Electric vehicle charging station load prediction method
Through QR decomposition and iterative least squares algorithm, the hidden layer node weight matrix of the limit learning machine model is optimized, which solves the problems of prediction accuracy and low efficiency in the traditional model, and achieves more efficient load prediction of electric vehicle charging stations.
Patent Information
- Application Number
- CN202510596416.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-09
- Publication Date
- 2025-08-12
AI Technical Summary
In the prior art, the electric vehicle charging station load prediction model based on the traditional incremental extreme learning machine has the problem of reducing the accuracy and efficiency of charging station load prediction. The randomly generated hidden layer node parameters affect the stability of the model, resulting in large network training errors.
The output weight matrix of the nodes of the limit learning machine hidden layer is obtained by using QR decomposition and the limit learning machine model is decomposed into orthogonal matrix Q and upper triangle matrix R through QR decomposition. The weight matrix is optimized by iterative least squares method to improve the stability and efficiency of model training.
While reducing the computational burden, the generalization performance of the algorithm is improved, more accurate load prediction of electric vehicle charging stations is achieved, and more efficient prediction model training methods are provided.
Smart Images
Figure BDA0005394713910000021 
Figure BDA0005394713910000037 
Figure BDA0005394713910000042
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of power systems, and in particular to a load forecasting method for an electric vehicle charging station. Background Art
[0002] By accurately predicting the load at electric vehicle charging stations, power grid companies can predict the future growth and distribution of power load in different regions. For example, in newly built residential and commercial areas where electric vehicles are prevalent, load forecasts can be used to rationally plan substation capacity, layout and specifications of transmission lines, and other factors. This can help avoid power shortages caused by insufficient grid capacity or resource waste caused by excessive grid construction, making grid resource allocation more scientific and efficient.
[0003] Designing prediction models and learning algorithms is a key issue in research on electric vehicle charging station load forecasting. Existing prediction models based on traditional incremental extreme learning machines (ELMs) contain many redundant nodes that reduce the accuracy and efficiency of charging station load forecasting. Randomly generated hidden layer node parameters affect the stability of the ELMs, leading to large network training errors. Therefore, designing efficient prediction models is crucial for charging station load forecasting. Summary of the Invention
[0004] In view of this, the present invention provides a method for predicting the load of an electric vehicle charging station, which uses an extreme learning machine model to predict the load of an electric vehicle charging station.
[0005] The present invention provides a method for predicting the load of an electric vehicle charging station, comprising the following steps:
[0006] Step 1: Collect the load values of electric vehicle charging stations before the prediction time point to form a historical data set, use the load value of the electric vehicle charging station at a certain time point as output, and use the load value of the electric vehicle charging station before the time point as input to establish a training sample set;
[0007] Step 2: Establish an extreme learning machine model and use the training sample set established in step 1 to train the extreme learning machine model. The extreme learning machine model is shown in the following formula:
[0008] Y 1×N =β 1×k ·J k×N
[0009]
[0010] Among them, Y 1×N ∈R 1×N represents the output matrix of the extreme learning machine model, N is the number of samples, k is the number of hidden layer nodes, J k×N is the hidden layer node output matrix, x4×N ∈R 4×N is the input data of the extreme learning machine model, α k×4 is the input weight matrix between the hidden layer nodes, λ k×N is the bias matrix of the hidden layer nodes, is an optional nonlinear activation function, α k×4 and λ k×N All are randomly generated;
[0011] Step 3: In actual use, the load values of the electric vehicle charging station at Q consecutive time points before the time point to be predicted are obtained, and the values are input into the trained extreme learning machine model to obtain the predicted value of the electric vehicle charging station load at the time point to be predicted.
[0012] Furthermore, the historical data set is constructed by collecting the load values of the electric vehicle charging station M days before the predicted time point, recording the load values of the electric vehicle charging station from 0:00 to 24:00 every day, and then calculating the average of the recorded results as the load value of the electric vehicle charging station on that day. A total of data at M time points are recorded to form a historical data set.
[0013] Furthermore, the value of M is an integer greater than 4.
[0014] Furthermore, the method of taking the load value of the electric vehicle charging station before the time point as input is: taking the load values of the electric vehicle charging station at Q consecutive time points before the time point as input.
[0015] Furthermore, the value of Q is an integer not less than 2.
[0016] Furthermore, the method of training the extreme learning machine model using the training sample set established in step 1 in step 2 is: using QR decomposition and iterative least squares to solve the output weight matrix of the hidden layer nodes to complete the training of the extreme learning machine model.
