Power load intelligent prediction method based on big data

Abstract

The invention relates to the technical field of power prediction, and provides an intelligent power load prediction method based on big data, and the method is characterized in that the method comprises the following steps: S10, obtaining historical power load records and historical power behavior records of all power utilization units in a region; s20, establishing a power probability model based on historical power behavior records; s30, establishing a power load prediction model based on the power probability model; s40, identifying parameters of the power probability model and parameters of the power load prediction model based on historical power behavior records; and S50, inputting a historical power load record to the power load prediction model to output a power load prediction value. According to the invention, the power probability model and the load prediction model are established and the respective parameters are identified, so that the intelligent prediction of the power load is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of power forecasting, and in particular relates to a method for intelligent power load forecasting based on big data. Background Art

[0002] In modern power systems, power load forecasting is a crucial task. Accurate load forecasting can help power companies optimize resource allocation, improve power supply reliability and reduce operating costs. Traditional power load forecasting methods are mostly based on empirical rules or simple statistical analysis, and usually rely on linear regression or time series models of historical data. These methods are inadequate in dealing with complex nonlinear relationships, seasonal changes and emergencies.

[0003] In addition, with the increasing complexity of power load forecasting models, parameter selection and optimization become more difficult, which can easily lead to overfitting. In particular, traditional methods do not consider the impact of different types of days due to parameter selection and optimization, and cannot better capture the changes in electricity consumption behavior in different time periods, resulting in low accuracy of the forecasting model. Summary of the invention

[0004] In view of the above-mentioned defects of the prior art, the present invention proposes an intelligent prediction method for power load based on big data. The technical solution steps designed by the present invention include:

[0005] S10: Obtain the historical power load records and historical power behavior records of each power user in the region;

[0006] S20: Establish a power probability model based on historical power behavior records;

[0007] S30: Establishing a power load prediction model based on the power probability model;

[0008] S40: Identifying parameters of a power probability model and parameters of a power load forecasting model based on historical power behavior records;

[0009] S50: Input historical power load records into the power load prediction model to output a power load prediction value.

[0010] Preferably, the S20 includes:

[0011] The component functions of the power probability model are defined based on historical power behavior records, including the working day component function and the holiday component function. The formula is as follows:

[0012]

[0013] In the formula, f WD，n (x k |d n , en ) is the working day component function, f HD，n (c|d n , e n ) is the holiday component function, d n is the logarithmic expectation of the nth component function, e n is the standard deviation of the nth component function, x k is the observed value of power consumption.

[0014] Preferably, the S30 includes:

[0015] The power load of each power user in the region is weighted and summed, based on the dependent parameter a represented by the distribution. n d n and e n The power probability model is used to construct the overall power load forecasting model. The formula is as follows:

[0016]

[0017] t g，m ∈T=[t g，1 , t g，2 ,…,t g，M ]~f A (a n , d n , e n )

[0018] In the formula, f B (x k ) is the power load forecasting model, P m (x k ) is the power load of the power user, M is the number of power users, is the total power mean of the electricity consumption unit, is the average total electricity consumption time of the electricity-consuming unit, f is the electricity consumption time sequence of each electricity-consuming unit in the region, and ~f A (a n , d n , e n ) is the dependent parameter a of the distribution representation n d n and e n The power probability model.

[0019] Preferably, the parameters for identifying the power probability model in S40 include:

[0020] The expectation and maximization processing of the holiday component function, and then the expectation and maximization processing of the working day component function are used as a calculation process. By repeating the iterative calculation process until convergence, the parameter a of the power probability model of the local optimal solution is obtained. n dn and e n .

[0021] Preferably, the holiday component function is subjected to expectation processing and maximization processing, and the formula is as follows:

[0022]

[0023] In the formula, Expect k，n is the expectation of the holiday component function, Maximize is the maximization of the holiday component function, and K is the total amount of data for obtaining the optimal solution parameters.

[0024] Preferably, the working day component function is subjected to expectation processing and maximization processing, and the formula is as follows:

[0025]

[0026] In the formula, Expect k,n ′ is the expectation of the workday composition function, and Maximize′ is the maximization of the workday composition function.

[0027] Preferably, the parameters for identifying the power load forecasting model in S40 include:

[0028] Create an iterative parameter matrix and output f B (x k ) function, establish the position search of the scout, follower and predator, continuously update the position of the iterative parameter matrix and output the corresponding f B (x k ) function, find the best combination of iterative parameters, and determine the parameters of the power load forecasting model and M;

[0029] The iterative parameter matrix is ​​established and output f B (x k ) functions include:

[0030]

[0031] Where Matrix(v) is the iteration parameter matrix, f B (x k ), Parameters for the first set of iterations and M, V are the maximum number of iterations.

