A method for power load forecasting based on cloud computing

Abstract

The invention discloses a power load forecasting method based on cloud computing, which uses an orthogonalized least squares OLS algorithm to calculate a central vector, sets a threshold and precision parameters to ensure the accuracy of the calculation result, and then integrates a k-means clustering algorithm into a MapReduce architecture in a cloud computing environment of Hadoop to calculate a correction value of the central vector, obtain an extended parameter according to the correction value of the central vector, and ensure the accuracy of the result according to an equation, then obtain a radial basis function value using the central vector and the extended parameter, and obtain a weight value therefrom, and finally substitute the radial basis function value and the weight value into an RBF neural network to calculate the prediction result. The present application constructs a method for power load forecasting in a Hadoop environment based on an OLS algorithm, a k-means clustering algorithm, a MapReduce architecture and an RBF neural network, which improves the running speed while ensuring the accuracy of the prediction.

Description

Technical Field

[0001] The present invention relates to the field of power load data prediction, and in particular to a power load prediction method based on cloud computing. Background Art

[0002] The rational use of power load data plays an important role in maintaining the stability of the power grid system (Short-term power load forecasting based on SRDHLGRU neural network and differential error compensation [J / OL]. China Test: 1-7 [2022-07-07]). However, due to the complexity and variability of power load data, traditional power load data processing systems have been unable to properly process power load data. At this time, there is an urgent need for a processing system that can make full use of power load data. The continuous development of cloud computing technology has made the realization of this system possible. Cloud computing technology is a cloud service method based on software and hardware resources and using the Internet as a service channel. It has been widely used in various fields (Microservice architecture of Web applications in cloud environments [J]. Electronic Technology and Software Engineering, 2019 (15): 131-132). For power grid companies, cloud computing can greatly improve their mastery of power grid resources, enable them to understand user needs, and then improve the power system network, which plays an important role in the development of the power system (Power load forecasting based on big data platform [J]. Modern Electronic Technology, 2018, 41 (20): 153-156). Cloud computing plays a unique role in the field of power load forecasting. The use of cloud computing technology can greatly speed up the prediction of power load data and improve the accuracy of power load forecasting to a certain extent. Some scholars have studied the power load forecasting system from the perspective of cloud computing, such as: Application of short-term load forecasting model based on OpenStack neural network in orderly power consumption management [J]. Smart Grid, 2017, 5(07): 683-688 and Discussion on the development of medium- and long-term load forecasting methods for power systems [J]. Shandong Electric Power Technology, 2017, 44(01): 13-18+40, but it has problems such as limited scope of application and lack of universality.

[0003] When conducting research, the above scholars all adopted different methods to combine various algorithms or models with cloud computing platforms or architectures. The results obtained improved the accuracy and speed of power load forecasting based on cloud computing, and to a certain extent improved the performance of the power system. However, there are certain problems in their research process, such as the constructed system is too complex, the system is not universal, and the scope of application is limited. In response to these problems, this application proposes a solution. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a cloud computing-based power load forecasting method that can make full use of power load data and improve the efficiency and accuracy of power load forecasting.

[0005] Technical solution: The power load forecasting method based on cloud computing described in the present invention is characterized by comprising the following steps:

[0006] S1: Select power load data;

[0007] S2: Use the orthogonal least squares method (OLS) to find the center vector of the radial basis function, and set a threshold to detect the center vector of the radial basis function. If the threshold is not met, adjust the number of neurons in the hidden layer and recalculate;

[0008] S3: Use the k-means algorithm in a cloud computing environment and combine it with the MapReduce architecture to adjust the center vector of the radial basis function and calculate the expansion parameters. Use the equation to calculate the expansion parameters. If the equation is satisfied, proceed to the next step. If not, reselect the input data.

[0009] S4: Calculate weights and run RBF neural network;

[0010] S5: Output power load data prediction value.

[0011] Preferably, the specific calculation steps of S2 are as follows:

[0012] S2.1: Take the actual power load data as the sample number T, and set A t is the input vector, B t is the expected vector, both are vectors of the same dimension, e t is the absolute error between the input vector and the input vector, the value range of t is 1, 2, …T, and the formula is:

[0013]

[0014] S2.2: Set the number of neurons in the hidden layer of the RBF neural network to H, which is usually 1 in the initial case, and w h is the weight matrix, O h (t) is the regression operator, and the value range of h is 1, 2, ...H, so B t The calculation formula is:

[0015]

[0016] S2.3: Express equation (2) as a matrix:

[0017] B=OWH+e (3)

[0018] S2.4: Use the orthogonal triangular QR decomposition method to express O as a matrix Q and an upper triangular matrix R. The specific formula is as follows:

[0019] B=QRWH+e (4)

[0020] S2.5: Express RW as q h , according to the orthogonal least squares method, the least squares solution is obtained, the formula is as follows:

[0021]

[0022] S2.6: Use the Schmidt orthogonalization method to reduce the error and determine the value s h , the specific formula is as follows:

[0023]

[0024] S2.7: Set the threshold z and select the input vector that satisfies formula (7) as the center vector c k , the value of k is an integer less than T:

[0025]

[0026] If there is no center vector c that meets the conditions k , then adjust B t and H are recalculated.