[0017] Furthermore, the QR decomposition method is:
[0018] Definition U N×k Represents the hidden layer node output matrix J k×N The inverse matrix of J k×N *U N×k =I k×k , where I k ×k is the identity matrix, r is J k×N rank;
[0019] Step 1.1, let i add 1, when i≤r, let Calculate ||q1||2, ||q1||2 is a vector 2-norm, if ‖‖q1‖‖2 is not less than ε, stop the loop and execute step 1.2 to obtain the first column q1 of the matrix Q, otherwise execute step 1.1;
[0020] Step 1.2, use ‖‖q1‖‖2 to calculate the vector q i Unitize;
[0021] Step 1.3, let j = i and calculate the first row of matrix R:
[0022] Step 1.4: Starting from the second column of matrix Q, let k add 1 to matrix Q. If k ≤ r, proceed to step 1.5; otherwise, proceed to step 1.6.
[0023] Step 1.5, use the following two formulas to transform the vector Orthogonalize with the first i-1 columns of matrix Q to obtain the candidate for the i-th column of matrix Q:
[0024]
[0025] q s =q s -r k Q(:,k);
[0026] Step 1.6: If ‖‖q s If ‖‖2 is not less than ε, then execute step 1.2 to obtain the i-th column of matrix Q; otherwise, let j = k and execute step 1.3 to obtain the i-th row of matrix R;
[0027] Step 1.7, orthogonalize the matrix R, and make the matrix C an n-order diagonal matrix, and construct the matrix K = [A T A,A T ];
[0028] Step 1.8. Perform the following calculations on the matrix C and matrix K to perform updates:
[0029]
[0030] K(j,:)=K(j,:)+const×K(i,:)
[0031] C(j,:)=C(j,:)+const×C(i,:);
[0032] Step 1.9: Transpose the matrix C to get the final C matrix, and transpose the matrix consisting of all columns starting from the n+1th column in the matrix K to get the final B matrix, so that R = BC -1 .
[0033] Furthermore, the iterative least squares method for solving the output weight matrix of the hidden layer nodes is:
[0034] Step 2.1: Convert the inverse equation J*U=I into QBC -1 U=I,let I=Q T I, U = C -1 U, J = B, the inverse matrix is calculated using the following formula The value of the element in row i and column j is:
[0035]
[0036] in, is a vector and vector Perform inner product operation, is a vector The square of the 2-norm of ;
[0037] Step 2.2: Let i increment by 1. When i≤N, execute step 2.1; when i>N, execute step 2.3.
[0038] Step 2.3: Let j increment by 1. When j ≤ k, execute step 2.1; when j > k, execute step 2.4.
[0039] Step 2.4: Use the following two formulas to calculate and output the output weight matrix of the hidden layer nodes:
[0040] U=C*U
[0041] β 1×k =Y 1×N *U N×k
[0042] End this process.
[0043] Beneficial effects:
[0044] The present invention utilizes QR decomposition and iterative least squares to obtain the output weight matrix of the hidden layer nodes of the extreme learning machine to implement the training of the extreme learning machine model. It can improve the generalization performance of the algorithm while reducing the excessive computational burden and improving computational efficiency. To a certain extent, it can meet the needs of charging station load forecasting, and at the same time provide new ideas and new approaches for more accurate charging station load forecasting. DETAILED DESCRIPTION
[0045] The present invention is described in detail with reference to the following examples.
[0046] The present invention provides a method for predicting the load of an electric vehicle charging station, which uses an extreme learning machine based on QR decomposition and iterative least squares to implement the load prediction of the charging station, and specifically includes the following steps:
[0047] Step 1: Collect the load values of the electric vehicle charging station M days before the predicted time point (the unit of "time point" in this embodiment is "day"), record the load values of the electric vehicle charging station from 0:00 to 24:00 every day, and then calculate the average value of the recorded results as the load value of the electric vehicle charging station on that day. Record the data of M time points in total, M>4, and form a historical data set K={k1,k2,···,k m ,···,k M}, where k m is the load value of the electric vehicle charging station on the mth day.
[0048] From the historical data set K={k1,k2,···,k m ,···,k M}, where the output is the load value of the electric vehicle charging station at the jth time point, expressed as y j =k j The corresponding input is the load value of the electric vehicle charging station at Q consecutive time points before the jth time, where Q is a positive integer greater than or equal to 2. Here, a sliding window with a width of Q = 4 time points can be set to move in steps of 1 time point to select the time point as the input. Then the input sample at the jth time is expressed as x j =[k p-1 ,k p-2 ,k p-3 ,k p-4 ], where (p-4)>j. Then the training sample set established is j represents the jth sample, and N is the total number of samples in the training sample set.
[0049] Step 2: Establish an extreme learning machine model and use the training sample set established in step 1 to train the extreme learning machine model. The extreme learning machine model is shown in the following formula:
[0050] Y 1×N =β 1×k ·J k×N
[0051]
[0052] Among them, Y 1×N ∈R 1×N represents the output matrix of the extreme learning machine model, N is the number of samples, k is the number of hidden layer nodes, J k×N is the hidden layer node output matrix, x 4×N ∈R 4×N is the input data of the extreme learning machine model, α k×4 is the input weight matrix between the hidden layer nodes, λk×N is the bias matrix of the hidden layer nodes, is an optional nonlinear activation function; α k×4 and λ k×N They are all randomly generated and remain unchanged after generation.
