Power load data preprocessing method based on secondary decomposition and reconstruction data

Abstract

The present invention provides a method for preprocessing electric load data based on secondary decomposition and reconstruction of data, decomposing the original data and performing secondary decomposition and reconstruction of the components. The secondary decomposition method proposed by the present invention is to decompose the original sequence twice, use two decomposition methods, and select the high-complexity components of the first decomposition for secondary decomposition, which can further reduce the non-stationarity and complexity of the data, so that the internal useful information and rules can be extracted, thereby improving the prediction accuracy. At the same time, based on the secondary decomposition technology, the present invention proposes a secondary decomposition and reconstruction strategy, analyzes and reconstructs each component obtained after decomposition, extracts the regular information of the data, and reduces the non-stationarity of the data. The present invention reveals the change trend in the electric load data, mines the internal change rules of the data, and provides support for the operation, decision-making and subsequent load forecasting of the power system.

Description

Technical Field

[0001] The present invention relates to the technical field of power load data preprocessing, and in particular to a power load data preprocessing method based on secondary decomposition and reconstruction of data. Background Art

[0002] Power load data refers to the amount of electricity consumed within a specific time frame, usually recorded on an hourly, daily or monthly basis. This data plays an important role in assessing changes in power grid demand, planning power supply and optimizing energy use. Power load data can help power companies and power supply departments to rationally allocate power resources and ensure the stability and reliability of power supply. However, power load data usually exhibits characteristics such as nonlinearity, non-stationarity and random fluctuations. The original data is highly complex and weakly correlated, so data preprocessing is required for power load data.

[0003] Current technologies mainly use signal decomposition technology to decompose raw data, thereby extracting the inherent laws and useful information of the data. However, most technologies only use the primary decomposition method, and the residual components obtained still contain a lot of non-stationarity. Although some scholars have proposed secondary decomposition methods, the existing secondary decomposition models do not analyze the characteristics of the components after decomposition, but instead treat each component in the same way, which will result in the data information of each component cannot be reasonably extracted. Summary of the invention

[0004] The main purpose of the present invention is to provide a method for preprocessing power load data based on secondary decomposition and reconstruction of data, so as to at least solve the problem in the prior art that the power load data has the characteristics of nonlinearity, non-stationarity and strong random volatility, resulting in high complexity and weak correlation of the load data, and thus the data information of the power load data cannot be reasonably extracted.

[0005] In order to achieve the above object, the present invention provides a method for preprocessing power load data based on secondary decomposition and reconstruction data, comprising:

[0006] Step S1: Use STL to decompose the power load data to obtain trend component, seasonal component and residual component;

[0007] Step S2: Use ICEEMDAN to perform secondary decomposition on the residual component to obtain multiple intrinsic mode components;

[0008] Step S3: Calculate the SE value of each eigenmode component by sample entropy, and calculate the MIC value of each eigenmode component and residual component by maximum information number;

[0009] Step S4: comparing the SE value of each eigenmode component with the SE judgment value, comparing the MIC value of each eigenmode component and the residual component with the MIC judgment value, and reconstructing the multiple eigenmode components into high-complexity components or low-complexity components according to the comparison results;

[0010] Step S5: input the trend component, seasonal component, high complexity component and low complexity component into the prediction model for prediction.

[0011] Furthermore, step S1 specifically includes:

[0012] Through Loess, discrete power load data Decomposition into trend components , Seasonal Components and the residual component , the specific calculation formula is as follows:

[0013] .

[0014] Furthermore, the specific steps of the decomposition method of STL are as follows: the decomposition of the STL algorithm consists of two recursive processes, namely the inner loop and the outer loop; the inner loop is used to update the trend component and seasonal component of the time series, and the outer loop is used to calculate the robust weights required for the next round of the inner loop;

[0015] set up , is the result of the kth iteration, then the iterative process of the k+1th inner loop is as follows:

[0016] Step S101: Calculate the detrended power load data sequence ;

[0017] Step S102: The seasonal subsequence of is Loess smoothed to obtain a temporary seasonal series ;

[0018] Step S103: Applying a low-pass filter yields ; Among them, the low-pass filter consists of three moving average processes and one Loess process;

[0019] Step S104: According to and , calculate the seasonal component, the calculation expression of the seasonal component of the k+1th iteration is:

[0020] ;

[0021] Step S105: Calculate the power load data sequence after removing the seasonal component ;

