Electric quantity prediction method and device based on comprehensive prediction model, equipment and medium

Abstract

The invention provides an electric quantity prediction method and prediction device based on a comprehensive prediction model, electronic equipment and a storage medium, and belongs to the technical field of electric quantity prediction. The prediction method comprises the following steps: collecting historical electric quantity data, and selecting a common single prediction model; defining a probability matrix, obtaining a historical prediction error of the single prediction model, and obtaining an estimated probability matrix in combination with the freshness function; analyzing the pre-estimated probability matrix, and solving a weight feature vector by using a feature vector method; using the weight feature vector to represent an approximate weight probability distribution function, obtaining the mathematical expectation of the weight feature vector, and screening out a better prediction model; combining the better prediction models to form a comprehensive model in combination with historical electric quantity data, and solving the weight of each model by using a genetic algorithm to obtain a comprehensive prediction model; and performing medium-and-long-term electric quantity prediction by using the comprehensive prediction model. The invention provides a comprehensive prediction model based on an improved prediction decision, and accurate medium and long term electric quantity prediction can be carried out.

Description

Technical Field

[0001] The present invention belongs to the technical field of power forecasting, and specifically relates to a power forecasting method, a forecasting device, an electronic device and a storage medium based on a comprehensive forecasting model. Background Art

[0002] The electricity forecast is divided into medium- and long-term forecast and short-term forecast according to the forecast period. The medium- and long-term electricity forecast is an important support for the power department to make grid planning decisions. Accurate forecast is of great significance to the economic and stable operation of the power system. Especially since the marketization of electricity, some industrial and commercial users have participated in the market by purchasing electricity on behalf of the power grid. The medium- and long-term electricity forecast is the theoretical guidance in the process of purchasing electricity on behalf of the power grid, and is an important guarantee for the economic interests of the power grid and users.

[0003] At present, there are many different methods for electricity forecasting, and there are dozens of commonly used methods. For example, parameter-based models include regression analysis, time series method, correlation analysis method, etc., and non-parametric models include gray system method, artificial neural network, fuzzy mathematics, support vector machine, etc. Faced with a complex and changing forecasting environment, a single forecasting method is difficult to obtain satisfactory results. Therefore, multiple forecasting models can be combined to form a combined model for forecasting. However, how to select a suitable algorithm model from the numerous forecasting methods and how to allocate the weights between the various forecasting models are the key factors affecting the forecasting effect of the combined model. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide an electricity prediction method, prediction device, electronic device and storage medium based on a comprehensive prediction model in view of the deficiencies in the prior art.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] A method for electricity forecasting based on a comprehensive forecasting model comprises: collecting historical electricity data and selecting a commonly used single forecasting model; defining a probability matrix, obtaining the historical forecasting error of the single forecasting model, and obtaining an estimated probability matrix in combination with a freshness function; analyzing the estimated probability matrix and obtaining a weight eigenvector using an eigenvector method; using the weight eigenvector to represent an approximate weight probability distribution function, obtaining the mathematical expectation of the weight eigenvector, and screening out a better forecasting model; combining the historical electricity data, combining the better forecasting models to form a comprehensive model, and using a genetic algorithm to obtain the weights of each model to obtain a comprehensive forecasting model; and using the comprehensive forecasting model to perform medium- and long-term electricity forecasting.

[0007] Furthermore, the method for obtaining the historical prediction error of a single prediction model includes: the prediction error of the t-th prediction value of the ith single prediction model in is the tth prediction value of the ith single prediction model, y it is the tth true value of the ith single prediction model.

[0008] Furthermore, the method of obtaining the estimated probability matrix includes: the weight of the i-th single model or where ε it is the historical forecast error, α(1-α) t is the freshness function, t=1,2,...,n, n is the number of historical electricity data corresponding to a single prediction model; based on the weights of all single prediction models, the estimated probability matrix is ​​obtained.

[0009] Furthermore, the method for obtaining the estimated probability matrix includes: introducing an evaluation method to obtain a comprehensive coefficient of credibility evaluation r s =f(r 1,s ,r 2,s ,…,r m,s ), r s ∈[0,1], where r m,s represents the credibility evaluation coefficient of the mth evaluation method introduced in the sth historical period; the weight of the i-th single model or where ε it is the historical forecast error, α(1-α) t is the freshness function, t=1,2,...,n, r s is the comprehensive coefficient of credibility evaluation, n is the number of historical electricity data corresponding to a single prediction model; based on the weights of all single prediction models, the estimated probability matrix is ​​obtained.

