity consumption to obtain the mutual information dependence of the i-th impact data.

[0055] The mutual information dependence of each impact data is calculated by using the same calculation method as the mutual information dependence of the i-th impact data.

[0056] S4, identifying each mutation point of the electricity consumption data in the time interval; based on the correlation degree between the mutation points in each impact data and the mutation points of the electricity consumption data in the time interval, and the trend similarity of each impact data at the same time of the electricity consumption mutation point and the mutual information dependence, the prediction effective contribution of each impact data in the time interval is determined.

[0057] The greater the time sequence correlation between the mutation condition of the impact data and the mutation condition of the electricity consumption, the stronger the trend and periodic correlation between the impact data and the electricity consumption data when the electricity consumption data mutates; at this time, the participation weight of the impact data in the prediction process should be increased.

[0058] All power consumption data in the time interval is arranged in time sequence to form a power consumption data sequence, and a mutation point detection algorithm based on Bayes is used to obtain each mutation point in the power consumption data sequence, and the time point of the power consumption mutation point is recorded as the main variable mutation time point. In this embodiment, an STL sequence decomposition algorithm is used to decompose the power consumption and the influence data at the main variable mutation time point, and the trend item and the periodic item of the power consumption and the influence data at the main variable mutation time point are obtained respectively. The STL sequence decomposition algorithm is a known technology, and the specific process is not described here.

[0059] The time points of all mutation points of the influence data in the time interval are arranged in time sequence to form a first mutation time sequence of the influence data, and the time points of all mutation points of the power consumption data in the time interval are arranged in time sequence to form a second mutation time sequence. The metric distance between the first mutation time sequence and the second mutation time sequence of the influence data is calculated. In this embodiment, the Euclidean distance is used to calculate the metric distance, and the implementer can select other available metric distance calculation methods, such as DTW distance, and this embodiment does not limit this.

[0060] Further, taking the i-th influence data as an example, the similarity between the trend item of all power consumption mutation points in the time interval and the trend item of the i-th influence data at the same time is calculated, which is recorded as the first similarity, and the similarity between the periodic item of all power consumption mutation points in the time interval and the periodic item of the i-th influence data at the same time is calculated, which is recorded as the second similarity. In this embodiment, the first similarity and the second similarity are calculated by using the cosine similarity, and the implementer can select other available similarity calculation methods, such as Pearson correlation coefficient.

[0061] The metric distance of the i-th influence data is mapped to a positive value, the sum of the first similarity and the second similarity of the i-th influence data is calculated, and the ratio of the sum to the positive value is calculated as the trend and periodic correlation degree of the i-th influence data in the time interval.

[0062] The purpose of mapping the metric distance to a positive value is to avoid the denominator being zero, which affects the calculation of the ratio. In this embodiment, the mapping method of the metric distance to a positive value is to calculate the sum of the metric distance and a preset value greater than 0, and the sum is taken as the positive value. In this embodiment, the preset value greater than 0 is set to 0.01, and the implementer can set it according to the actual situation. The implementer can select other available mapping methods, and this embodiment does not limit this.

[0063] Based on the above analysis, the prediction effective contribution of the influence data in the time interval is calculated, which is used to represent the mutual information threshold effect condition and the trend and periodic correlation degree between the influence data and the power consumption data respectively, and the specific expression is:

[0064] wherein, is the prediction effective contribution of the i-th influencing data in the time interval, is the mutual information dependency of the i-th influencing data in the time interval, is the trend cycle correlation of the i-th influencing data in the time interval, and norm() is a normalization function, so that the value of is within the range of (0, 1).

[0065] It should be understood that the prediction effective contribution reflects the obvious condition of the threshold effect and the trend cycle correlation between each influencing data and the power consumption data in the time interval; the mutual information dependency reflects the condition of the threshold effect and the mutual information dependency of each influencing data on the change of the power consumption data in the time interval; and the trend cycle correlation represents the time sequence correlation between each influencing data and the power consumption data in the time interval, and the trend cycle similarity between the influencing data and the power consumption mutation condition; in the process of predicting the power consumption by using the PCR principal component regression algorithm, when the prediction contribution weight of the influencing data on the power consumption data is stronger, the condition of the threshold effect of the influencing data on the power consumption data is more obvious, the mutual information dependency between the influencing data and the power consumption data is greater, i.e., the calculated mutual information dependency is greater; meanwhile, the time sequence correlation between the influencing data and the power consumption data is stronger, the trend cycle similarity between the influencing data and the power consumption mutation condition is higher, i.e., the calculated trend cycle correlation is greater, and finally the prediction effective contribution is greater.