[0032] Preferably, the position search formula of the scout is as follows:

[0033]

[0034] In the formula, is the position of the scout after the search, is the position of the scout before the search, β is a uniform random number in [0, 1], γ represents the predator as a random number in [0, 1], δ is a standard normal random number, and Warn is the warning value of the scout discovering the predator.

[0035] Preferably, the follower's position search formula is as follows:

[0036]

[0037] In the formula, is the follower position after searching, is the follower position before the search, The iterative parameter position with the worst optimization effect in the current search. is the iterative parameter position with the best optimization effect in the current search, and ε is a random number in [0, 1].

[0038] Preferably, the predator position search formula is as follows:

[0039]

[0040] In the formula, is the position of the predator after searching, is the predator's position before searching, f is the current search B (x k ) function, f best is the optimal f B (x k ) function, f worst The lowest f B (x k ) function, θ is the standard normal distribution number, and σ takes the value of |-1, 1|.

[0041] Beneficial effects:

[0042] 1. By processing the working day and holiday components separately, the model can better capture the changes in electricity consumption patterns and significantly improve the prediction accuracy of power load;

[0043] 2. This application adopts the method of expectation maximization processing and iterative calculation, so that the parameters of the power probability model are better estimated, and the stability and robustness of the model are improved;

[0044] 3. By establishing a parameter matrix and dynamically updating it, the model can quickly adapt to real-time data changes, ensuring the prediction effect in different seasons, holidays and emergencies. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a flow chart of a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0046] The embodiments of the present invention are described in detail below. The following embodiments are implemented on the premise of the technical solution of the present invention, and detailed implementation methods and specific operation processes are given, but the protection scope of the present invention is not limited to the following embodiments.

[0047] The present invention designs a method for intelligent prediction of power load based on big data, such as Figure 1 As shown, the technical solution includes the following steps, specifically including:

[0048] S10: Obtain the historical power load records and historical power behavior records of each power user in the region;

[0049] S20: Establish a power probability model based on historical power behavior records;

[0050] S30: Establishing a power load prediction model based on the power probability model;

[0051] S40: Identifying parameters of a power probability model and parameters of a power load forecasting model based on historical power behavior records;

[0052] S50: Input historical power load records into the power load prediction model to output a power load prediction value.

[0053] Specifically, the historical power load record is the data of actual power consumption of power users in a certain period of time;

[0054] The historical electricity behavior records are the user's electricity usage behavior pattern data, including but not limited to peak hour electricity usage habits, the impact of specific events (such as holidays, weather changes) on electricity usage, and the user's electricity equipment types and usage frequency.

[0055] Preferably, S20 includes:

[0056] The component functions of the power probability model are defined based on historical power behavior records, including the working day component function and the holiday component function. The formula is as follows:

[0057]

[0058] In the formula, f WD，n (x k |d n , e n ) is the working day component function, f HD，n (c|d n , e n ) is the holiday component function, d nis the logarithmic expectation of the nth component function, e n is the standard deviation of the nth component function, x k is the observed value of power consumption.

[0059] Specifically, the formula of the power probability model is as follows:

[0060]

[0061]

[0062] In the formula, f A (x k ) is the power probability model, N is f A The number of component functions in (x), a n f A (x k ) is the weight of the nth component function, f n (x|b n ) is f A (c) The nth component function, b n f A (x k ) is the parameter value of the component function in .

[0063] In addition, n is the logarithmic expectation of the nth component function, reflecting the central tendency of electricity consumption (in logarithmic form), e n is the standard deviation of the nth component function, reflecting the dispersion of power consumption data, x k is the observed value of power consumption. It should be noted that when it is in the weekday component function, it is the power usage on weekdays, and when it is in the holiday component function, it is the power usage on holidays. When there is no power behavior restriction on weekdays and holidays, the observed value is the total power usage. Through the difference in power behavior on weekdays and holidays, a power probability model is established to predict power demand.

[0064] Preferably, S30 includes:

[0065] The power load of each power user in the region is weighted and summed, based on the dependent parameter a represented by the distribution. n d n and e n The power probability model is used to construct the overall power load forecasting model. The formula is as follows:

[0066]

[0067] In the formula, f B (x k ) is the power load forecasting model, P m (xk ) is the power load of the power user, M is the number of power users, is the total power mean of the electricity consumption unit, is the average total electricity consumption time of the electricity-consuming unit, T is the electricity consumption time sequence of each electricity-consuming unit in the region, and ~f A (a n , d n , e n ) is the dependent parameter a of the distribution representation n d n and e n The power probability model.