[0010] Furthermore, the reliability evaluation comprehensive coefficient r s =f(r T,s ,r D,s ,r P,s ,…), where: r T,s is the temperature credibility evaluation coefficient of the sth historical period, r T,s =|T s -T f |, T s is the comprehensive temperature coefficient, T f is the temperature credibility evaluation coefficient evaluated by experts, T s,max is the highest temperature in the sth historical period, T s,avg is the average temperature in the sth historical period, T s,min is the lowest temperature in the sth historical period, is the temperature variance in the sth historical period, c1 and c2 are constants used to map the temperature comprehensive coefficient to the interval [0,1]; rD,s is the credibility evaluation coefficient of natural disasters in the sth historical period, r P,s is the policy credibility evaluation coefficient of the sth historical period.

[0011] Furthermore, the comprehensive prediction model is expressed as Where p is the number of better prediction models, ω i ' is the final weight of the i-th single model, y i is the predicted value of the ith single model.

[0012] Furthermore, the approximate weight probability distribution function is 0<x<1; the mathematical expectation of the weighted eigenvector W is The better prediction model is the single prediction model with a weight greater than E(W).

[0013] A power forecasting device based on a comprehensive forecasting model comprises an acquisition unit, a construction unit and a forecasting unit; the acquisition unit is used to acquire historical power data and to acquire a commonly used single forecasting model; the construction unit is used to acquire an estimated probability matrix based on a freshness function and the historical forecasting error of the single forecasting model; the estimated probability matrix is ​​analyzed and a weight eigenvector is obtained using an eigenvector method; the weight eigenvector is used to represent an approximate weight probability distribution function, obtain the mathematical expectation of the weight eigenvector, and screen out a better forecasting model; the historical power data is combined to combine the better forecasting models into a comprehensive model, and a genetic algorithm is used to obtain the weights of each model to finally obtain a comprehensive forecasting model; the forecasting unit is used to use the comprehensive forecasting model to perform medium- and long-term power forecasting.

[0014] An electronic device comprises a processor and a memory; the memory is used to store executable instructions, and the processor is used to execute the instructions to implement the prediction method.

[0015] A computer-readable storage medium stores instructions, and when the instructions are executed, the prediction method is implemented.

[0016] Among the combined forecasting methods, the most widely used are variance-covariance combination and variable weight combination. Among them, the variance-covariance combination method is based on the principle of "minimum historical error", which is prone to overfitting, and the obtained weights are not optimal, and the obtained results have negative weight problems. The weight determination method of the variable weight combination method is relatively complicated, and it is extremely difficult to obtain when there are many types of single forecasting models.

[0017] The prediction decision method can use the probability matrix (Odds-matrix) to evaluate the prediction quality of multiple models and complete the model screening. However, there are two problems with the prediction decision method in the prior art. First, the matrix estimation process is not clearly specified. Second, when there are large fluctuations in historical data, the prediction effect of the model screened by this method is not ideal. There is also an existing technology that improves the adaptability and anti-interference of the model by smoothing the historical data with large fluctuations, but this method does not consider the effective information in the historical data with large fluctuations, and the prediction effect of electricity in special cases is poor, and the matrix estimation method is not mentioned.

[0018] Compared with the prior art, the present invention has the following beneficial effects:

[0019] The present invention draws on the ideas of information theory, introduces the freshness function, improves the estimation process of the probability matrix, and establishes a comprehensive prediction model for medium- and long-term electricity forecasting; on this basis, the credibility of historical data is evaluated through expert experience, the prediction decision algorithm is further improved, and the model's ability to adjust predictions for special situations is improved. Taking the monthly electricity forecast of a provincial power grid as an example for prediction analysis, the results show that the model and method mentioned in the present invention are very practical and effective.