[0066] S5, combining the redundancy imbalance and the prediction effective contribution, correcting the regression coefficient in the process of predicting the power consumption data of the power grid by using the PCR principal component regression algorithm.

[0067] In the process of predicting the power consumption based on the multi-dimensional data by using the PCR principal component regression algorithm, when the redundancy imbalance of any influencing data is more subtle, and the prediction effective contribution of the influencing data on the power consumption data as the prediction principal variable is stronger, it indicates that the sample imbalance condition and the noise redundancy degree in the influencing data are more blurred, the correlation contribution in the prediction process of the power consumption data as the prediction principal variable is greater, and the influencing weight in the prediction process of the power consumption data should be increased.

[0068] Therefore, the embodiment constructs a prediction weight adjustment factor, which is used to represent the weight adjustment degree of each influence data when predicting the power consumption data by using the PCR principal component regression algorithm. The prediction weight adjustment factor is obtained by the prediction effective contribution and the redundancy imbalance degree of each influence data. Specifically, the ratio of the prediction effective contribution and the redundancy imbalance degree of each influence data in the time interval is taken as the prediction weight adjustment factor of each influence data. The greater the prediction weight adjustment factor is, the higher the data quality of the influence data is, and the greater the prediction effective contribution weight of the principal variable data is.

[0069] In the embodiment, the time range of the current month and the previous month is taken as a preset historical period, the power consumption data of the substation power grid in the historical period is taken as a principal variable, i.e., a dependent variable, each influence data is taken as an independent variable and as an input of the PCR principal component regression algorithm, the covariance matrix of all the standardized independent variables is calculated by using the PCR principal component regression algorithm, the eigenvalue decomposition is performed, the Kaiser criterion is used to select the number of principal components, the first K characteristic vectors are extracted and the principal component scores are calculated, a linear regression model, i.e., a linear regression equation, is established, and the prediction result of the power consumption data of the substation power grid in the next time interval in the current historical period is obtained. K represents the number of selected characteristic vectors.

[0070] In the embodiment, the product of the mean of the prediction weight adjustment factor of each influence data corresponding to all time intervals in the historical period and the regression coefficient of each independent variable in the linear regression equation is taken as the corrected regression coefficient, the sum of the corrected regression coefficients of all independent variables in the linear regression equation is ensured to be 1 by using the distribution fitting method, and the value of K is obtained by using the cross-validation method. The PCR principal component regression, the Kaiser criterion and the cross-validation method are all known technologies, and details are not described herein.

[0071] Based on the same inventive concept as the above method, the embodiment of the present application further provides a power energy consumption prediction medium based on multi-dimensional data, wherein a computer program is stored in the medium, and the computer program is executed by a processor to implement the steps of any one of the above power energy consumption prediction methods based on multi-dimensional data.

[0072] Based on the same inventive concept as the above method, the embodiment of the present application further provides a power energy consumption prediction system based on multi-dimensional data, which comprises a memory, a processor and a computer program stored in the memory and running on the processor. When the processor executes the computer program, the steps of any one of the above power energy consumption prediction methods based on multi-dimensional data are implemented.

[0073] It should be noted that the above-mentioned order of the embodiments of the present application is only for description, and does not represent the advantages and disadvantages of the embodiments. And the above describes specific embodiments of the present application. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are also possible or can be advantageous.

[0074] Each of the embodiments in the specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the difference from other embodiments.

[0075] The above is only the preferred embodiment of the present application, and is not used to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for predicting power consumption based on multidimensional data, characterized in that, The method includes the following steps: Real-time collection of power grid consumption data and its various impact data; Analyze the distribution of all mutation points in the various impact data within the time series interval, and the differences in the distribution of mutation points between the various impact data and the remaining impact data, to determine the redundancy and imbalance of the various impact data within the time series interval. By analyzing the impact of abrupt changes in various influencing data on electricity consumption data within a time series interval, and the cross-correlation between various influencing data and electricity consumption data, the mutual information dependence of various influencing data within a time series interval is calculated. Identify each abrupt change point in the electricity consumption data within a time series interval; based on the temporal correlation between the abrupt change points in the various impact data within the time series interval and the abrupt change points in the electricity consumption data, as well as the trend similarity between the abrupt change points in the electricity consumption data and the various impact data at the same time, and the mutual information dependence, determine the predictive effective contribution of each impact data within the time series interval; By combining the aforementioned redundancy imbalance with the aforementioned effective contribution of prediction, the regression coefficients in the process of predicting the electricity consumption data of the power grid using the PCR principal component regression algorithm are corrected.