[0022] Step S106: Calculate the trend component and use the Loess smoothing method to After smoothing, the trend component of k+1 iteration is obtained as ;

[0023] Assume that the seasonal component and trend component obtained after one inner cycle are , , then the residual component is calculated as follows:

[0024]

[0025] Step S107: Determination Whether it converges, if it converges, the decomposition result is output, if it does not converge, the robust weight is calculated , and returns to execute S102; wherein, Convergence requires the following formula to be satisfied:

[0026]

[0027] In the subsequent inner loop process, when performing Loess smoothing in step S102 and step S106, the weight value should be multiplied by the robust weight at the corresponding moment based on the original weight value. , in order to reduce the impact of outliers on decomposition, the specific calculation method of robust weights is as follows:

[0028]

[0029] in, is the robust weight corresponding to time point v, is the residual component corresponding to the time point v, median() is the median function, and B(u) is the bisquare weight function.

[0030] Furthermore, step S2 specifically includes:

[0031] Step S201: Residual component sequence Perform EMD decomposition to obtain the first remainder. The specific method is shown in the following formula:

[0032]

[0033] Among them, r1 is the first remainder, is the residual component, is the standard deviation (STD) of the added Gaussian white noise, is the local mean of the residual component, is the function of the kth mode produced by the decomposition, is the process of adding Gaussian white noise, To calculate the overall average;

[0034] Step S202: When k =1, the first modal component is calculated based on p and r1 IMF 1 , the specific method is shown as follows:

[0035]

[0036] Step S203: When k =2, the second modal component is calculated based on r1 and r2 IMF 2 , the specific method is shown as follows:

[0037]

[0038] Step S204: When , calculate the corresponding remainders respectively, K The specific calculation method of the remainder is as follows:

[0039]

[0040] Step S205: Calculate the kth modal component according to the corresponding remainder. The specific calculation method is shown in the following formula:

[0041]

[0042] Step S206: Repeat step S204 and step S205 until the decomposition is completed to obtain multiple eigenmode components.

[0043] Furthermore, the specific steps of the method for calculating sample entropy in step S3 are as follows:

[0044] Assuming that an intrinsic mode component is composed of N data, the sequence formed by the kth intrinsic mode component is:

[0045]

[0046] The intrinsic mode components are organized into a vector sequence of a one-dimensional array m, and the vector sequence is specifically shown in the following formula:

[0047]

[0048] in, Represents the m consecutive values ​​of x starting from the i-th;

[0049] Calculate the distance between two vector sequences. The specific calculation method is shown in the following formula:

[0050] ;

[0051] Count the number of vector sequences whose distance is less than or equal to r , the specific calculation method is shown in the following formula:

[0052]

[0053] in, is the probability of m points matching two sequences, r is the similarity tolerance, which represents the measure of similarity;

[0054] When N is a finite number, the specific calculation method of the SE value of the eigenmode component sequence is as follows:

[0055]

[0056] Where m is the embedding dimension of the moving window for reconstructing the time series.

[0057] Furthermore, the specific steps of the method for calculating the maximum number of information in step S3 are as follows:

[0058] set up and the residual component There is a relationship between them, making them an ordered set , and x and y are in and The direction is divided into grids to calculate the MIC values ​​of the intrinsic mode components and residual components. The specific calculation method of the MIC values ​​of the intrinsic mode components and residual components is shown in the following formula:

[0059]

[0060] in, is and The maximum mutual information of all distributions generated by D on the grid.

[0061] Furthermore, step S4 specifically includes:

[0062] Calculate the SE mean value according to the SE values ​​of each eigenmode component as the SE judgment value, and calculate the MIC mean value according to the MIC values ​​of each eigenmode component and the residual component as the MIC judgment value;

[0063] The SE value of each eigenmode component is compared with the SE judgment value, and the MIC value of each eigenmode component and residual component is compared with the MIC judgment value. The eigenmode components with MIC values ​​lower than the MIC judgment value and SE values ​​higher than the SE judgment value are reconstructed as high-complexity components, and the rest are reconstructed as low-complexity components.

