Electric charge prediction method, system and equipment based on hierarchical fusion of multiple models, and medium

Abstract

According to the layered fusion multi-model-based electricity charge prediction method, system and equipment and the medium, the electricity charge receivable in the first time, the actually-received electricity charge in the first time and the actually-received electricity charge in the second time are predicted respectively, layered prediction is carried out on the electricity charge by expanding to different time units, refined prediction is achieved, and the prediction efficiency is improved. The method is suitable for a multi-demand scene, and has relatively high availability and expansibility; the electric charge receivable in the first time is predicted based on a plurality of electric charge receivable prediction models, and the limitation of a single model in applicable scenes and applicable data types can be overcome to a certain extent, so that the accuracy and stability of electric charge income prediction are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field related to electricity, and in particular relates to an electricity fee prediction method, system, equipment and medium based on hierarchical fusion multi-model. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] For power companies, electricity revenue is the main source of cash inflow. The reasonable forecast of electricity revenue is of strategic significance. The forecast of electricity revenue is the basis for power supply companies to prepare budgets and an important reference for electricity price adjustment. With the continuous advancement of electricity marketization, competition among power supply companies is becoming increasingly fierce, and the strategic decision-making of enterprises needs more accurate forecast data support. The construction goals of the "5A" level electricity revenue forecasting model currently proposed are advanced, actual, automatic, adjustable, and accurate; scientific management of the "two settlement pools"; the promotion of the integration and capital management efficiency of the "three markets", namely the internal, domestic and overseas markets; and the strengthening of the financial risk prevention and control of the "three-dimensional safety line". All of these require improving the accuracy of the power grid company's electricity revenue forecast to meet high-level construction goals, support scientific management mechanisms, promote market integration, and assist in ensuring capital security.

[0004] In summary, it is necessary to conduct in-depth research on the forecasting method for electricity revenue and improve its forecasting accuracy. Based on the historical laws of electricity revenue in the past, the forecast of the future trend of enterprise electricity revenue can provide good decision-making support for enterprises in cost control, financing, investment and business expansion. Summary of the invention

[0005] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides an electricity charge prediction method, system, device and medium based on hierarchical fusion of multiple models. By expanding to different time units, hierarchical prediction of electricity charges is carried out to achieve refined prediction, which is suitable for multiple demand scenarios and has strong usability and scalability.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] In a first aspect, the present invention provides an electricity rate prediction method based on hierarchical fusion multi-model, comprising:

[0008] Obtain the influencing data related to the electricity charges receivable within the first time, and obtain the prediction results of the electricity charges receivable within the first time based on multiple electricity charges receivable prediction models and their corresponding prediction weights;

[0009] According to the constructed mathematical model of the electricity charges receivable within the first time and the electricity charges actually collected within the first time, according to the prediction result of the electricity charges receivable within the first time, the prediction result of the electricity charges actually collected within the first time is obtained;

[0010] Determine the relevant indicators of the electricity charges actually collected in the second time, assign category labels through a clustering algorithm based on the historical data corresponding to the determined relevant indicators, calculate the Markov state transition probability matrix based on the assigned category labels and the category conversion relationship represented by the time series, and establish a Markov prediction model;

[0011] According to the constructed Markov prediction model and the prediction result of the actually collected electricity charges within the first time period, the prediction result of the actually collected electricity charges within the second time period is obtained.

[0012] In a second aspect, the present invention provides an electricity fee prediction system based on hierarchical fusion multi-model, comprising:

[0013] The first prediction module is configured to: obtain relevant impact data of the electricity fee receivable within a first time, and obtain a prediction result of the electricity fee receivable within the first time based on a plurality of electricity fee receivable prediction models and their corresponding prediction weights;

[0014] The second prediction module is configured to: obtain a prediction result of the actually collected electricity fee within the first time according to the prediction result of the electricity fee receivable within the first time based on the constructed mathematical model of the electricity fee receivable within the first time and the actually collected electricity fee within the first time;

[0015] A Markov prediction model building module is configured to: determine relevant indicators of the actual electricity charges collected within the second time, assign category labels through a clustering algorithm according to the historical data corresponding to the determined relevant indicators, calculate the Markov state transition probability matrix according to the assigned category labels and the category conversion relationship represented by the time series, and establish a Markov prediction model;

[0016] The third prediction module is configured to obtain a prediction result of the actual electricity charge collected within the second time period according to the constructed Markov prediction model and the prediction result of the actual electricity charge collected within the first time period.

