A method for load prediction within busbar area under high proportion of distributed photovoltaic access

Abstract

The present invention discloses a method for predicting load within a bus area under high-proportion distributed photovoltaic access, which belongs to the technical field of distributed photovoltaic power generation load prediction, and includes the following steps: collecting the load change curve over time and the distributed photovoltaic power generation curve within the distribution network bus area to construct a data set; based on the data set, by calculating the mutual information value of the distributed photovoltaic output and the bus area load, the correlation between the data is judged; after performing empirical mode decomposition on the data with strong correlation, the data frequency is identified to obtain the prediction results of the high-frequency component, the low-frequency component and the residual term respectively; the prediction results obtained respectively are added by vector to obtain the final prediction value of the load prediction within the bus area. The present invention effectively avoids information loss, retains the original characteristics of the data, and improves the accuracy of the prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of distributed photovoltaic power generation load prediction, and in particular to a method for predicting load within a busbar area under high-proportion distributed photovoltaic access. Background Art

[0002] The large-scale access of distributed photovoltaics in distribution networks has become a future trend. However, the increase in the penetration rate of distributed photovoltaics will lead to changes in the load pattern within its bus area, which will in turn affect the normal power supply in the area. Therefore, accurate prediction of the load within the bus area under the scenario of large-scale photovoltaic access is a prerequisite for the economic and reliable operation of the distribution network. Compared with the load at the power system level, it is difficult to adjust the smoothness of the load curve through demand-side management within the bus area, resulting in obvious fluctuations in the overall load curve after aggregation, and it is difficult to predict the load in the bus area. Existing studies mainly use time series analysis and machine learning to predict the load in the bus area, but these methods still have shortcomings in dealing with the volatility of photovoltaic power generation, cannot effectively capture instantaneous changes, and have high requirements on data quality. Summary of the invention

[0003] In order to solve the above problems, the purpose of the present invention is to provide a load forecasting method within the bus area under a high proportion of distributed photovoltaic access, make judgments based on mutual information theory, use empirical mode decomposition and Lempel-Ziv complexity algorithm to decompose and identify target components, and use aggregated gradient boosting tree algorithm and extreme learning machine algorithm to train and predict data.

[0004] In order to achieve the above technical objectives, the present application provides a method for predicting load within a busbar area under high-proportion distributed photovoltaic access, comprising the following steps:

[0005] Collect the load variation curve over time and distributed photovoltaic power generation curve within the distribution network bus area to build a data set;

[0006] Based on the data set, the correlation between the data is judged by calculating the mutual information value between the distributed photovoltaic output and the busbar area load;

[0007] After empirical mode decomposition of the strongly correlated data, data frequency identification is performed to obtain the prediction results of high-frequency components, low-frequency components and residual terms respectively;

[0008] The prediction results obtained separately are added by vector to obtain the final prediction value of the load forecast within the bus area.

[0009] Preferably, in the process of obtaining the mutual information value, the information entropy value of the distributed photovoltaic output sequence and the conditional information entropy of the distributed photovoltaic output sequence to the bus area load sequence are obtained; and the difference between the information entropy value and the conditional information entropy is used as the mutual information value of the distributed photovoltaic output and the bus area load, and the strong correlation between the distributed photovoltaic output sequence and the bus area load sequence is judged through the mutual information value.

[0010] Preferably, in the process of performing empirical mode decomposition, the distributed photovoltaic output sequence and the bus area load sequence with strong correlation are decomposed into intrinsic mode function components and residual terms respectively.

[0011] Preferably, in the process of obtaining the intrinsic mode function components and residual terms, according to the original function of the distributed photovoltaic output sequence or the busbar area load sequence, upper and lower envelopes are formed for all maximum and minimum points of the original function based on the cubic difference function, and the average values ​​of the upper and lower envelopes are calculated. According to the difference between the original function and the average value, the intrinsic mode function is constructed, and the intrinsic mode function is separated into the eigenmode function components and residual terms.

[0012] Preferably, in the process of performing empirical mode decomposition, extreme points are cyclically screened until the standard deviation of two adjacent screening results reaches [0.2, 0.3], so as to realize the process of screening out the eigenmode function through the original function.

[0013] Preferably, in the process of data frequency identification, a Lempel-Ziv complexity algorithm is used to identify high and low frequency components.