2. The power consumption prediction method based on multidimensional data as described in claim 1, characterized in that, The determination of the redundancy imbalance of various influencing data within the time series interval includes: The dispersion of all abrupt change points in the various impact data within the time series interval is calculated and multiplied by the asymmetry of the distribution of all abrupt change points in the various impact data within the time series interval to obtain the sample imbalance of the various impact data within the time series interval. Calculate the mean of all mutation points in any one of the influencing data within the time series interval, determine the cumulative result of the Js divergence between the mutation point in the one influencing data and the mutation points in all other influencing data, map the mean to a positive number, and calculate the ratio of the cumulative result to the positive number as the noise redundancy degree of the one influencing data within the time series interval; wherein, the number of types of influencing data is greater than 2. The redundancy imbalance is determined by using the sample imbalance and the noise redundancy, wherein the redundancy imbalance is positively correlated with the sample imbalance and noise redundancy of each influencing data within the time series interval.

3. The power consumption prediction method based on multidimensional data as described in claim 2, characterized in that, The degree of distribution asymmetry is the absolute value of the skewness of the time series composed of all mutation points in the various influencing data within the time series interval.

4. The power consumption prediction method based on multidimensional data as described in claim 2, characterized in that, The redundancy imbalance is the normalized result of the product of the sample imbalance of each influencing data within the time series interval and the degree of noise redundancy.

5. The power consumption prediction method based on multidimensional data as described in claim 1, characterized in that, The calculation of the mutual information dependency of each influencing data within the time series interval includes: The time range between each mutation point and its next mutation point in the various impact data within the time series interval is taken as the short-term impact interval of each mutation point, and the range of electricity consumption data within the short-term impact interval of each mutation point is calculated. Obtain the segmentation threshold of all mutation points in the various impact data within the time series interval, and calculate the difference between the mean of the range values ​​corresponding to all mutation points greater than or equal to the segmentation threshold and the mean of the range values ​​corresponding to all mutation points less than the segmentation threshold. The difference is used as the exponent of an exponential function with the natural constant as the base. The mutual information entropy between the various influencing data and the electricity consumption data within the time series interval is calculated, and multiplied by the calculation result of the exponential function to obtain the mutual information dependence of the various influencing data.

6. The power consumption prediction method based on multidimensional data as described in claim 1, characterized in that, The determination of the predictive effectiveness contribution of each influencing data within the time series interval includes: Calculate the metric distance between the times of all abrupt change points corresponding to various impact data within the time series interval and the times of all abrupt change points corresponding to electricity consumption data; perform trend decomposition on the electricity consumption abrupt change points and their respective impact data at the same time within the time series interval to obtain trend terms and periodic terms; The similarity between the trend term of the electricity consumption mutation point within the time series interval and the trend term of each influencing data at the same time is calculated and denoted as the first similarity. The similarity between the periodic term of the electricity consumption mutation point within the time series interval and the periodic term of each influencing data at the same time is calculated and denoted as the second similarity. The sum of the first similarity and the second similarity is determined. The metric distance is mapped to a positive value. The ratio of the sum to the positive value is calculated and used as the trend periodic correlation of each influencing data within the time series interval. The effective contribution of the prediction is obtained by using the trend cycle correlation and the mutual information dependence.

7. The power consumption prediction method based on multidimensional data as described in claim 6, characterized in that, The predictive effective contribution is the normalized value of the product of the trend cycle correlation and the mutual information dependence.

8. The power consumption prediction method based on multidimensional data as described in claim 1, characterized in that, The correction of the regression coefficients in the process of predicting power grid electricity consumption data using the PCR principal component regression algorithm includes: The ratio of the predictive effective contribution to the redundancy imbalance of each impact data point within a time series interval is calculated and used as the prediction weight adjustment factor for each impact data point within the time series interval. The mean of the prediction weight adjustment factors for each impact data point within a preset historical period is calculated and multiplied by the original regression coefficients corresponding to each impact data point to obtain the corrected regression coefficients for each impact data point. The duration of the preset historical period is longer than the duration of the time series interval.

9. A power consumption prediction medium based on multidimensional data, wherein the medium stores a computer program, characterized in that, When the computer program is executed by the processor, it implements the power consumption prediction method as described in any one of claims 1-8.

10. A power consumption prediction system based on multidimensional data, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1-8.

Images