[0017] In a third aspect, the present invention provides an electronic device comprising a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in the first aspect is performed.

[0018] In a fourth aspect, the present invention provides a computer-readable storage medium for storing computer instructions, wherein when the computer instructions are executed by a processor, the method described in the first aspect is performed.

[0019] One or more of the above technical solutions have the following beneficial effects:

[0020] In the present invention, by predicting the electricity charges to be collected within the first time, the electricity charges actually collected within the first time, and the electricity charges actually collected within the second time respectively, and by expanding to different time units, a layered prediction of the electricity charges is performed to achieve refined prediction, which is suitable for multiple demand scenarios and has strong usability and scalability.

[0021] In the present invention, the electricity charges receivable within the first time are predicted based on multiple electricity charges receivable prediction models, which can overcome the limitations of a single model in applicable scenarios and applicable data types to a certain extent, thereby improving the accuracy and stability of electricity charge revenue prediction.

[0022] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0024] Figure 1 This is a flow chart of an electricity fee prediction method based on hierarchical fusion multi-model in Embodiment 1 of the present invention;

[0025] Figure 2 A schematic diagram of the relationship between the electricity charges payable and the electricity charges actually collected in the first embodiment of the present invention;

[0026] Figure 3 Schematic diagram of the electricity fee prediction system in the first embodiment of the present invention. DETAILED DESCRIPTION

[0027] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0028] It should be noted that the terms used herein are for describing specific embodiments only and are not intended to be limiting of exemplary embodiments according to the present invention.

[0029] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0030] Embodiment 1

[0031] This embodiment discloses a method for predicting electricity charges based on hierarchical fusion of multiple models, including:

[0032] Obtain the influencing data related to the electricity charges receivable within the first time, and obtain the prediction results of the electricity charges receivable within the first time based on multiple electricity charges receivable prediction models and their corresponding prediction weights;

[0033] According to the constructed mathematical model of the electricity charges receivable within the first time and the electricity charges actually collected within the first time, according to the prediction result of the electricity charges receivable within the first time, the prediction result of the electricity charges actually collected within the first time is obtained;

[0034] Determine the relevant indicators of the electricity charges actually collected in the second time, assign category labels through a clustering algorithm based on the historical data corresponding to the determined relevant indicators, calculate the Markov state transition probability matrix based on the assigned category labels and the category conversion relationship represented by the time series, and establish a Markov prediction model;

[0035] According to the constructed Markov prediction model and the prediction result of the actually collected electricity charges within the first time period, the prediction result of the actually collected electricity charges within the second time period is obtained.

[0036] like Figure 3 As shown, the electricity price prediction method based on layered fusion multi-model proposed in this embodiment is implemented based on the Realinsight digital platform and deployed on the Tomcat8.x server of the power information intranet. The back-end adopts the microservice architecture of SpringCloud, SpringBoot and JPA, and the front-end uses Echarts components, jQuery and Ajax technology; the communication protocol adopts HTTP, the data format is JSON, and the database uses PostgreSQL and ElasticSearch; it is divided into six parts: display layer, control layer, business logic service layer, persistence layer, data layer and basic platform layer.

[0037] The electricity bill forecast includes the forecast of receivable electricity bills and the forecast of actually collected electricity bills. The two parts are explained in detail below.

[0038] The basic process of electricity bill receivable forecasting is as follows: preprocess the original data to meet the principles of data homogeneity and comparability; select forecasting models of different technical route categories, including time series forecasting models, machine learning regression models, etc., and optimize parameters around the optimization goal of minimum mean square error to obtain the optimal forecasting model of each forecasting model under the condition of minimum mean square error, and obtain the forecast result based on the optimal forecasting model; then, according to the proportional relationship between the error sizes of the above-mentioned optimal forecasting models, assign different weights to the forecast results of each forecasting model, and finally perform weighted summation to obtain the forecast result of electricity bill receivable.