[0014] Preferably, in the process of predicting the high-frequency component, based on the XGBOOST algorithm, an XGBOOST model is constructed through the high-frequency component to obtain the prediction result of the high-frequency component.

[0015] Preferably, in the process of constructing the XGBOOST model, the error term and the regularization term are used as the loss function of the model, the model is trained using an addition strategy, and the objective function is gradually optimized using a forward distribution algorithm to construct the XGBOOST model.

[0016] Preferably, in the process of obtaining the prediction results of the low-frequency components and the residual terms, the prediction results are obtained based on the low-frequency components and the residual terms respectively based on the extreme learning machine model, wherein the extreme learning machine is a single hidden layer feedforward neural network model consisting of an input layer, a hidden layer and an output layer, and the neurons are fully connected.

[0017] Preferably, the present invention implements a method for predicting load within a busbar area by using a system for predicting load within a busbar area under high-proportion distributed photovoltaic access, the system comprising:

[0018] The data acquisition module is used to collect the load variation curve over time and the distributed photovoltaic power generation curve within the distribution network bus area to build a data set;

[0019] The data processing module is used to determine the correlation between data by calculating the mutual information value between the distributed photovoltaic output and the busbar area load based on the data set;

[0020] The component prediction module is used to perform empirical mode decomposition on the data with strong correlation, identify the data frequency, and obtain the prediction results of high-frequency components, low-frequency components and residual terms respectively;

[0021] The load forecasting module is used to obtain the final forecast value of the load forecast within the bus area by adding the forecast results obtained separately through vectors.

[0022] The present invention discloses the following technical effects:

[0023] The present invention effectively avoids information loss, retains the original characteristics of the data, and improves the accuracy of prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0025] Figure 1 It is a schematic diagram of the method flow of the present invention. DETAILED DESCRIPTION

[0026] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0027] The present invention provides a method for predicting load within a busbar area under high-proportion distributed photovoltaic access, comprising the following steps:

[0028] Collect the load variation curve over time and distributed photovoltaic power generation curve within the distribution network bus area to build a data set;

[0029] Based on the data set, the correlation between the data is judged by calculating the mutual information value between the distributed photovoltaic output and the busbar area load;

[0030] After empirical mode decomposition of the strongly correlated data, data frequency identification is performed to obtain the prediction results of high-frequency components, low-frequency components and residual terms respectively;

[0031] The prediction results obtained separately are added by vector to obtain the final prediction value of the load forecast within the bus area.

[0032] Further preferably, the method for load prediction within the bus area under a high proportion of distributed photovoltaic access provided by the present invention obtains the information entropy value of the distributed photovoltaic output sequence and the conditional information entropy of the distributed photovoltaic output sequence to the bus area load sequence in the process of obtaining the mutual information value; and uses the difference between the information entropy value and the conditional information entropy as the mutual information value of the distributed photovoltaic output and the bus area load, and judges the strong correlation between the distributed photovoltaic output sequence and the bus area load sequence through the mutual information value.

[0033] Further preferably, the method for load prediction within the bus area under a high proportion of distributed photovoltaic access provided by the present invention, in the process of performing empirical mode decomposition, decomposes the distributed photovoltaic output sequence and the bus area load sequence with strong correlation into intrinsic mode function components and residual terms respectively.

[0034] Further preferably, the method for load prediction within the bus area under a high proportion of distributed photovoltaic access provided by the present invention, in the process of obtaining the intrinsic mode function components and residual terms, forms upper and lower envelopes for all maximum and minimum points of the original function based on the cubic difference function according to the original function of the distributed photovoltaic output sequence or the bus area load sequence, and calculates the average values ​​of the upper and lower envelopes, constructs the intrinsic mode function according to the difference between the original function and the average value, and separates the intrinsic mode function into the eigenmode function components and residual terms.

[0035] Further preferably, the load prediction method within the bus area under a high proportion of distributed photovoltaic access provided by the present invention, in the process of performing empirical mode decomposition, cyclically screens extreme points until the standard deviation of two adjacent screening results is [0.2, 0.3], so as to realize the process of screening out the eigenmode function through the original function.

[0036] Further preferably, in the method for load forecasting within a bus area under a high proportion of distributed photovoltaic access provided by the present invention, in the process of data frequency identification, a Lempel-Ziv complexity algorithm is used to identify high and low frequency components.