[0020] According to the mid- and long-term changes in electricity consumption, the present invention proposes an improved prediction decision algorithm and establishes a comprehensive prediction model based on improved prediction decision. The model not only processes the influence of historical data according to the principle of "near is large and far is small" by introducing a freshness function, but also evaluates the credibility of historical data based on expert experience, and searches for historical data that is most similar to recent predictions, thereby obtaining a more accurate probability matrix and completing the model screening. The monthly prediction results of a provincial power grid show that the improved prediction decision algorithm proposed by the present invention can more accurately screen out models suitable for recent predictions, and the formed comprehensive prediction model has high accuracy and practicality.

[0021] The present invention can accurately predict medium- and long-term electricity volume. In the face of the new situation of imbalance in electricity supply and demand and the operation of the spot market, high-precision electricity volume forecasting can provide an important basis for electricity market transactions and provide a guarantee for the stable operation of the electricity market. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The present invention is further described in detail below in conjunction with the accompanying drawings.

[0023] Figure 1 : Flow chart of embodiment 1 of the present invention;

[0024] Figure 2 : A schematic diagram of the architecture of the comprehensive prediction model in Example 1 of the present invention;

[0025] Figure 3 : A schematic diagram of a process for introducing expert experience in Example 1 of the present invention;

[0026] Figure 4 : A schematic diagram of Embodiment 2 of the present invention;

[0027] Figure 5 : Schematic diagram of embodiment 3 of the present invention. DETAILED DESCRIPTION

[0028] In order to better understand the present invention, the content of the present invention is further clearly described below in conjunction with the embodiments and the accompanying drawings, but the protection content of the present invention is not limited to the following embodiments. In the following description, a large number of specific details are provided to provide a more thorough understanding of the present invention. However, it is obvious to those skilled in the art that the present invention can be implemented without one or more of these details.

[0029] Example 1: See Figure 1-Figure 3 ,The purpose of this embodiment is to provide a method for power ,prediction based on a comprehensive prediction model.

[0030] The purpose of the prediction decision in the prediction method of this embodiment is to select some models with better prediction effects from multiple single prediction models and combine them to form a comprehensive prediction model. First, the probability matrix (Odds-matrix) in decision theory is used to evaluate the pros and cons of single prediction models to form a weighted probability distribution function. Then, by comparing with the mathematical expectation of the probability distribution function, the better prediction model for a certain period and a certain region is selected. Finally, the optimization algorithm is used to obtain the weights of each single model in the comprehensive model to form the final optimal comprehensive model.

[0031] like Figure 1 As shown, the prediction method includes:

[0032] Step S1, collect historical electricity data and select a commonly used single prediction model.

[0033] The historical data is the monthly historical electricity data. Commonly used single prediction models include artificial neural network, linear model, N-order curve model, hyperbola model, fuzzy mathematics method, etc. Let the number of single prediction models be q.

[0034] Step S2, define a probability matrix, combine the freshness function and the historical prediction error of a single prediction model, and obtain an estimated probability matrix.

[0035] Specifically, this step S2 includes:

[0036] S21. Define the probability matrix.

[0037] The probability matrix method originates from decision theory and is a method for solving multi-objective optimization problems. The concept of probability is the ratio of the probability of a random event to the probability of it not occurring. For prediction, the accuracy of a model represents the probability of the model's predicted results. The higher the accuracy of the model's predicted results, the greater the probability of its predicted results. Correspondingly, the greater the weight of the model when making a correct decision. Based on this idea, the decision matrix O and the weight ratio matrix R in the probability matrix are defined.

[0038]

[0039] In the formula, q represents the number of models to be screened, ω i It represents the weight of the i-th single prediction model. The larger the weight, the better the prediction effect of the single model. When the decision matrix O is estimated accurately, O=R should be satisfied. However, in actual operation, it is difficult to ensure that the estimation of the decision matrix O is completely accurate. Most estimation methods can only satisfy O≈R, that is, Therefore O ij It represents the probability that the i-th single model is better than the j-th single model.

[0040] S22. Combine historical electricity data to obtain historical prediction errors of a single prediction model.

[0041] The decision matrix O can be estimated using the raw error of each single forecast model. Historical forecast errors (relative errors) of single forecast models In the formula is the tth prediction value of the ith single prediction model, y it is the tth true value of the ith single prediction model.

[0042] Weight ω i The estimation of can be done in absolute value or root mean square form. The absolute value expression is The root mean square expression is n is the number of historical electricity data corresponding to a single prediction model.