[0039] like Figure 1-Figure 2As shown, the input of the electricity fee receivable prediction model is the original data including the electricity fee data of different categories and the data of related influencing factors; the output of the electricity fee receivable prediction model is the electricity fee receivable prediction result with the month as the prediction unit.

[0040] Specifically, the data contains the following category fields: internal explanatory variables, external explanatory variables, and the target variable, i.e., the electricity charges receivable.

[0041] As shown in Table 1, the internal explanatory variables include user profile information, electricity consumption information, payment information, electricity price information, etc.; the external explanatory variables include season information, temperature information, holiday information, etc.; the target variable is the electricity bill receivable.

[0042]

[0043] In order to ensure the homogeneity and comparability of data, the original data is preprocessed into data that can be directly used in subsequent steps. The preprocessing method is as follows:

[0044] 1. For different types of data during the period when the purchase and sale policy existed, data restoration shall be performed according to different accounting rules;

[0045] 2. Restore the data affected by the epidemic according to the historical month-on-month growth rate;

[0046] 3. Remove duplicate data, fill missing values ​​and correct outliers.

[0047] This embodiment constructs a prediction model algorithm library to facilitate the subsequent establishment of a fusion prediction model, as shown in Table 2, including Prophet, Holt-Winters, ARIMA, random forest, XGBoost, LightGBM, LSTM, linear regression and gray model.

[0048] Table 2:

[0049] Model Name Model explanation Prophet A Time Series Forecasting Algorithm Holt-Winters A time series prediction algorithm ARIMA A time series prediction algorithm Random Forest An ensemble learning algorithm XGBoost A Gradient Reinforcement Learning Algorithm LightGBM A Gradient Reinforcement Learning Algorithm LSTM Long Short-Term Memory Neural Networks Linear Regression A basic regression analysis algorithm Gray Model A prediction algorithm for uncertain systems

[0050] The above preprocessed input data is combined with the models contained in the prediction model algorithm library to perform parameter tuning training on each model around the optimization goal of minimizing the mean square error (MSE), and the three models with the best performance are selected as sub-models of the fusion prediction model. The selected models and their optimal mean square errors after training are shown in Table 3.

[0051] Table 3 Each sub-model of the fusion prediction model and the corresponding optimal mean square error

[0052]

[0053]

[0054] The optimal mean square error of the sub-models is related to each other. Each sub-model is given a weight. The weight w of the i-th sub-model is i The calculation expression is:

[0055]

[0056] Among them, γ j Represents the inverse of the optimal mean square error of the jth sub-model.

[0057] In summary, the predicted electricity bill P r for:

[0058] p r =w1p r1 +w2p r2 +w3p r3 (2)

[0059] Specifically, it can be:

[0060] p r =0.4025p r1 +0.3376p r2 +0.2599p r3

[0061] Among them, P r1 , P r2 , P r3 They represent the predicted values ​​of the receivable electricity charges of the sub-models Prophet, random forest, and linear regression, respectively; w1, w2, and w3 represent the weights of the sub-models Prophet, random forest, and linear regression, respectively.

[0062] In this embodiment, the actual electricity charges collected include monthly actual electricity charges, annual actual electricity charges, etc. This implementation predicts the monthly actual electricity charges, and then predicts smaller time units such as daily electricity charges to be collected. Alternatively, the annual actual electricity charges are predicted, and then smaller time units such as monthly actual electricity charges are predicted.

[0063] The overall process of predicting the actual electricity charges is as follows: according to the composition structure of relevant costs, the mathematical relationship between the receivable electricity charges and the actual electricity charges is derived, and the actual electricity charges prediction model equation containing the coefficients to be predicted is given; a method based on multivariate linear regression is proposed to calibrate the coefficients of the actual electricity charges prediction model equation in combination with the prediction results of the receivable electricity charges, and finally the prediction results of the actual electricity charges are calculated based on the calibrated model equation; on the basis of the above-mentioned actual electricity charges prediction results, the actual electricity charges of smaller prediction units are predicted, and a certain method is used to complete the selection of relevant characteristic indicators, and a clustering algorithm is used to assign category labels to the time periods to be predicted, and the Markov state transition probability matrix is ​​calculated based on the time series and category conversion relationship. At the same time, different weights are assigned to each category according to the size relationship of the indicators of each category, and finally the Markov prediction model is used to obtain the actual electricity charges prediction results of smaller prediction units.