[0064] The beneficial effects of the technical solution of the present invention are:

[0065] Aiming at the problem that the load data has high complexity and weak correlation due to the nonlinearity, non-stationarity and strong random volatility of the power load data, the present invention proposes a power load data preprocessing method based on secondary decomposition and reconstruction data, decomposes the original data and performs secondary decomposition and reconstruction on the components. The secondary decomposition method proposed by the present invention decomposes the original sequence twice, uses two decomposition methods, and selects the high-complexity components of the first decomposition for secondary decomposition, which can further reduce the non-stationarity and complexity of the data, so that the internal useful information and laws can be extracted, thereby improving the prediction accuracy. At the same time, based on the secondary decomposition technology, the present invention proposes a secondary decomposition and reconstruction strategy, analyzes and reconstructs the components obtained after decomposition, extracts the regular information of the data, and reduces the non-stationarity of the data. The present invention reveals the change trend in the power load data, mines the internal change laws in the data, and provides support for the operation, decision-making and subsequent load forecasting of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] The drawings constituting a part of the present application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0067] Figure 1 is a flowchart of a method for preprocessing electric load data based on secondary decomposition and reconstruction data, which can be selected according to an embodiment of the present invention;

[0068] Figure 2 is an STL data decomposition flow chart of an electric load data preprocessing method based on secondary decomposition and reconstruction data, which can be selected according to an embodiment of the present invention;

[0069] Figure 3 It is a decomposition result diagram of performing STL primary decomposition on original short-term power load data according to an optional power load data preprocessing method based on secondary decomposition and reconstruction data according to an embodiment of the present invention;

[0070] Figure 4 This is a decomposition result diagram of performing ICEEMDAN secondary decomposition on residual components according to an optional power load data preprocessing method based on secondary decomposition and reconstruction data according to an embodiment of the present invention. DETAILED DESCRIPTION

[0071] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0072] like Figure 1As shown, the present invention provides a method for preprocessing power load data based on secondary decomposition and reconstruction data, comprising:

[0073] Step S1: Use STL to decompose the power load data to obtain trend component, seasonal component and residual component;

[0074] Step S2: Use ICEEMDAN to perform secondary decomposition on the residual component to obtain multiple intrinsic mode components;

[0075] Step S3: Calculate the SE value of each eigenmode component by sample entropy, and calculate the MIC value of each eigenmode component and residual component by maximum information number;

[0076] Step S4: comparing the SE value of each eigenmode component with the SE judgment value, comparing the MIC value of each eigenmode component and the residual component with the MIC judgment value, and reconstructing the multiple eigenmode components into high-complexity components or low-complexity components according to the comparison results;

[0077] Step S5: input the trend component, seasonal component, high complexity component and low complexity component into the prediction model for prediction.

[0078] like Figure 2 As shown, step S1 specifically includes:

[0079] Using STL to analyze discrete power load data The iterative Loess local weighted regression technique is used to smooth the trend component, and then Loess smoothing is performed to extract the seasonal component of the change. Loess decomposes it into trend components , Seasonal Components and the residual component ,in v Represents the sampling points of the original sequence. These three groups of components can also be expressed as discrete ones, and the calculation is shown in the following formula.

[0080] .

[0081] like Figure 2 As shown in the figure, the specific steps of the STL decomposition method are as follows: The STL algorithm decomposition consists of two recursive processes, namely the inner loop and the outer loop; the inner loop is used to update the trend component and seasonal component of the time series, and the outer loop is used to calculate the robust weights required for the next round of the inner loop;

[0082] set up , is the result of the kth iteration, then the iterative process of the k+1th inner loop is as follows:

[0083] Step S101: Calculate the detrended power load data sequence ;

[0084] Step S102: The seasonal subsequence of is Loess smoothed to obtain a temporary seasonal series ;

[0085] Step S103: Applying a low-pass filter yields ; Among them, the low-pass filter consists of three moving average processes and one Loess process;

[0086] Step S104: According to and , calculate the seasonal component, the calculation expression of the seasonal component of the k+1th iteration is:

[0087] ;

[0088] Among them, minus This is to prevent low frequency power from entering the seasonal quantity.

[0089] Step S105: Calculate the power load data sequence after removing the seasonal component ;

[0090] Step S106: Calculate the trend component and use the Loess smoothing method to After smoothing, the trend component of k+1 iteration is obtained as ;

[0091] Assume that the seasonal component and trend component obtained after one inner cycle are , , then the residual component is calculated as follows:

[0092]

[0093] Step S107: Determine whether convergence has occurred. and is the iteration of the continuous trend component, then If the following conditions are met, it can be considered to be converged and the decomposition result is output. If not, the robust weight is calculated and S102 is executed:

[0094]

[0095] Then, a robust weight is introduced for each time point, and the size of the robust weight reflects the residual component In the extreme case of , the robust weight is calculated as follows:

[0096]

[0097] in, is the robust weight corresponding to time point v, is the residual component corresponding to the time point v, median() is the median function, and B(u) is the bisquare weight function;

[0098] In the subsequent inner loop process, when performing Loess smoothing in step S102 and step S106, the weight value should be multiplied by the robust weight at the corresponding moment based on the original weight value. , in order to reduce the impact of outliers on the decomposition.