[0027] Preferably, the specific calculation steps of S3 are as follows:

[0028] S3.1: Combine the power load data sample set and the center vector c confirmed by OLS k Input, denoted as c k (n), n represents the number of iterations;

[0029] S3.2: Calculate the Euclidean distance D, A t (n) indicates that in t The randomly selected data represents the input data of the nth iteration. If the D value between the input vector and a certain center vector is the smallest, then this input vector is included in this cluster point set J k The specific formula is as follows:

[0030] D=arg||A t (n)-c k (n)|| (8)

[0031] S3.3: Get the new cluster center c k (n)', Represents the set J k The number of neutron subsets is calculated as follows:

[0032]

[0033] S3.4: If equation (10) is satisfied, then c k (n)' is the new center vector, and the process goes to step S3.5. If the condition is not satisfied, the process goes back to step S3.2 and randomly extracts data again:

[0034] c k (n)'=c k (n+1)' (10)

[0035] S3.5: Calculate the distance D between the center vectors c , X represents the overlap coefficient n / 2, and the calculation formula of the coefficient S is as follows:

[0036] D c =||c k (n)'-c k || (11)

[0037] S=D c X (12)

[0038] Preferably, the cloud computing environment in S3 is a Hadoop cloud computing environment.

[0039] Preferably, calculating the weight in S4 comprises the following steps:

[0040] S4.1: Calculate radial basis function, C h ' indicates that this neuron is A t The corresponding center vector, the specific formula is as follows:

[0041]

[0042] S4.2: Calculate the weight w th :

[0043] w th =(F T F)A t F (14)

[0044] Preferably, the power load data prediction value output in S5 is the output value Y of the RBF neural network:

[0045]

[0046] Beneficial effects:

[0047] In the Hadoop environment, this application constructs a power load forecasting method for a power load forecasting system based on cloud computing based on the OLS algorithm, k-means clustering algorithm, MapReduce architecture and RBF neural network. This method improves the operating speed while ensuring the accuracy of the prediction, and can better process large amounts of data, thereby improving the operating speed and prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a flow chart of this application. DETAILED DESCRIPTION

[0049] The present application is further described below in conjunction with specific embodiments. The embodiments are only used to explain the present invention and do not limit the protection scope of the present invention.

[0050] like Figure 1 As shown, it is a flow chart of the present application. In this embodiment, it specifically includes the following steps:

[0051] S1: Select power load data;

[0052] S2: Use the orthogonal least squares method OLS to find the center vector of the radial basis function, and set a threshold to detect the center vector of the radial basis function. If the threshold is not met, adjust the number of neurons in the hidden layer and recalculate. Specifically:

[0053] S2.1: Take the actual power load data as the sample number 100, and set A t is a three-dimensional input vector, B t is the three-dimensional expected vector, both of which are vectors of the same dimension, e t is the absolute error between the input vector and the input vector, the value range of t is 1, 2, ... 100, and the formula is:

[0054]

[0055] S2.2: Set the number of neurons in the hidden layer of the RBF neural network to 1. In this case, the value of h is 1, and w h is the weight matrix, O h (t) is the matrix of regression operator 3*100, and we can get B t The calculation formula is:

[0056]

[0057] S2.3: Express equation (2) as a matrix:

[0058] B=OWH+e (3)

[0059] S2.4: Use the orthogonal triangular QR decomposition method to represent O as a 3*3-order matrix Q and a 3rd-order upper triangular matrix R. The specific formula is as follows:

[0060] B=QRWH+e (4)

[0061] S2.5: Express RW as q h , according to the orthogonal least squares method, the least squares solution is obtained, the formula is as follows:

[0062]

[0063] S2.6: Use the Schmidt orthogonalization method to reduce the error and determine the value s h , the specific formula is as follows:

[0064]

[0065] S2.7: Set the threshold to 0.9 and select the input vector that satisfies formula (7) as the center vector c k :

[0066]

[0067] If there is no center vector c that meets the conditions k , then adjust B t And record H as 2,3… and recalculate.