[0064] This embodiment is explained by firstly predicting the monthly electricity charges and then the daily electricity charges to be collected.

[0065] The mathematical expressions involving the actual electricity charges collected this month, the electricity charges collected last month, and the electricity charges collected this month are as follows: W represents the actual electricity charges collected this month, X represents the electricity charges collected last month, Y represents the electricity charges collected this month, and a and b are dimensionless parameters. Therefore, the monthly actual electricity charges prediction model equation is established:

[0066] W=a·X+b·Y (3)

[0067] Among them, a and b are dimensionless parameters to be calibrated.

[0068] Based on the idea of ​​multiple linear regression, the training is carried out with the minimum mean square error as the optimization goal, the parameters are calibrated, and the optimal parameter set is obtained.

[0069] In this embodiment, the corresponding optimal parameters are shown in Table 4:

[0070] Table 4: Parameter calibration results of the monthly actual electricity bill forecasting model equation

[0071] parameter Parameter Explanation Calibrate the optimal value a Last month's electricity bill receivable coefficient 0.6031 b This month's electricity bill coefficient 0.3969

[0072] According to the above-mentioned monthly electricity fee receivable forecast results, the monthly electricity fee actually collected forecast model equation is calculated to obtain the monthly electricity fee actually collected forecast results, and based on this monthly electricity fee actually collected forecast result, the actual electricity fee collected forecast is carried out on a daily basis, specifically including:

[0073] Step 1: Selection of relevant characteristic indicators.

[0074] The mutual information of the candidate indicators is calculated, and the four indicators with the largest mutual information are selected for subsequent clustering analysis.

[0075] Mutual information is an important concept in information theory, which is used to measure the mutual dependence between two random variables. It is applicable not only to real-valued random variables, but also to any form of random variables. Its calculation formula is as follows:

[0076]

[0077] or:

[0078]

[0079] Among them, X is the indicator to be screened, and Y is the target variable, so that the dependency relationship between each indicator and the target variable can be obtained.

[0080] Consider the mutual dependence between the user profile information, electricity information, payment information, electricity price information, season information, weather information, temperature information, holiday information and the target variable daily electricity charges. The larger the value of mutual information, the greater the dependence between the variables. Therefore, the mutual information of the selected indicators is calculated, and n indicators with the largest mutual information, such as 4 indicators, are selected for subsequent cluster analysis.

[0081] In this embodiment, the selected indicators and their corresponding mutual information are shown in Table 5.

[0082] Table 5: Selected indicators and their corresponding mutual information

[0083] Indicator (variable) name Variable Types Mutual Information weather Categorical variables 0.3227 Average temperature Discrete variables 0.2816 season Categorical variables 0.2179 Holidays Categorical variables 0.1656

[0084] Step 2: In the training set, use the data corresponding to the above four indicators and use the DBSCAN density clustering algorithm to assign category labels to each day.

[0085] In this embodiment, according to the above four indicators, each day is divided into three categories, namely, three states in the Markov prediction model.

[0086] Specifically, the indicators corresponding to each day are converted, the weather indicators are converted into numerical values: sunny = 1, cloudy = 2, overcast = 3, etc.; the average temperature indicator is standardized to fall within the interval [0,1], the season indicator is converted into numerical values: spring = 1, summer = 2, autumn = 3, winter = 4; holidays are converted into numerical values: holiday = 1, non-holiday = 0; after the indicator conversion, the overall standardization is performed to ensure that the values ​​of different attributes are on the same scale and the clustering results will not be affected by the dimension.

[0087] This embodiment uses the DBSCAN clustering algorithm to perform cluster analysis, and assigns different category labels to each day, which are state 1, state 2, and state 3. Specifically, the neighborhood radius and the minimum number of samples, which are the key parameters of DBSCAN, are selected. In the embodiment, the neighborhood radius and the minimum number of samples are set; the density-based clustering DBSCAN algorithm is used, and the algorithm automatically selects a sample point as the center of the circle and draws a circle with the set radius. If the number of sample points contained in the circle is greater than or equal to the minimum number of samples, all sample points in the circle are used as the center of the circle, and the circle is continued to be drawn. It is determined whether the number of samples in the circle meets the conditions, and the samples contained in the circle that meets the conditions are clustered with the previous samples, and so on, until the conditions are no longer met, then the next sample point is found and another sample cluster is started.