[0099] like Figure 3 As shown, Figure 3 It is the decomposition result of STL primary decomposition of the original short-term power load data. Figure 3 It can be seen that the decomposed trend component shows an upward and downward trend with the original sequence, the waveform of the seasonal component is relatively regular and shows periodicity, while the residual component has a large volatility and no obvious trend.

[0100] Since the residual component has a large volatility and still contains a lot of useful information and rules that have not been extracted, it is subjected to secondary decomposition using ICEEMDAN. Step S2 specifically includes:

[0101] Step S201: Residual component sequence Perform EMD decomposition to obtain the first remainder. The specific method is shown in the following formula:

[0102]

[0103] Among them, r1 is the first remainder, is the residual component, is the standard deviation (STD) of the added Gaussian white noise, is the local mean of the residual component, is the function of the kth mode produced by the decomposition, is the process of adding Gaussian white noise, To calculate the overall average;

[0104] Step S202: When k =1, the first modal component is calculated based on p and r1 IMF 1 , the specific method is shown as follows:

[0105]

[0106] Step S203: When k=2, the second modal component is calculated based on r1 and r2 IMF 2 , the specific method is shown as follows:

[0107]

[0108] Step S204: When , calculate the corresponding remainders respectively, K The specific calculation method of the remainder is as follows:

[0109]

[0110] Step S205: Calculate the kth modal component according to the corresponding remainder. The specific calculation method is shown in the following formula:

[0111]

[0112] Step S206: Repeat step S204 and step S205 until the decomposition is completed to obtain multiple eigenmode components.

[0113] like Figure 4 As shown, Figure 4 The decomposition result of ICEEMDAN secondary decomposition is shown in Figure 2. The final residual component is decomposed into 12 intrinsic mode components and one residual term. Then the sample entropy of each component is calculated to analyze the complexity; and the MIC value of each component and the residual component is calculated to analyze the correlation.

[0114] Sample Entropy (SE) is a method to measure the complexity of time series. It determines the complexity of data by measuring the probability of generating new patterns in the data. The greater the probability, the greater the complexity of the data.

[0115] The Maximum Information Coefficient (MIC) is used to describe the degree of correlation between any two groups of random variables under big data. It is not easily affected by outliers and has the characteristics of universality and fairness in data measurement.

[0116] The specific steps of the method for calculating sample entropy in step S3 are as follows:

[0117] Assume that the intrinsic mode component consists of N data, then the intrinsic mode component , the intrinsic mode components are composed into a vector sequence of a one-dimensional array m, and the vector sequence is specifically shown in the following formula:

[0118]

[0119] in, Represents the m consecutive values ​​of x starting from the i-th;

[0120] Calculate the distance between two vector sequences. The specific calculation method is shown in the following formula:

[0121] ;

[0122] Count the number of vector sequences whose distance is less than or equal to r , the specific calculation method is shown in the following formula:

[0123]

[0124] in, is the probability of m points matching two sequences, r is the similarity tolerance, which represents the measure of similarity;

[0125] When N is a finite number, the specific calculation method of the SE value of the eigenmode component sequence is as follows:

[0126]

[0127] Where m is the embedding dimension of the moving window for reconstructing the time series.

[0128] The specific steps of the method for calculating the maximum number of information in step S3 are as follows:

[0129] Set the kth eigenmode component and the residual component There is a relationship between them, making them an ordered set , and x and y are in and The direction is divided into grids to calculate the MIC values ​​of the intrinsic mode components and residual components. The specific calculation method of the MIC values ​​of the intrinsic mode components and residual components is shown in the following formula:

[0130]

[0131] in, is and The maximum mutual information of all distributions generated by D on the grid.

[0132] To determine whether a SE value and MIC are low or high, it is generally necessary to combine the specific application and background. Because there is no fixed standard for the absolute thresholds of SE and MIC, its specific meaning may vary due to factors such as signal type, application field, noise level, etc. A commonly used method is to compare SE and MIC with a certain reference value, such as the average SE and average MIC of the same signal. The present invention compares the MIC value of the current IMF with the average of all IMFs, and compares the SE value of the current IMF with the average of all SEs.