[0037] Further preferably, in the process of predicting the high-frequency component, based on the XGBOOST algorithm, an XGBOOST model is constructed through the high-frequency component to obtain the prediction result of the high-frequency component.

[0038] Further preferably, in a method for load prediction within a bus area under a high proportion of distributed photovoltaic access provided by the present invention, in the process of constructing an XGBOOST model, the error term and the regularization term are used as the loss function of the model, the model is trained using an addition strategy, and the objective function is gradually optimized using a forward distribution algorithm, thereby constructing the XGBOOST model.

[0039] Further preferably, the load prediction method within the bus area under a high proportion of distributed photovoltaic access provided by the present invention, in the process of obtaining the prediction results of the low-frequency components and the residual terms, obtains the prediction results based on the low-frequency components and the residual terms respectively based on the extreme learning machine model, wherein the extreme learning machine is a single hidden layer feedforward neural network model consisting of an input layer, a hidden layer and an output layer, and the neurons are fully connected.

[0040] The present invention also discloses a load prediction system within a bus area under high-proportion distributed photovoltaic access, which is used to implement a load prediction method within a bus area. The system includes:

[0041] The data acquisition module is used to collect the load variation curve over time and the distributed photovoltaic power generation curve within the distribution network bus area to build a data set;

[0042] The data processing module is used to determine the correlation between data by calculating the mutual information value between the distributed photovoltaic output and the busbar area load based on the data set;

[0043] The component prediction module is used to perform empirical mode decomposition on the data with strong correlation, identify the data frequency, and obtain the prediction results of high-frequency components, low-frequency components and residual terms respectively;

[0044] The load forecasting module is used to obtain the final forecast value of the load forecast within the bus area by adding the forecast results obtained separately through vectors.

[0045] Example: Figure 1 As shown, the present invention provides a method for predicting load within a busbar area under high-proportion distributed photovoltaic access, comprising the following steps:

[0046] 1. Data collection and division. Collect the load change curve over time and distributed photovoltaic power generation curve within the jurisdiction of the 10kV distribution network busbar. Divide the data into training data set and test data set in a ratio of 8:2.

[0047] 2. Calculate the mutual information value I(X;Y) between the distributed photovoltaic output and the bus area load.

[0048] First, the information entropy value H(X) of the distributed photovoltaic output sequence X and the conditional information entropy H(X / Y) of the sequence X to the bus area load sequence Y are calculated.

[0049]

[0050] p(x) and p(y) are the marginal probability distributions of x and y occurring separately respectively; p(x, y) is the joint probability distribution of x and y occurring simultaneously.

[0051] According to the mutual information value definition I(X;Y)=H(X)-H(X / Y), the mutual information value of distributed photovoltaic output and bus area load is further calculated.

[0052]

[0053] If I(X;Y) is greater than 0.8 and there is a strong correlation, proceed to the next step.

[0054] 3. Perform empirical mode decomposition on the photovoltaic output time series and load time series in the training set data. Assume that the given data is r(t), which can represent the function of the photovoltaic output time series or load time series with respect to time t. After decomposition, r(t) consists of m intrinsic mode function components u(t) and residual term h(t); the processing steps are as follows:

[0055] 3.1 Filter the original function r(t) to finally form a new function s(t): First, calculate the intrinsic mode function s1(t) of r(t). Based on the cubic difference function, form upper and lower envelopes for all maximum and minimum points of r(t), and calculate the average value a 1(t) of the upper and lower envelopes. The formula is as follows:

[0056] s 1 (t) = r(t) - a 1 (t);

[0057] To prevent missing the decomposition process 1 The inflection point in (t) is converted into a new extreme point, and the above process is repeated to filter the extreme points in a cycle. 1 (t) is used as the new r(t), and the previous operation is repeated n times until the standard deviation of two adjacent screening results is [0.2, 0.3]. The formula is as follows. n (t) and an (t) are the nth eigenmode function and envelope average value respectively. n (t) is the new function s(t):

[0058] s n (t) = r(t) - a n (t);

[0059] 3.2 Separate the intrinsic mode function components u(t) and the residual h(t). Let u 1 (t) is equal to s(t), and u 1 (t) is separated from the original signal r(t) to obtain the residual h 1 (t), the calculation formula is as follows:

[0060] h 1 (t) = r(t) - u 1 (t);

[0061] h 1 (t) is used as a new function, and the above steps are repeated to obtain the subsequent eigenmode function components until h n When (t) is monotonic or less than the given error standard [0.2, 0.3], stop decomposition. The formula is as follows:

[0062]

[0063] 4. Data frequency identification. In order to coarse-grain the data to create sequences of different scales, the Lempel-Ziv complexity algorithm is used to identify high- and low-frequency components, thereby selecting the optimal model for predicting each component.