[0088] Among them, when the clustering algorithm is used to calculate the sample distance, the weather, average temperature, season and holiday characteristics of each sample day after encoding are used to calculate the distance between samples.

[0089] This embodiment uses the sample silhouette coefficient mean as the model evaluation method, and the calculation formula for the silhouette coefficient of each sample point is as follows:

[0090]

[0091] Among them, i represents the i-th sample point, a i represents the average distance between the sample point and other sample points in its cluster, b i It represents the average distance between the sample point and all sample points in other clusters of its nearest neighbors. The value range of the silhouette coefficient is [-1, 1]. The larger the value, the better the sample point matches its own cluster, and the greater the difference with the adjacent cluster, that is, the better the clustering effect.

[0092] The mean calculation formula of the sample silhouette coefficient is:

[0093]

[0094] Among them, n is the number of sample points.

[0095] Next, using the grid search method, when the neighborhood radius is 0.5 and the minimum number of samples is 60, the mean of the silhouette coefficient is the largest, which is 0.904, and the clustering effect is the best. At this time, the DBSCAN algorithm clusters the samples of this embodiment into three sample clusters. This embodiment assigns three labels to the three sample clusters, which are represented as state 1, state 2, and state 3 respectively.

[0096] For example, if a day is sunny, the standardized temperature is 0.7, the season is spring and it is a holiday, then the label of the day is 1; if a day is sunny, the standardized temperature is 0.5, the season is autumn and it is a holiday, then the label of the day is 1; if a day is sunny, the standardized temperature is 1, the season is summer and it is not a holiday, then the label of the day is 2; if a day is rainy, the standardized temperature is 0, the season is winter and it is not a holiday, then the label of the day is 3. By analogy, if the distance between features is close, the algorithm will assign the same label to the samples, such as 1; if the distance between features is far, they will be assigned different labels, such as 2, 3.

[0097] For new samples, labels are assigned to the new samples in the following way: the average distance between the new sample and all points in each cluster is calculated, and the new sample is assigned to the cluster with the smallest average distance.

[0098] Step 3: Based on the category label and the category conversion relationship represented by the time series, calculate the Markov state transition probability matrix.

[0099] Specifically, since there are three types of category labels for each day, that is, there are three states in Markov prediction, according to this result, a 3*3 state transition count matrix is ​​initialized; the training set with category labels is traversed in time sequence. If the state of a certain day changes compared with the previous day, the corresponding position in the state transition count matrix is ​​increased by 1 count. After obtaining the final state transition count matrix, the joint probability distribution is calculated and converted into a state transition probability matrix. The state transition probability can be expressed by the following formula:

[0100]

[0101] Among them, the state transition probability P ij Num is the number of samples that change from state i, i.e., category label i, to state j, i.e., category label j ij The ratio of (i.e., the value of the i column and j row of the state transition frequency matrix) to the total number of samples Num is used to obtain the Markov state transition probability matrix.

[0102] Formula (6) is used to convert the state transition count matrix into the state transition probability matrix. The specific calculation process is as follows: For example, the previous day is state 1, the next day is state 2, and each time the state changes, then Num 12 Just add 1. Use the same method to get the number of transitions of other states, draw the state transition number matrix, and then use this formula to get the state transition probability matrix. Thus, the state transition number matrix is ​​transformed into the state transition probability matrix, and the corresponding values ​​in the probability matrix in Table 6 are obtained.

[0103] In this embodiment, the obtained Markov state transition probability matrix is ​​shown in Table 6, where the vertical direction is the transition-out state and the horizontal direction is the transition-in state.

[0104] Table 6: Markov state transition probability matrix

[0105] State 1 State 2 State 3 State 1 0.3320 0.3402 0.3279 State 2 0.3918 0.2694 0.3388 State 3 0.2870 0.4174 0.2957

[0106] Step 4: Assign weights to the actual electricity charges based on the status.

[0107] In this embodiment, the weight w of the day corresponding to state i si The calculation formula is:

[0108]

[0109] in, is the daily electricity charges of the corresponding samples in the training set that are transferred from state i to state j, is the sum of the daily electricity charges of all samples in the training set, j is the number of samples corresponding to state i, and n is the total number of samples.