[0068] Specifically, according to the power load forecasting model formula, the parameters that need to be determined are M.a n d n and e n , as the training time goes by, the parameters of the power probability model and the power load forecasting model will change accordingly. Due to the rapid growth of data volume and the characteristics of real-time interaction, this scheme realizes online updating of the parameters of the power load forecasting model and the power probability model by accessing the data interface. In view of the significant difference in the usage patterns of the weekday component function and the holiday component function, two independent models are used to determine their respective update frequencies, so as to adjust the corresponding parameters independently and alternately.

[0069] Preferably, the parameters for identifying the power probability model in S40 include:

[0070] The expectation and maximization processing of the holiday component function, and then the expectation and maximization processing of the working day component function are used as a calculation process. By repeating the iterative calculation process until convergence, the parameter a of the power probability model of the local optimal solution is obtained. n d n and e n .

[0071] Preferably, the holiday component function is subjected to expectation processing and maximization processing, and the formula is as follows:

[0072]

[0073] In the formula, Expect k，n is the expectation of the holiday component function, Maximize is the maximization of the holiday component function, and K is the total amount of data for obtaining the optimal solution parameters.

[0074] Preferably, the working day component function is subjected to expectation processing and maximization processing, and the formula is as follows:

[0075]

[0076] In the formula, Expect k，n ′ is the expectation of the workday composition function, and Maximize′ is the maximization of the workday composition function.

[0077] Specifically, the expected value of the holiday component function for power load is calculated, which helps to evaluate the expected level of power demand in a specific time period, and finds the parameters that maximize the expected value through the optimization algorithm. This process helps to determine the model parameters that best explain the load behavior under given data. This process is repeated, and the parameters are updated multiple times to gradually approach the optimal solution until convergence, and the optimal parameters of the power probability model are obtained. n d n and e n , which in disguise determines the parameters of the final power load forecasting model, and for the other parameters of the power load forecasting model and M, according to the following step S40, by establishing an iterative parameter matrix, output f B (x k ) function, searches according to the positions of the scout, follower, and predator, iterates and updates the position of the parameter matrix and outputs the corresponding f B (x k ) function to find the best combination of iterative parameters, that is, to determine the parameters and M.

[0078] Preferably, the parameters for identifying the power load forecasting model in S40 include:

[0079] Create an iterative parameter matrix and output f B (x k ) function, establish the position search of the scout, follower and predator, continuously update the position of the iterative parameter matrix and output the corresponding f B (x k ) function, find the best combination of iterative parameters, and determine the parameters of the power load forecasting model and M;

[0080] Create an iterative parameter matrix and output f B (x k ) functions include:

[0081]

[0082] Where Matrix(v) is the iteration parameter matrix, f B (x k ), Parameters for the first set of iterations and M, V are the maximum number of iterations.

[0083] Preferably, the scout's position search formula is as follows:

[0084]

[0085] In the formula, is the position of the scout after the search, is the position of the scout before the search, β is a uniform random number in [0, 1], γ represents the predator as a random number in [0, 1], δ is a standard normal random number, and Warn is the warning value of the scout discovering the predator.

[0086] Preferably, the follower's position search formula is as follows:

[0087]

[0088] In the formula, is the follower position after searching, is the follower position before the search, The iterative parameter position with the worst optimization effect in the current search. is the iterative parameter position with the best optimization effect in the current search, and ε is a random number in [0, 1].

[0089] Preferably, the predator's position search formula is as follows:

[0090]

[0091] In the formula, is the position of the predator after searching, is the predator's position before searching, f is the current search B (x k ) function, f best is the optimal f B (x k ) function, f worst The lowest f B (x k ) function, θ is the standard normal distribution number, and σ takes the value of |-1, 1|.

[0092] Specifically, due to f B (x k ) is the power load forecasting model, so the goal is to minimize f B (x k ) Optimize the search of Matrix to make the model's prediction of power load closer to the actual value. Optimize the search of Matrix includes: establish Scout, Follower and Predator. Scout records the current search f B (xk ), the follower records the f of the scout B (x k ) solution and assign the predator Predator. In addition, μ is used in the formula to affect the search position update of the predator, and the optimization avoids falling into the local optimum.