[0043] After obtaining the weights of all single prediction models, we combine The probability matrix can be estimated and the estimated probability matrix can be obtained.

[0044] S22. Define a freshness function.

[0045] According to the theory of information theory, the newer the data is, the more useful information it contains, and therefore the more important it is when making predictions. When estimating the decision matrix, the impact of prediction errors in different time periods should not be considered the same, but should follow the principle of "larger in the near future and smaller in the far future", paying more attention to the recent prediction errors of the model. For this reason, the freshness function is introduced to obtain a new matrix estimation method.

[0046] The freshness function is used to describe the importance of data when making predictions. The freshness function defined in this step is α(1-α) t , t=1,2,...,n, is a freshness function in the form of exponential smoothing.

[0047] S23. Estimate the probability matrix by combining the freshness function and the historical prediction error to obtain an estimated probability matrix.

[0048] Combined with freshness function α(1-α) t and historical forecast error ε it , estimate the probability matrix and get the weight ω i The absolute value estimation expression of Weight ω i The root mean square estimation expression of Based on the weight ω i , get the estimated probability matrix.

[0049] Step S3: Analyze the estimated probability matrix and use the eigenvector method to obtain the weight eigenvector.

[0050] Let the weighted feature vector W = [ω1, ω2, …, ωq] T , since O≈R, we have (O-qI)W≈0, where I is the identity matrix.

[0051] Although it is impossible to guarantee that the estimated matrix O is completely correct, the error between O and R can be minimized through appropriate methods, which is equivalent to introducing a small perturbation on O, thereby causing a small change in the main eigenvalue of the matrix. The weight vector is the eigenvector corresponding to the main eigenvalue λ1 of the matrix O, that is, OW=λ1W. The main eigenvalue λ1 and the weight eigenvector W can be obtained by iteratively solving the power method.

[0052] Step S4, using the weight feature vector to represent the approximate weight probability distribution function, obtaining the mathematical expectation of the weight feature vector, and screening out a better prediction model.

[0053] Since the weight ω of a single prediction model i is a number between [0,1]. Assuming that the weight of a single prediction model approximately follows the β distribution, that is, W~B(a,b), the weight probability distribution function is 0<x<1, use the maximum likelihood estimation method to estimate the parameters (a, b), then the mathematical expectation of the weight eigenvector W is

[0054] Select weight ω i >p single prediction models of E(W) are taken as the better prediction models.

[0055] Step S5, combining historical electricity data, combining the better prediction models to form a comprehensive model, and using a genetic algorithm to obtain the weight of each model, and finally obtaining a comprehensive prediction model.

[0056] The final comprehensive prediction model is Where p is the number of better prediction models, ω i ' is the final obtained single model weight, y i is the predicted value of the ith single model. The architecture of the final comprehensive prediction model is as follows Figure 2 shown.

[0057] Step S6, using the comprehensive forecasting model to perform medium- and long-term electricity forecasting.

[0058] The prediction method proposed in this embodiment draws on the ideas of information theory, introduces the freshness function, improves the estimation process of the probability matrix, and establishes a comprehensive prediction model for electricity forecasting. The medium- and long-term electricity forecast using the comprehensive prediction model will be more accurate.

[0059] However, for some special periods and special regions, such as the occurrence of historically rare high (low) temperature weather in a certain month or a large-scale natural disaster (earthquake, flood, etc.) in a certain place, it is bound to have a huge impact on its electricity consumption. It is difficult to accurately and comprehensively cover such special information by estimating the probability matrix using only the freshness function and relative error.

[0060] In order to reflect the above special information in the forecasting process, the credibility of historical data can be evaluated by introducing the similarity between the historical period and the period to be predicted, and the relative error can be weighted in the probability matrix estimation process to obtain a probability matrix with more comprehensive information. However, this process requires a large amount of historical data and the processing process is relatively complicated. In comparison, experts engaged in power forecasting have rich practical experience. The judgment of information in the power forecasting process, especially the judgment of special information, can also serve as an important basis in the matrix estimation process, so as to make up for the information loss problem in the pure data estimation process.