[0110] Step 5: Establish a Markov prediction model based on the Markov state transition probability matrix.

[0111] Based on the characteristic indicators of the day to be predicted in the test set, the above Markov prediction model is used to obtain the actual electricity charge prediction result of the day to be predicted.

[0112] Assuming that the category label of the previous day is i, the formula for the actual electricity charge on the day to be predicted with the label j is as follows:

[0113]

[0114] Among them, E j is the actual electricity charge on the day to be predicted, Z is the predicted monthly actual electricity charge, N day is the number of days in the month, w sj is the weight corresponding to state j, p ij is the state transition probability from state i on the previous day to state j on the current day.

[0115] Embodiment 2

[0116] The purpose of this embodiment is to provide an electricity fee prediction system based on hierarchical fusion multi-model, including:

[0117] The first prediction module is configured to: obtain relevant impact data of the electricity fee receivable within a first time, and obtain a prediction result of the electricity fee receivable within the first time based on a plurality of electricity fee receivable prediction models and their corresponding prediction weights;

[0118] The second prediction module is configured to: obtain a prediction result of the actually collected electricity fee within the first time according to the prediction result of the electricity fee receivable within the first time based on the constructed mathematical model of the electricity fee receivable within the first time and the actually collected electricity fee within the first time;

[0119] A Markov prediction model building module is configured to: determine relevant indicators of the actual electricity charges collected within the second time, assign category labels through a clustering algorithm according to the historical data corresponding to the determined relevant indicators, calculate the Markov state transition probability matrix according to the assigned category labels and the category conversion relationship represented by the time series, and establish a Markov prediction model;

[0120] The third prediction module is configured to obtain a prediction result of the actual electricity charge collected within the second time period according to the constructed Markov prediction model and the prediction result of the actual electricity charge collected within the first time period.

[0121] In further embodiments, there is also provided:

[0122] An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in Embodiment 1 is performed. For the sake of brevity, no further description is given here.

[0123] It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0124] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0125] A computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the method described in embodiment 1 is completed.

[0126] The method in the first embodiment can be directly embodied as a hardware processor, or a combination of hardware and software modules in the processor. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware. To avoid repetition, it will not be described in detail here.

[0127] A computer program product includes a computer program, and when the computer program is executed by a processor, the method described in the first embodiment is implemented.

[0128] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer executable instructions, such as instructions included in a program module, which are executed in a device on a real or virtual processor of the target to perform the process / method as described above. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc. that perform specific tasks or implement specific abstract data types. In various embodiments, the functions of program modules can be combined or divided between program modules as needed. Machine executable instructions for program modules can be executed in local or distributed devices. In distributed devices, program modules can be located in local and remote storage media.

[0129] The computer program code for implementing the method of the present invention can be written in one or more programming languages. These computer program codes can be provided to the processor of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code, when executed by the computer or other programmable data processing device, causes the function / operation specified in the flow chart and / or block diagram to be implemented. The program code can be executed completely on a computer, partially on a computer, as an independent software package, partially on a computer and partially on a remote computer or completely on a remote computer or server.

[0130] In the context of the present invention, computer program codes or related data may be carried by any appropriate carrier to enable a device, apparatus or processor to perform the various processes and operations described above. Examples of carriers include signals, computer readable media, etc. Examples of signals may include electrical, optical, radio, acoustic or other forms of propagation signals, such as carrier waves, infrared signals, etc.

[0131] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0132] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A method for predicting electricity charges based on hierarchical fusion of multiple models, characterized in that: include: Obtain the influencing data related to the electricity charges receivable within the first time, and obtain the prediction results of the electricity charges receivable within the first time based on multiple electricity charges receivable prediction models and their corresponding prediction weights; According to the constructed mathematical model of the electricity charges receivable within the first time and the electricity charges actually collected within the first time, according to the prediction result of the electricity charges receivable within the first time, the prediction result of the electricity charges actually collected within the first time is obtained; Determine the relevant indicators of the electricity charges actually collected in the second time, assign category labels through a clustering algorithm based on the historical data corresponding to the determined relevant indicators, calculate the Markov state transition probability matrix based on the assigned category labels and the category conversion relationship represented by the time series, and establish a Markov prediction model; According to the constructed Markov prediction model and the prediction result of the actually collected electricity charges within the first time period, the prediction result of the actually collected electricity charges within the second time period is obtained.