[0093] The preferred specific embodiments of the present invention are described in detail above. It should be understood that ordinary technicians in the field can make many modifications and changes based on the concept of the present invention without creative work. Therefore, all technical solutions that can be obtained by technicians in the technical field based on the concept of the present invention through logical analysis, reasoning or limited experiments on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A method for intelligent prediction of power load based on big data, characterized in that: The following steps are involved: S10: Obtain the historical power load records and historical power behavior records of each power user in the region; S20: Establish a power probability model based on historical power behavior records; S30: Establishing a power load prediction model based on the power probability model; S40: Identifying parameters of a power probability model and parameters of a power load forecasting model based on historical power behavior records; S50: Input historical power load records into the power load prediction model to output a power load prediction value.

2. According to the big data-based intelligent power load prediction method of claim 1, it is characterized in that: The S20 includes: The component functions of the power probability model are defined based on historical power behavior records, including the working day component function and the holiday component function. The formula is as follows: In the formula, f WD,n (x k |d n ,e n ) is the working day component function, f HD,n (c|d n ,e n ) is the holiday component function, d n is the logarithmic expectation of the nth component function, e n is the standard deviation of the nth component function, x k is the observed value of power consumption.

3. The method for intelligent prediction of power load based on big data according to claim 2 is characterized in that: The S30 includes: The power load of each power user in the region is weighted and summed, based on the dependent parameter a represented by the distribution. n d n and e n The power probability model is used to construct the overall power load forecasting model. The formula is as follows: In the formula, f B (x k ) is the power load forecasting model, P m (x k ) is the power load of the power user, M is the number of power users, is the total power mean of the electricity consumption unit, is the average total electricity consumption time of the electricity-consuming unit, T is the electricity consumption time sequence of each electricity-consuming unit in the region, and ~f A (a n , d n , e n ) is the dependent parameter a of the distribution representation n d n and e n The power probability model.

4. The method for intelligent prediction of power load based on big data according to claim 3 is characterized in that: The parameters for identifying the power probability model in S40 include: The expectation and maximization processing of the holiday component function, and then the expectation and maximization processing of the working day component function are used as a calculation process. By repeating the iterative calculation process until convergence, the parameter a of the power probability model of the local optimal solution is obtained. n d n and e n .

5. The method for intelligent prediction of power load based on big data according to claim 4 is characterized in that: The holiday component function is subjected to expectation processing and maximization processing, and the formula is as follows: In the formula, Expect k，n is the expectation of the holiday component function, Maximize is the maximization of the holiday component function, and K is the total amount of data for obtaining the optimal solution parameters.

6. The method for intelligent prediction of power load based on big data according to claim 5 is characterized in that: The expectation processing and maximization processing of the working day component function are performed, and the formula is as follows: In the formula, Expect k,n ′ is the expectation of the workday composition function, and Maximize′ is the maximization of the workday composition function.

7. The method for intelligent prediction of power load based on big data according to claim 4 is characterized in that: The parameters for identifying the power load forecasting model in S40 include: Create an iterative parameter matrix and output f B (x k ) function, establish the position search of the scout, follower and predator, continuously update the position of the iterative parameter matrix and output the corresponding f B (x k ) function, find the best combination of iterative parameters, and determine the parameters of the power load forecasting model and M; The iterative parameter matrix is ​​established and output f B (x k ) functions include: Where Matrix(v) is the iteration parameter matrix, f B (x k ), Parameters for the first set of iterations and M, V are the maximum number of iterations.

8. The method for intelligent prediction of power load based on big data according to claim 7 is characterized in that: The position search formula of the scout is as follows: In the formula, is the position of the scout after the search, is the position of the scout before the search, β is a uniform random number in [0, 1], γ represents the predator as a random number in [0, 1], δ is a standard normal random number, and Warn is the warning value of the scout discovering the predator.

9. The method for intelligent prediction of power load based on big data according to claim 8 is characterized in that: The follower position search formula is as follows: In the formula, is the follower position after searching, is the follower position before the search, The iterative parameter position with the worst optimization effect in the current search. is the iterative parameter position with the best optimization effect in the current search, and ε is a random number in [0, 1].

10. The method for intelligent prediction of power load based on big data according to claim 9, characterized in that: The position search formula of the predator is as follows: In the formula, is the position of the predator after searching, is the predator's position before searching, f is the current search B (x k ) function, f best is the optimal f B (x k ) function, f worst The lowest f B (x k ) function, θ is the standard normal distribution number, and σ is in the range of [-1, 1].

Images