[0061] Therefore, an improved implementation method is proposed: Figure 3 As shown, in step S23, based on the combination of the freshness function and the relative error, expert experience is also added to perform a new round of allocation of the credibility of the relative error of each historical data to obtain an improved estimated probability matrix.

[0062] When expert experience is involved, different evaluation methods are introduced, and each evaluation method obtains a corresponding credibility evaluation coefficient. For example, in some special areas, a credibility evaluation coefficient suitable for the local area can be added. After introducing expert experience, the credibility evaluation comprehensive coefficient r of the sth historical period is obtained s =f(r1,s ,r 2,s ,…,r m,s ), r s ∈[0,1], where r m,s It represents the credibility evaluation coefficient corresponding to the mth evaluation method introduced in the sth historical period.

[0063] In different regions and periods, the values ​​of each credibility evaluation coefficient and the distribution of each coefficient in the comprehensive coefficient are different, but for a specific region and a specific period, the credibility evaluation coefficient r s Can be uniquely determined based on expert experience.

[0064] Combined with freshness function α(1-α) t , relative error ε it and the comprehensive coefficient of credibility evaluation r s , estimate the probability matrix and get the weight ω i The absolute value estimation expression of Weight ω i The root mean square estimation expression of Based on the weight ω i , you can get the estimated probability matrix.

[0065] The evaluation methods introduced in this embodiment include a temperature evaluation method, a natural disaster evaluation method, and a policy evaluation method.

[0066] Temperature evaluation method:

[0067] Considering the existence of the accumulated temperature effect, the evaluation of temperature data should not only include the highest temperature, the lowest temperature, and the average temperature, but also the temperature fluctuation during a certain period. Therefore, the comprehensive temperature coefficient T s The formula for obtaining is Where, T s,max is the highest temperature in the sth historical period, T s,avg is the average temperature in the sth historical period, T s,min is the lowest temperature in the sth historical period, is the temperature variance in the sth historical period, c1 and c2 are constants, which are used to map the comprehensive temperature coefficient to the interval [0,1] and need to be determined according to the specific conditions of the prediction location and prediction period.

[0068] According to the obtained temperature comprehensive coefficient T s Based on the experience of experts, the temperature of each year in history is classified. Taking winter as an example, it is divided into three categories: warm winter, normal winter, and cold winter. The corresponding temperature comprehensive coefficients are 1, 0.5, and 0, respectively. The corresponding temperature credibility evaluation coefficient is r T,s =|T s -T f |, where rT,s is the temperature credibility evaluation coefficient of the sth historical period, T f It is the temperature credibility evaluation coefficient for the future forecast period evaluated by experts.

[0069] Natural disaster assessment methods:

[0070] Although the frequency of large-scale natural disasters is low, once they occur, they will have a great impact on electricity consumption. If there are natural disasters in the historical period, the corresponding electricity data of that period will deviate greatly from the normal value, which will affect the normal prediction. Therefore, the natural disaster credibility evaluation coefficient of the sth historical period is introduced.

[0071] Policy evaluation methods:

[0072] The introduction or revision of certain policies will also have an impact on electricity consumption. The policy credibility evaluation coefficient r of the sth historical period is introduced. P,s The specific quantitative value needs to be determined based on the specific conditions of a certain region and period.

[0073] The temperature credibility evaluation coefficient, natural disaster credibility evaluation coefficient and policy credibility evaluation coefficient are combined to obtain the final credibility evaluation comprehensive coefficient r s =f(r T,s ,r D,s ,r P,s ,…).

[0074] The performance of the comprehensive prediction model in this test method is verified by combining examples below.

[0075] Taking a provincial power grid as an example, a monthly electricity consumption forecast analysis was conducted. The historical data is the monthly electricity consumption of a provincial power grid from 2015 to December 2021. The single forecast model selects 20 commonly used monthly electricity consumption forecast models, as shown in the following table:

[0076] Linear Model Index Model 1 Index Model 2 Logarithmic Model Hyperbolic Model 1 Hyperbolic Model 2 Power function model Extended S-curve Extended GOMPERTZ Curve Extended Inverse Tangent Curve S-curve GOMPERTZ curve Parabolic Model N-th degree curve Artificial Neural Networks Moving Average Method Exponential Smoothing Grey system method Growth Rate Method Fuzzy Mathematics

[0077] Only the probability matrix of relative error estimation was used to screen 20 models, and 13 better prediction models were obtained, namely: linear model, exponential model 1, power function model, extended S-curve, extended GOMPERTZ curve, extended inverse tangent, parabola model, N-order curve, artificial neural network, moving average method, exponential smoothing method, grey system method, and growth rate method.