[0053] The extreme learning machine model is trained using a training sample set of electric vehicle charging station load values. The present invention utilizes QR decomposition and iterative least squares to solve the output weight matrix of hidden layer nodes, thereby completing the training of the extreme learning machine model.
[0054] Step 3: In actual use, the load values of the electric vehicle charging station at Q consecutive time points before the time point to be predicted are obtained, and the load values of the electric vehicle charging station at the time point to be predicted are input into the trained extreme learning machine model to obtain the output load value of the electric vehicle charging station at the time point to be predicted.
[0055] Specifically, in step 2, the output weight matrix of the hidden layer nodes is solved by using QR decomposition and iterative least squares to complete the training of the extreme learning machine model, which specifically includes the following steps:
[0056] Definition U N×k Represents the hidden layer node output matrix J k×N The inverse matrix of J k×N *U N×k =I k×k , where I k ×k is the identity matrix, r is J k×N rank.
[0057] For example, if k = 4 and N = 4, the identity matrix and inverse matrix can be expressed as:
[0058]
[0059] Step 1.1, let i add 1, when i≤r, let Calculate ‖‖q1‖‖2, where ‖‖q1‖‖2 is a vector If the 2-norm is not less than ε, stop the loop and execute step 1.2 to obtain the first column q1 of the matrix Q. Otherwise, execute step 1.1.
[0060] Step 1.2: Use the following formula to calculate the vector q i Unitize:
[0061]
[0062] Step 1.3: Let j = i and calculate the first row of the matrix R using the following formula:
[0063]
[0064] Step 1.4: Starting from the second column, add 1 to the matrix k. If k ≤ r, proceed to step 1.5; otherwise, proceed to step 1.6.
[0065] Step 1.5, use the following two formulas to transform the vector Orthogonalize with the first i-1 columns of matrix Q to obtain the candidate for the i-th column of matrix Q:
[0066]
[0067] q s =q s -r k Q(:,k)
[0068] Step 1.6. Calculate ‖‖q s ‖‖2. If it is not less than ε, then we consider that we have found a new basis vector and execute step 1.2 to obtain the i-th column of matrix Q. If it is less than ε, then let j = k and execute step 1.3 to obtain the i-th row of matrix R. The above algorithm can obtain an orthogonal matrix Q and an upper triangular matrix R, satisfying J ≈ QR.
[0069] Step 1.7, orthogonalize the matrix R, and make the matrix C an n-order diagonal matrix, and construct the matrix K = [A T A,A T ], execute the loop, outer layer i = 1 to n-1, inner layer j = i+1 to n, for each row i selected by the outer loop, the inner loop processes all rows j, if K(i,i)>0, execute step 1.8.
[0070] Step 1.8: Use the following three formulas to calculate and update the matrix C and matrix K:
[0071]
[0072] K(j,:)=K(j,:)+const×K(i,:)
[0073] C(j,:)=C(j,:)+const×C(i,:)
[0074] Step 1.9: Transpose the matrix C to get the final C matrix, and transpose the matrix consisting of all columns starting from the n+1th column in the matrix K to get the final B matrix, so that R = BC -1 .
[0075] According to the above steps, the original inverse equation J*U=I can be written as QBC -1 U=I,let I=Q T I, U = C -1 U, J = B, use the following steps to calculate U:
[0076] Step 2.1, calculate the inverse matrix using the following formula The value of the element in row i and column j is:
[0077]
[0078] in, is a vector and vector Perform inner product operation, is a vector The square of the 2-norm of .
[0079] Step 2.2: Let i be incremented by 1. When i≤N, execute step 2.1; when i>N, execute step 2.3.
[0080] Step 2.3: Let j add 1. When j≤k, execute step 2.1; when j>k, execute step 2.4. Step 2.4: Use the following two formulas to calculate and output the output weight matrix of the hidden layer nodes.
[0081] U=C*U
[0082] β 1×k =Y 1×N *U N×k
[0083] End this process.
[0084] In the above process, the output weight matrix of the hidden layer nodes is solved based on QR decomposition and iterative least squares algorithm, which makes the model easier to converge, thereby improving the learning speed of the extreme learning machine model and shortening the training time of the model.
[0085] In summary, the above are only preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.