2. The electricity rate prediction method based on hierarchical fusion multi-model according to claim 1, characterized in that: By utilizing the influencing data of electricity charges receivable in the first time in history, multiple prediction models with different technical routes are tuned and trained based on the optimization goal of minimizing the mean square error, and the electricity charges receivable prediction model used for electricity charges receivable prediction is determined; based on the size relationship of the optimal mean square error of the electricity charges receivable prediction model, corresponding prediction weights are assigned to each electricity charges receivable prediction model.

3. The electricity rate prediction method based on hierarchical fusion multi-model according to claim 1, characterized in that: A mathematical model is constructed based on the relationship between the electricity charges receivable and its coefficients issued at different time points and the electricity charges actually collected in the first time; based on multiple linear regression, the coefficients are solved with the minimum mean square error as the optimization goal to obtain the solved mathematical model.

4. The electricity rate prediction method based on hierarchical fusion multi-model according to claim 1, characterized in that: The relevant indicators for determining the actual electricity charges collected within the second time are specifically: Determine a plurality of to-be-screened indicators related to a target variable, wherein the target variable is the actual electricity charges collected within the second time; Using mutual information method, calculate the dependency relationship between each index to be screened and the target variable; According to the calculated dependency relationship, the relevant index of the target variable is determined.

5. The electricity fee prediction method based on hierarchical fusion multi-model according to claim 1, characterized in that: The category labels are assigned through the clustering algorithm. According to the assigned category labels and the category conversion relationship represented by the time series, the Markov state transition probability matrix is ​​calculated, which is: Traverse the training set with category labels in time sequence. If the state of a certain day changes compared with the previous day, increase the count at the corresponding position of the state transition count matrix until the final state transition count matrix is ​​obtained. The final state transition count matrix is ​​converted into a Markov state transition probability matrix.

6. The electricity rate prediction method based on hierarchical fusion multi-model according to claim 1, characterized in that: According to the constructed Markov prediction model and the prediction result of the actual electricity charges collected in the first time, the prediction result of the actual electricity charges collected in the second time is obtained, which is specifically: Determine the weight corresponding to the category label of the day to be predicted; The prediction result of the actual electricity charges within the second time period is calculated based on the state transition probability between the prediction day and the previous day, the weight corresponding to the determined category label of the day to be predicted, and the prediction result of the actual electricity charges within the first time period.

7. The method for predicting electricity charges based on hierarchical fusion of multiple models as claimed in claim 6, characterized in that: The weight corresponding to the category label of the day to be predicted is determined as follows: The weights corresponding to different category labels are determined according to the ratio of the sum of the actually collected electricity charges within the second time of the samples corresponding to the state transition between different category labels in the training set and the sum of the actually collected electricity charges within the second time of all samples in the training set.

8. An electricity fee prediction system based on hierarchical fusion multi-model, characterized in that: include: The first prediction module is configured to: obtain relevant impact data of the electricity fee receivable within a first time, and obtain a prediction result of the electricity fee receivable within the first time based on a plurality of electricity fee receivable prediction models and their corresponding prediction weights; The second prediction module is configured to: obtain a prediction result of the actually collected electricity fee within the first time according to the prediction result of the electricity fee receivable within the first time based on the constructed mathematical model of the electricity fee receivable within the first time and the actually collected electricity fee within the first time; A Markov prediction model building module is configured to: determine relevant indicators of the actual electricity charges collected within the second time, assign category labels through a clustering algorithm according to the historical data corresponding to the determined relevant indicators, calculate the Markov state transition probability matrix according to the assigned category labels and the category conversion relationship represented by the time series, and establish a Markov prediction model; The third prediction module is configured to obtain a prediction result of the actual electricity charge collected within the second time period according to the constructed Markov prediction model and the prediction result of the actual electricity charge collected within the first time period.

9. An electronic device, characterized in that: The method comprises a memory and a processor and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method according to any one of claims 1 to 7 is completed.

10. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the method described in any one of claims 1 to 7.

Images