[0078] Since historical errors were not handled according to the principle of “larger near and smaller far”, models with poor recent prediction results were also included in the selected better prediction models. This result will be unfavorable to the recent electricity forecast results.

[0079] After considering the freshness function, a total of 6 better prediction models were selected, as shown in the following table:

[0080]

[0081] The results show that after the introduction of the freshness function, not only the models with large relative errors among the 20 models were eliminated, but also the models with good historical prediction results but poor recent prediction results were eliminated. The comparison of model predictions is shown in the "Comprehensive prediction model based on freshness function" in the following table.

[0082]

[0083]

[0084] In addition, since the electricity consumption in December 2019 was low, it was abnormal data, and the corresponding relative error was large. Even after being processed by the freshness function, it could not meet the principle of "large near and small far". Therefore, the improved prediction decision algorithm was used to evaluate the credibility of historical data by introducing expert experience and intervene in the probability matrix estimation process. In this example, the credibility levels are divided into "high, medium, and low", and the corresponding credibility is set to 0.9, 0.5, and 0.1 respectively. The model screening results after the improved prediction decision are shown in the "Comprehensive prediction model based on freshness function and improved prediction decision" in the above table.

[0085] Therefore, the present invention proposes an improved prediction decision algorithm based on the medium- and long-term changes in electricity volume, and establishes a comprehensive prediction model based on improved prediction decision. This model not only processes the impact of historical data according to the principle of "near is big and far is small" by introducing a freshness function, but also evaluates the credibility of historical data based on expert experience, and searches for historical data that is most similar to recent predictions, thereby obtaining a more accurate probability matrix and completing the model screening. The monthly prediction results of a provincial power grid show that the improved prediction decision algorithm can more accurately screen out models suitable for recent predictions, and the formed comprehensive prediction model has high accuracy and practicality.

[0086] Embodiment 2: The purpose of this embodiment is to provide an electric quantity prediction device based on a comprehensive prediction model, which is used to execute the prediction method of embodiment 1. Figure 4 As shown, the prediction device 200 includes: an acquisition unit 201 , a construction unit 202 and a prediction unit 203 .

[0087] The acquisition unit 201 is used to acquire historical power data and to acquire a commonly used single prediction model. Specifically, in combination with Embodiment 1, the acquisition unit 201 is used to execute step S1.

[0088] Construction unit 202 is used to construct a comprehensive prediction model. Specifically, construction unit 202 is used to define a probability matrix, combine the freshness function and the historical prediction error of a single prediction model, and obtain an estimated probability matrix; to analyze the estimated probability matrix, and use the eigenvector method to obtain the weight eigenvector; to use the weight eigenvector to represent the approximate weight probability distribution function, obtain the mathematical expectation of the weight eigenvector, and screen out a better prediction model; to combine the historical electricity data, combine the better prediction models to form a comprehensive model, and use a genetic algorithm to obtain the weights of each model, and finally obtain a comprehensive prediction model. In combination with Example 1, construction unit 202 is used to execute steps S2-S5.

[0089] The prediction unit 203 is used to perform medium- and long-term electricity prediction using a comprehensive prediction model. Specifically, in combination with Embodiment 1, the prediction unit 203 is used to execute step S6.

[0090] Regarding the device in this embodiment, the specific manner in which each unit performs operations has been described in detail in Example 1 and will not be elaborated here.

[0091] Those skilled in the art should understand that for the convenience and simplicity of description, the present application only uses the division of the above-mentioned functional modules or units as examples in the device embodiments. In actual applications, the above-mentioned functions can be assigned to different functional modules or units as needed, that is, the internal structure of the device can be divided into different functional modules or units to complete all or part of the functions described above.

[0092] Example 3: See Figure 5 The purpose of this embodiment is to provide an electronic device 300, including at least one processor 301 and one or more memories 302 for storing executable instructions of the processor 301. The processor 301 is used to execute the instructions in the memory 302 to implement the prediction method described in Example 1.