[0133] Step S4 specifically includes:

[0134] Calculate the SE mean value according to the SE values ​​of each eigenmode component as the SE judgment value, and calculate the MIC mean value according to the MIC values ​​of each eigenmode component and the residual component as the MIC judgment value;

[0135] The SE value of each eigenmode component is compared with the SE judgment value, and the MIC value of each eigenmode component and residual component is compared with the MIC judgment value. The eigenmode components with MIC values ​​lower than the MIC judgment value and SE values ​​higher than the SE judgment value are reconstructed as high-complexity components, and the rest are reconstructed as low-complexity components.

[0136] The specific determination results of the eigenmode components of this embodiment are shown in Table 1 below.

[0137]

[0138]

[0139] Step S5: input the trend component, seasonal component, high complexity component and low complexity component into the prediction model for prediction.

[0140] In order to verify the effectiveness of the data preprocessing method of the present invention, the present invention will use a specific embodiment to input the processed data into the prediction model for prediction. The present invention compares the model prediction effects of not performing data preprocessing, using a single decomposition, and using the secondary decomposition and reconstruction strategy proposed by the present invention under three models. The results of the three groups of comparative experiments are shown in Table 2 below. As can be seen from Table 2, for the SVM prediction model, the prediction error of the original load data after STL primary decomposition is lower than that without decomposition, and STL-ICEEMDAN further reduces the prediction error, indicating that the secondary decomposition and reconstruction strategy can effectively reduce the nonlinearity and non-stationarity of the data, thereby performing stable predictions. On the basis of the secondary decomposition of STL-ICEEMDAN, the errors of the LSTM and GRU prediction models are reduced by 0.19, 0.56, 0.53 and 0.91, 0.82, 0.44 respectively compared with only performing STL primary decomposition, thereby verifying the effectiveness of the secondary decomposition and reconstruction strategy of the present invention.

[0141]

[0142]

[0143] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for preprocessing power load data based on secondary decomposition and reconstruction data, characterized in that: include: Step S1: Use STL to decompose the power load data to obtain trend component, seasonal component and residual component, wherein step S1 specifically includes: The discrete power load data Y is converted into V Decomposed into trend component Tv, seasonal component Sv and residual component R v , the specific calculation formula is as follows: Y v =S v +T v +Rv; The specific steps of the STL decomposition method are as follows: the STL algorithm decomposition consists of two recursive processes, namely an inner loop and an outer loop; the inner loop is used to update the trend component and seasonal component of the time series, and the outer loop is used to calculate the robust weights required for the next round of the inner loop; Let S v (k) , T v (k) is the result of the kth iteration, then the iterative process of the k+1th inner loop is as follows: Step S101: Calculate the detrended power load data sequence Y v -T v (k) ; Step S102: Y v -T v (k) The seasonal subsequence of is smoothed by Loess to obtain the temporary seasonal sequence C v (k+1) ; Step S103: v (k+1) Apply a low-pass filter to get L v (k+1) ; Among them, the low-pass filter consists of three moving average processes and one Loess process; Step S104: According to C v (k+1) and L v (k+1) , calculate the seasonal component, the calculation expression of the seasonal component of the k+1th iteration is: S v (k+1) =C v (k+1) -L v (k+1) ; Step S105: Calculate the power load data sequence Y after removing the seasonal component v -S v (k) ; Step S106: Calculate the trend component and use the Loess smoothing method to v -S v (k) After smoothing, the trend component of k+1 iteration is T v (k+1) ; Assume that the seasonal component and trend component obtained after one inner cycle are S v , T v , then the residual component is calculated as follows: R v =Y v -S v -T v Step S107: Determine T v (k) Whether it converges, if it converges, the decomposition result is output, if it does not converge, the robust weight ρ is calculated v , and returns to execute S102; wherein, T v (k) Convergence needs to satisfy the following formula: In the subsequent inner loop process, when performing Loess smoothing in step S102 and step S106, the weight value should be multiplied by the robust weight ρ at the corresponding moment based on the original weight value. v , in order to reduce the impact of outliers on decomposition, the specific calculation method of robust weights is as follows: h=6median(|R v |) r v =B(|R v | / h) Among them, ρ v is the robust weight corresponding to time point v, Rv is the residual component corresponding to time point v, median() is the median function, and B(u) is the bisquare weight function; Step S2: Use ICEEMDAN to perform secondary decomposition on the residual component to obtain multiple intrinsic mode components; Step S3: Calculate the SE value of each eigenmode component by sample entropy, and calculate the MIC value of each eigenmode component and residual component by maximum information number; Step S4: comparing the SE value of each eigenmode component with the SE judgment value, comparing the MIC value of each eigenmode component and the residual component with the MIC judgment value, and reconstructing the multiple eigenmode components into high-complexity components or low-complexity components according to the comparison results; Step S5: input the trend component, seasonal component, high complexity component and low complexity component into the prediction model for prediction.