[0068] S3: Use the k-means algorithm in the Hadoop cloud computing environment, combined with the MapReduce architecture to adjust the central vector of the radial basis function and calculate the expansion parameters. Use the equation to calculate the expansion parameters. If the equation is satisfied, proceed to the next step. If it is not satisfied, reselect the input data. Specifically:

[0069] S3.1: Combine the power load data sample set and the center vector c confirmed by OLS k Input, denoted as c k (n), n represents the number of iterations, assuming the maximum iteration is 10;

[0070] S3.2: Calculate the Euclidean distance D, A t (3) indicates that in A t The randomly selected data in represents the input data of the third iteration, such as the input vector A t (3) If the D value with a certain center vector c1(3) is the smallest, then this input vector is classified into this cluster point set J1. The specific formula is as follows:

[0071] D=arg||A t (n)-c k (n)|| (8)

[0072] S3.3: Get the new cluster center c k (n)', Represents the set J k The number of subsets in the , if the input vector with the smallest D value with c1(10) is A2(10)A3(10), then is 2, and the calculation formula is as follows:

[0073]

[0074] S3.4: If equation (10) is satisfied, then c k (10)' is the new center vector, such as c k (10)'=c k (11) Enter step S3.5. If the result is not satisfied, return to step S3.2 and randomly extract data again:

[0075] c k (n)'=c k (n+1)' (10)

[0076] S3.5: Calculate the distance D between the center vectors c , X represents the overlap coefficient 5, and the calculation formula of the coefficient S is as follows:

[0077] D c =||c k (n)'-c k || (11)

[0078] S=D c X (12)

[0079] S4: Calculate weights and run the RBF neural network, specifically:

[0080] S4.1: Calculate radial basis function, C h ' indicates that this neuron is A t The corresponding center vector c2(10)', the specific formula is as follows:

[0081]

[0082] S4.2: Calculate the weight w th :

[0083] w th =(F T F)A t F (14)

[0084] S5: Output the predicted value of power load data, which is the output value Y of the RBF neural network:

[0085]

[0086] Finally, it should be noted that the described embodiments are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

Claims

1. A method for predicting power load based on cloud computing, characterized in that: The specific steps include: S1: Select power load data; S2: Use the orthogonal least squares method OLS to find the center vector of the radial basis function, and set a threshold to detect the center vector of the radial basis function. If the threshold is not met, adjust the number of neurons in the hidden layer and recalculate. The specific calculation steps are as follows: S2.1: Take the actual power load data as the sample number T, and set A t is the input vector, B t is the expected vector, both are vectors of the same dimension, e t is the absolute error between the input vector and the input vector, the value range of t is 1, 2, …T, and the formula is: S2.2: Set the number of neurons in the hidden layer of the RBF neural network to H, which is usually 1 in the initial case, and w h is the weight matrix, O h (t) is the regression operator, and the value range of h is 1, 2, ...H, so B t The calculation formula is: S2.3: Express equation (2) as a matrix: B=OWH+e (3) S2.4: Use the orthogonal triangular QR decomposition method to express O as a matrix Q and an upper triangular matrix R. The specific formula is as follows: B=QRWH+e (4) S2.5: Express RW as q h , according to the orthogonal least squares method, the least squares solution is obtained, the formula is as follows: S2.6: Use the Schmidt orthogonalization method to reduce the error and determine the value s h , the specific formula is as follows: S2.7: Set the threshold z and select the input vector that satisfies formula (7) as the center vector c k , the value of k is an integer less than T: If there is no center vector c that meets the conditions k , then adjust B t and H and then recalculate; S3: Use the k-means algorithm in a cloud computing environment and combine it with the MapReduce architecture to adjust the center vector of the radial basis function and calculate the expansion parameters. Use the equation to calculate the expansion parameters. If the equation is satisfied, proceed to the next step. If not, reselect the input data. S4: Calculate weights and run RBF neural network; S5: Output power load data prediction value.

2. The method for predicting power load based on cloud computing according to claim 1, characterized in that: The specific calculation steps of S3 are as follows: S3.1: Combine the power load data sample set and the center vector c confirmed by OLS k Input, denoted as c k (n), n represents the number of iterations; S3.2: Calculate the Euclidean distance D, A t (n) indicates that in t The randomly selected data represents the input data of the nth iteration. If the D value between the input vector and a certain center vector is the smallest, then this input vector is included in this cluster point set J k The specific formula is as follows: D=arg||A t (n)-c k (n)|| (8) S3.3: Get the new cluster center c k (n)', Represents the set J k The number of neutron subsets is calculated as follows: S3.4: If equation (10) is satisfied, then c k (n)' is the new center vector, and the process goes to step S3.

5. If the condition is not satisfied, the process goes back to step S3.2 and randomly extracts data again: c k (n)'=c k (n+1)' (10) S3.5: Calculate the distance D between the center vectors c , X represents the overlap coefficient n / 2, and the calculation formula of the coefficient S is as follows: D c =||c k (n)'-c k || (11) S=D c X (12) 3. The power load forecasting method based on cloud computing according to claim 1 is characterized in that: The cloud computing environment in S3 is a Hadoop cloud computing environment.

4. The power load forecasting method based on cloud computing according to claim 1 is characterized in that: The weight calculation in S4 comprises the following steps: S4.1: Calculate radial basis function, C h ' indicates that this neuron is A t The corresponding center vector, the specific formula is as follows: S4.2: Calculate the weight w th : w th =(F T F)A t F (14) 5. The method for predicting power load based on cloud computing according to claim 1, characterized in that: The predicted value of the power load data output in S5 is the output value Y of the RBF neural network:

Images