[0093] The electronic device 300 further includes a bus 304 , and the processor 301 and the memory 302 are connected to each other via the bus 304 , or are connected to each other in other ways.

[0094] The processor 301 is a central processing unit (CPU), a general-purpose processor, a network processor (NP), a digital signal processor (DSP), a microprocessor, a microcontroller, a programmable logic device (PLD), or any combination thereof. The processor 301 may also be other devices with processing functions, such as circuits, devices, or software modules, without limitation. In one example, the processor 301 may include one or more CPUs, such as Figure 5 CPU0 and CPU1 in.

[0095] Among them, the memory 302 can be a read-only memory (ROM) or other types of static storage devices that can store static information and / or instructions, or a random access memory (RAM) or other types of dynamic storage devices that can store information and / or instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), magnetic disk storage medium or other magnetic storage devices, etc., without limitation.

[0096] It should be noted that the memory 302 may exist independently of the processor 301 or may be integrated with the processor 301. The memory 302 may be used to store instructions or program codes or some data, etc. The memory 302 may be located inside the electronic device 300 or outside the electronic device 300, without limitation.

[0097] As an optional implementation, the electronic device 300 also includes a communication interface 303. The communication interface 303 is a wired interface (or port), such as a fiber distributed data interface (FDDI), a gigabit Ethernet interface (GE), etc. Alternatively, the communication interface 303 is a wireless interface. The communication interface 303 can be a module, a circuit, a communication interface, or any device capable of achieving communication. The communication interface 303 is used to communicate with other devices or other communication networks, which can be Ethernet, a radio access network (RAN), a wireless local area network (WLAN), etc.

[0098] As an optional implementation, the electronic device 300 further includes an input device 305 and an output device 306. Exemplarily, the input device 305 is a keyboard, a mouse, a microphone, a joystick, and the like, and the output device 306 is a display screen, a speaker, and the like.

[0099] It should be noted that the electronic device 300 may be a desktop computer, a portable computer, a network server, a mobile phone, a tablet computer, a wireless terminal, an embedded device, a chip system or a Figure 5 In addition, Figure 5 The components shown in the figure do not constitute a limitation on the terminal device, except Figure 5 In addition to the components shown, the electronic device 300 may include more or fewer components than shown, or combine certain components, or arrange the components differently.

[0100] Embodiment 4: The purpose of this embodiment is to provide a computer-readable storage medium.

[0101] All or part of the processes in the above method embodiments may be completed by computer instructions to instruct relevant hardware. The program may be stored in the computer-readable storage medium. When the program is executed, the prediction method described in Example 1 may be implemented.

[0102] The computer-readable storage medium may be an internal storage unit of the electronic device of Example 3, such as a hard disk or a memory. The computer-readable storage medium may also be an external storage device of the electronic device, such as a plug-in hard disk, a smart memory card (smart media card, SMC), a secure digital (secure digital, SD) card, a flash card (flash card), etc. Further, the computer-readable storage medium may also include both an internal storage unit of the above-mentioned electronic device and an external storage device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium may also be used to temporarily store data that has been output or is to be output.

[0103] In conjunction with the several embodiments provided in this application, it should be understood that the provided devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of the modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple modules or units or components can be combined or integrated into another device, or some features can be ignored or not executed.

[0104] In addition, in the embodiment of the device of the present application, each functional module or unit can be integrated into one unit, or each module or unit can exist physically separately, or two or more modules or units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional modules.

[0105] If the above-mentioned integrated unit is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium, including several instructions to enable a device (which can be a single-chip microcomputer, chip, etc.) or a processor (processor) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard drives, ROM, RAM, magnetic disks or optical disks.

[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Other modifications or equivalent substitutions made to the technical solution of the present invention by ordinary technicians in the field should be included in the scope of the claims of the present invention as long as they do not depart from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for power consumption prediction based on a comprehensive prediction model, characterized in that: include: Collect historical electricity data and select a commonly used single prediction model; Define the probability matrix, obtain the historical prediction error of a single prediction model, and combine it with the freshness function to obtain the estimated probability matrix; Analyze the estimated probability matrix and use the eigenvector method to obtain the weight eigenvector; Use the weight feature vector to represent the approximate weight probability distribution function, obtain the mathematical expectation of the weight feature vector, and screen out the better prediction model; Combined with historical electricity data, the better prediction models are combined to form a comprehensive model, and the genetic algorithm is used to obtain the weight of each model to obtain a comprehensive prediction model; Use comprehensive forecasting models to conduct medium- and long-term electricity forecasts.