[0064] First, calculate the average value of the output sequence in step 3, and set the data exceeding the average value in the sequence to 1, otherwise to 0. This will give a binary sequence of 1 and 0 {x 1 ,x 2 ,…,x i}. Set the temporary character variable P initialized to empty i and Q i For auxiliary calculation, assume that the initial character variable is empty and the initial complexity value C of the sequence is 0.

[0065] Secondly, through the string P i and Q i The relationship loop modifies the temporary complexity value C'.

[0066] Establish the connection between the character variable and the sequence by connecting the temporary character variable and the elements in the sequence i = {P i-1 x i},Q={Q i-1 x i}.

[0067] If Q i YesP i-1 If the substring is a string, the value of C' remains unchanged; otherwise, the value of C' increases by 1. After N cycles, all values ​​can be traversed and judged, and the complexity CN of the sequence can be calculated. Finally, C' is normalized to obtain the normalized complexity C, which is calculated as follows. m is the coarse-grained metric value. This method is a binary coarse-grained method, and m is 2.

[0068]

[0069] Finally, the data frequency is divided, k is the frequency division boundary parameter, the first k intrinsic mode function components are high-frequency components, and the kth to Kth are low-frequency components. The formula used is as follows.

[0070]

[0071] Where λ is the frequency judgment parameter, which is generally set to 0.8.

[0072] 5. Establish an XGBOOST model for high-frequency components to predict the high-frequency component results. In principle, the XGBOOST algorithm is an integration of many classification and regression trees. The regression tree determines the output value based on the feature vector and combines the prediction results of multiple decision trees to obtain better prediction performance. The most basic prediction model calculation of the XGBOOST algorithm is as follows.

[0073]

[0074] Where i = 1, 2, ..., n, n is the number of samples, M is the number of trees, f k is a function in the set F of trees.

[0075] The objective function is corrected by calculating the loss function fob. The loss function is related to f k The relationship is shown as follows.

[0076]

[0077] The loss function consists of an error term and a regularization term, where L is the error term and l represents the error term y i , relationship.

[0078] Using the addition strategy for model training, in the tth step, the regularization term Ω(f t ):

[0079]

[0080] T is the number of leaf nodes, ω is the score value of the leaf node, γ and λ are the built-in adjustment parameters of the tree model, which are adaptively adjusted according to the data situation to obtain the most accurate prediction results.

[0081] The forward distribution algorithm is used to gradually optimize the objective function. Based on the above principle, the decision tree to be learned is ft(xi), x i The prediction results The calculation is as follows.

[0082]

[0083] The final calculation method of the objective function is as follows.

[0084]

[0085] 6. Establish an extreme learning machine model for the remaining low-frequency components and residual terms to predict the low-frequency component results. The extreme learning machine is a single hidden layer feedforward neural network model consisting of an input layer, a hidden layer, and an output layer, and the neurons are fully connected. For the input, let the output function f L (x) is calculated as follows:

[0086]

[0087] h(x) is the output of the hidden layer, β=[β 1 ,…,β L ] T is the weight coefficient between the i-th hidden layer and the output layer, L is the number of neurons h in the hidden layer i (x) quantity; g i (x) and G(a i , b i , x) is the output function of hidden node i; a i , b i is the hidden layer parameter; β i is the output weight vector.

[0088] 7. Through vector addition, the prediction results of all high-frequency, low-frequency and residual components are integrated to obtain the final prediction value.

[0089] Table 1

[0090]

[0091] To verify the effectiveness of the proposed method, it is further compared with the prediction results using neural networks and time series algorithms. As shown in Table 1, it is found that the error index of this method is the smallest, that is, the accuracy is higher. The error indicators used are Mean Absolute Percentage Error (MAPE) and Mean Square Error (MSE).