2. The method for preprocessing electric load data based on secondary decomposition and reconstruction data according to claim 1 is characterized in that: Step S2 specifically includes: Step S201: For the residual component sequence p (i) =p+β0E1(w i ) to perform EMD decomposition and obtain the first remainder. The specific method is shown in the following formula: r1= <N(p i )> Where r1 is the first remainder, p is the residual component, β0 is the standard deviation (STD) of the added Gaussian white noise, N(·) is the local mean of the residual component, and E k (·) is the function of the kth mode generated by decomposition, W (i) is the process of adding Gaussian white noise, and <·> is the calculation of the overall average; Step S202: When k=1, the first modal component IMF1 is calculated according to p and r1. The specific method is shown in the following formula: IMF1=p-r1 Step S203: When k=2, the second modal component IMF2 is calculated according to r1 and r2. The specific method is shown in the following formula: IMF2=r1-r2=r1-<N(r1+β1E2(w i ))> Step S204: When k=3, ..., K, the corresponding remainders are calculated respectively. The specific calculation method of the Kth remainder is shown in the following formula: r k h <N(r k-1 +β k-1 E k (w i ))> Step S205: Calculate the kth modal component according to the corresponding remainder. The specific calculation method is shown in the following formula: IMF k =r k-1 -r k Step S206: Repeat step S204 and step S205 until the decomposition is completed to obtain multiple eigenmode components.

3. The method for preprocessing electric load data based on secondary decomposition and reconstruction data according to claim 1, characterized in that: The specific steps of the method for calculating the sample entropy in step S3 are as follows: Assuming that an intrinsic mode component is composed of N data, the sequence formed by the kth intrinsic mode component is: IMF k =x(1),x(2),…x(N) The intrinsic mode components are organized into a vector sequence of a one-dimensional array m, and the vector sequence is specifically shown in the following formula: X m (1),…,X m (N-M+1) Among them, X m (i) represents the m consecutive values ​​of x starting from the i-th one; Calculate the distance between two vector sequences. The specific calculation method is shown in the following formula: d[X m (i),X m (j)]=max k＝0,…,m-1 (|x(i+k)-x(j+k)|); The number of vector sequences whose distance is less than or equal to r is A m (r), the specific calculation method is shown in the following formula: Among them, A m (r) is the probability of m points matching the two sequences, r is the similarity tolerance, which represents the measure of similarity; When N is a finite number, the specific calculation method of the SE value of the eigenmode component sequence is as follows: Where m is the embedding dimension of the moving window for reconstructing the time series.

4. The method for preprocessing electric load data based on secondary decomposition and reconstruction data according to claim 3 is characterized in that: The specific steps of the method for calculating the maximum number of messages in step S3 are as follows: IMF k = x(1), x(2), ... x(N) and the residual component R v =y(1), y(2), ...y(N) have a relationship between them, let them be an ordered set D(x,y), and put x and y in the IMF k and R v The direction is divided into grids to calculate the MIC values ​​of the intrinsic mode components and residual components. The specific calculation method of the MIC values ​​of the intrinsic mode components and residual components is shown in the following formula: Among them, I(D,IMF k ,R v ) is in the IMF k and R v The maximum mutual information of all distributions generated by D on the grid.

5. The method for preprocessing electric load data based on secondary decomposition and reconstruction data according to claim 4 is characterized in that: Step S4 specifically includes: Calculate the SE mean value according to the SE values ​​of each eigenmode component as the SE judgment value, and calculate the MIC mean value according to the MIC values ​​of each eigenmode component and the residual component as the MIC judgment value; The SE value of each eigenmode component is compared with the SE judgment value, and the MIC value of each eigenmode component and residual component is compared with the MIC judgment value. The eigenmode components with MIC values ​​lower than the MIC judgment value and SE values ​​higher than the SE judgment value are reconstructed as high-complexity components, and the rest are reconstructed as low-complexity components.

Images