2. The power consumption prediction method based on the comprehensive prediction model according to claim 1 is characterized in that: Methods for obtaining historical forecast errors of a single forecast model include: The prediction error of the t-th prediction value of the i-th single prediction model in is the tth prediction value of the ith single prediction model, y it is the tth true value of the ith single prediction model.

3. The method for power consumption prediction based on a comprehensive prediction model according to claim 2 is characterized in that: Methods for obtaining the estimated probability matrix include: The weight of the i-th single model or where ε it is the historical forecast error, α(1-α) t is the freshness function, t=1,2,...,n, n is the number of historical electricity data corresponding to a single prediction model; Get the estimated probability matrix based on the weights of all individual prediction models.

4. The method for predicting power consumption based on a comprehensive prediction model according to claim 2 is characterized in that: Methods for obtaining the estimated probability matrix include: Introduce the evaluation method to obtain the comprehensive coefficient of credibility evaluation r s =f(r 1,s ,r 2,s ,…,r m,s ), r s ∈[0,1], where r m,s It represents the credibility evaluation coefficient corresponding to the mth evaluation method introduced in the sth historical period; The weight of the i-th single model or where ε it is the historical forecast error, α(1-α) t is the freshness function, t=1,2,...,n, r s is the comprehensive coefficient of credibility evaluation, n is the number of historical electricity data corresponding to a single prediction model; Get the estimated probability matrix based on the weights of all individual prediction models.

5. The method for power consumption prediction based on the comprehensive prediction model according to claim 4 is characterized in that: Reliability evaluation comprehensive coefficient r s =f(r T,s ,r D,s ,r P,s ,…),in: r T,s is the temperature credibility evaluation coefficient of the sth historical period, r T,s =|T s -T f |, T s is the comprehensive temperature coefficient, T f is the temperature credibility evaluation coefficient evaluated by experts, T s,max is the highest temperature in the sth historical period, T s,avg is the average temperature in the sth historical period, T s,min is the lowest temperature in the sth historical period, is the temperature variance in the sth historical period, c1 and c2 are constants used to map the temperature comprehensive coefficient to the interval [0,1]; r D,s is the credibility evaluation coefficient of natural disasters in the sth historical period, r P,s is the policy credibility evaluation coefficient of the sth historical period.

6. The method for electricity consumption prediction based on a comprehensive prediction model according to claim 1, characterized in that: The comprehensive prediction model is expressed as Where p is the number of better prediction models, ω i ' is the final weight of the i-th single model, y i is the predicted value of the ith single model.

7. The method for electricity consumption prediction based on a comprehensive prediction model according to claim 1, characterized in that: The approximate weight probability distribution function is The mathematical expectation of the weighted eigenvector W is The better prediction model is the single prediction model with a weight greater than E(W).

8. An electric quantity prediction device based on a comprehensive prediction model, characterized in that: It includes an acquisition unit, a construction unit and a prediction unit; The acquisition unit is used to acquire historical power data and to acquire a commonly used single prediction model; The construction unit is used to obtain an estimated probability matrix based on the freshness function and the historical prediction error of the single prediction model; to analyze the estimated probability matrix and obtain the weight eigenvector using the eigenvector method; to use the weight eigenvector to represent the approximate weight probability distribution function, obtain the mathematical expectation of the weight eigenvector, and screen out a better prediction model; to combine the better prediction models into a comprehensive model in combination with the historical electricity data, and use the genetic algorithm to obtain the weights of each model, and finally obtain the comprehensive prediction model; The prediction unit is used to perform medium- and long-term electricity quantity prediction using a comprehensive prediction model.

9. An electronic device, characterized in that: It comprises a processor and a memory; the memory is used to store executable instructions, and the processor is used to execute the instructions to implement the prediction method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The readable storage medium stores instructions, and when the instructions are executed, the prediction method according to any one of claims 1 to 7 is implemented.

Images