[0092] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0093] In the description of the present invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0094] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for predicting load within a busbar area under high-proportion distributed photovoltaic access, characterized in that: The following steps are involved: Collect the load variation curve over time and distributed photovoltaic power generation curve within the distribution network bus area to build a data set; Based on the data set, the correlation between the data is judged by calculating the mutual information value between the distributed photovoltaic output and the busbar area load; After empirical mode decomposition of the strongly correlated data, data frequency identification is performed to obtain the prediction results of high-frequency components, low-frequency components and residual terms respectively; The prediction results obtained separately are added by vector to obtain the final prediction value of the load forecast within the bus area.

2. According to the method for predicting load within the bus area under high-proportion distributed photovoltaic access according to claim 1, it is characterized by: In the process of obtaining the mutual information value, the information entropy value of the distributed photovoltaic output sequence and the conditional information entropy of the distributed photovoltaic output sequence to the bus area load sequence are obtained; and the difference between the information entropy value and the conditional information entropy is used as the mutual information value of the distributed photovoltaic output and the bus area load, and the strong correlation between the distributed photovoltaic output sequence and the bus area load sequence is judged through the mutual information value.

3. According to claim 2, a method for predicting load within a busbar area under high-proportion distributed photovoltaic access, characterized in that: In the process of performing empirical mode decomposition, the distributed photovoltaic output sequence and the bus area load sequence with strong correlation are decomposed into intrinsic mode function components and residual terms respectively.

4. According to claim 3, a method for predicting load within a busbar area under high-proportion distributed photovoltaic access, characterized in that: In the process of obtaining the intrinsic mode function components and residual terms, according to the original function of the distributed photovoltaic output sequence or the busbar area load sequence, upper and lower envelopes are formed for all maximum and minimum points of the original function based on the cubic difference function, and the average values ​​of the upper and lower envelopes are calculated. According to the difference between the original function and the average value, the intrinsic mode function is constructed, and the intrinsic mode function is separated into the eigenmode function components and residual terms.

5. According to claim 4, a method for predicting load within a busbar area under high-proportion distributed photovoltaic access, characterized in that: In the process of performing empirical mode decomposition, the extreme points are cyclically screened until the standard deviation of two adjacent screening results reaches [0.2, 0.3], so as to realize the process of screening out the eigenmode function through the original function.

6. According to claim 5, a method for predicting load within a busbar area under high-proportion distributed photovoltaic access, characterized in that: In the process of data frequency recognition, the Lempel-Ziv complexity algorithm is used to identify high and low frequency components.

7. According to claim 6, a method for predicting load within a busbar area under high-proportion distributed photovoltaic access, characterized in that: In the process of predicting the high-frequency component, based on the XGBOOST algorithm, an XGBOOST model is constructed through the high-frequency component to obtain the prediction result of the high-frequency component.

8. According to claim 7, a method for predicting load within a busbar area under high-proportion distributed photovoltaic access, characterized in that: In the process of constructing the XGBOOST model, the error term and the regularization term are used as the loss function of the model, the model is trained using an addition strategy, and the objective function is gradually optimized using a forward distribution algorithm to construct the XGBOOST model.

9. According to claim 8, a method for predicting load within a busbar area under high-proportion distributed photovoltaic access, characterized in that: In the process of obtaining the prediction results of the low-frequency components and the residual terms, the prediction results are obtained based on the low-frequency components and the residual terms respectively based on the extreme learning machine model, wherein the extreme learning machine is a single hidden layer feedforward neural network model consisting of an input layer, a hidden layer and an output layer, and the neurons are fully connected.

10. A method for predicting load within a busbar area under high-proportion distributed photovoltaic access according to any one of claims 1 to 9, characterized in that: A method for predicting load within a busbar area is implemented by a load prediction system within a busbar area under high-proportion distributed photovoltaic access. The system includes: The data acquisition module is used to collect the load variation curve over time and the distributed photovoltaic power generation curve within the distribution network bus area to build a data set; A data processing module is used to determine the correlation between data by calculating the mutual information value between the distributed photovoltaic output and the busbar area load based on the data set; The component prediction module is used to perform empirical mode decomposition on the data with strong correlation, identify the data frequency, and obtain the prediction results of high-frequency components, low-frequency components and residual terms respectively; The load forecasting module is used to obtain the final forecast value of the load forecast within the bus area by adding the forecast results obtained separately through vectors.

Images