Claims
1. A method for predicting load at an electric vehicle charging station, characterized in that: The following steps are involved: Step 1: Collect the load values of electric vehicle charging stations before the prediction time point to form a historical data set, use the load value of the electric vehicle charging station at a certain time point as output, and use the load value of the electric vehicle charging station before the time point as input to establish a training sample set; Step 2: Establish an extreme learning machine model and use the training sample set established in step 1 to train the extreme learning machine model. The extreme learning machine model is shown in the following formula: Y 1×N =β 1×k ·J k×N Among them, Y 1×N ∈R 1×N represents the output matrix of the extreme learning machine model, N is the number of samples, k is the number of hidden layer nodes, J k×N is the hidden layer node output matrix, x 4×N ∈R 4×N is the input data of the extreme learning machine model, α k×4 is the input weight matrix between the hidden layer nodes, λ k×N is the bias matrix of the hidden layer nodes, is an optional nonlinear activation function, α k×4 and λ k×N All are randomly generated; Step 3: In actual use, the load values of the electric vehicle charging station at Q consecutive time points before the time point to be predicted are obtained, and the values are input into the trained extreme learning machine model to obtain the predicted value of the electric vehicle charging station load at the time point to be predicted.
2. The method for predicting load of an electric vehicle charging station according to claim 1, characterized in that: The historical data set is constructed by collecting the load values of the electric vehicle charging station M days before the predicted time point, recording the load values of the electric vehicle charging station from 0:00 to 24:00 every day, and then averaging the recorded results as the load value of the electric vehicle charging station on that day. A total of data at M time points are recorded to form a historical data set.
3. The method for predicting load of an electric vehicle charging station according to claim 2, characterized in that: The value of M is an integer greater than 4.
4. The method for predicting load of an electric vehicle charging station according to claim 1, characterized in that: The method of using the load value of the electric vehicle charging station before the time point as input is: using the load values of the electric vehicle charging station at Q consecutive time points before the time point as input.
5. The method for predicting load of an electric vehicle charging station according to claim 4, characterized in that: The value of Q is an integer not less than 2.
6. The method for predicting load of an electric vehicle charging station according to claim 1, characterized in that: The method of training the extreme learning machine model using the training sample set established in step 1 in step 2 is: using QR decomposition and iterative least squares to solve the output weight matrix of the hidden layer nodes to complete the training of the extreme learning machine model.
7. The method for predicting load of an electric vehicle charging station according to claim 6, characterized in that: The QR decomposition method is: Definition U N×k Represents the hidden layer node output matrix J k×N The inverse matrix of J k×N *U N×k =I k×k , where I k×k is the identity matrix, r is J k×N rank; Step 1.1, let i add 1, when i≤r, let Calculate ||q1||2, ||q1||2 is a vector 2-norm of . If ||q1||2 is not less than ε, stop the loop and execute step 1.2 to obtain the first column q1 of the matrix Q. Otherwise, execute step 1.
1. Step 1.2, use ||q1||2 to calculate the vector q i Unitize; Step 1.3, let j = i and calculate the first row of matrix R: Step 1.4: Starting from the second column of matrix Q, let k add 1 to matrix Q. If k ≤ r, proceed to step 1.5; otherwise, proceed to step 1.
6. Step 1.5, use the following two formulas to transform the vector Orthogonalize with the first i-1 columns of matrix Q to obtain the candidate for the i-th column of matrix Q: q s =q s -r k Q(:,k); Step 1.6, if ||q s If ||2 is not less than ε, then execute step 1.2 to obtain the i-th column of matrix Q; otherwise, let j = k and execute step 1.3 to obtain the i-th row of matrix R; Step 1.7, orthogonalize the matrix R, and make the matrix C an n-order diagonal matrix, and construct the matrix K = [A T A,A T ]; Step 1.
8. Perform the following calculations on the matrix C and matrix K to perform updates: K(j,:)=K(j,:)+const×K(i,:) C(j,:)=C(j,:)+const×C(i,:); Step 1.9: Transpose the matrix C to get the final C matrix, and transpose the matrix consisting of all columns starting from the n+1th column in the matrix K to get the final B matrix, so that R = BC -1 .
8. The method for predicting load of an electric vehicle charging station according to claim 6, characterized in that: The method of iterative least squares solution to the output weight matrix of the hidden layer nodes is: Step 2.1: Convert the inverse equation J*U=I into QBC -1 U=I,let I=Q T I, U = C -1 U, J = B, the inverse matrix is calculated using the following formula The value of the element in row i and column j is: in, is a vector and vector Do the inner product operation, is a vector The square of the 2-norm of ; Step 2.2: Let i increment by 1. When i≤N, execute step 2.1; when i>N, execute step 2.
3. Step 2.3: Let j increment by 1. When j ≤ k, execute step 2.1; when j > k, execute step 2.
4. Step 2.4: Use the following two formulas to calculate and output the output weight matrix of the hidden layer nodes: U=C*U β 1×k =Y 1×N *U N×k End this process.