Power customer demand monitoring method and system influenced by multi-dimensional factors

Abstract

The invention discloses a method and a system for monitoring power customer appeals influenced by multi-dimensional factors. The method comprises the following steps: performing missing value completion based on weather weighting; coding the non-numerical characteristics of all weather types by adopting one-hot coding; the method comprises the following steps: dividing historical power customer appeal quantity into a high-frequency component and a low-frequency component on the basis of variational mode decomposition VMD optimized by an improved salat swarm optimization algorithm ISCSO; and an MLR model and a TCN model optimized based on ISCSO are constructed, the TCN model is used for predicting a high-frequency component of the power customer demand amount, and the MLR model is used for predicting a low-frequency component of the power customer demand amount. According to the method, weather types are coded through one-hot coding, so that the adaptability of the model to weather changes is enhanced; a TCN model is adopted for high-frequency components so as to capture long-time-sequence characteristics; an MLR model which is simple in structure and high in calculation speed is adopted for low-frequency components; accurate prediction of the power customer demand quantity data of the transformer area is realized.

Description

Technical Field

[0001] The invention relates to the field of power monitoring, and proposes a system and method for monitoring power customer demands influenced by multi-dimensional factors, in particular monitoring the quantity trend of power customer demands in a substation. Background Art

[0002] In the field of power systems and distribution, a substation usually refers to a specific area or unit in the distribution network, which is powered by a distribution transformer. Distribution transformers convert high-voltage electricity into low-voltage electricity suitable for home and commercial use. Each substation can serve a certain number of users, which depends on the capacity of the transformer and the demand of power customers in the area. In this case, the power supply in the substation may not match the actual power dispatch and the demand of power customers during certain periods of time. This situation affects the service quality of power companies and the satisfaction of power customers, and also poses new challenges to the management and operation of the power grid.

[0003] Monitoring the trend of power customer demands in the substation has important practical significance. First, it improves the stability of the power grid. By monitoring the trend of power customer demands, power companies can better understand and predict the load conditions of the power grid, take timely measures to avoid overload or instability, and ensure the stable operation of the power grid. Second, it optimizes the allocation of power resources and understands the changes in power customer demands in different time periods and regions, which helps power companies allocate power resources more in real time and improve the efficiency and reliability of power supply. Third, it improves the satisfaction of power customers and optimizes the power business environment. Through the above measures, power companies can provide more reliable and high-quality power services and improve the satisfaction and quality of life of power customers.

[0004] Existing technologies for monitoring power customer demands include:

[0005] The patent with the publication number CN118861695A proposes a method and system for analyzing hot spots of power customer demands based on user behavior pattern clustering, including: obtaining user load data, preprocessing the user load data, and constructing a load data set; reducing the dimension of the load data set to form a low-dimensional data set, and weighting the data features in the low-dimensional data set to construct a load feature set; constructing a joint parameter index based on the SSE index and the intra-cluster dispersion, and selecting the initial clustering number of clusters of the clustering algorithm; improving the firefly algorithm to optimize the determination of the initial center of the clustering algorithm; and constructing a clustering analysis model based on the determined initial center of the clustering algorithm and the constructed load feature set. However, the present invention only considers user load data and does not consider multidimensional data, such as adding non-numerical features such as weather characteristics, and is not adaptable enough to weather changes.

[0006] The patent with the publication number CN117172833A discloses a method for predicting electric power customer demands based on meteorological factors, including the following steps: using the grey correlation analysis method to analyze the correlation between the electric power customer demands and meteorological factors, establishing a calculation model for the correlation between the electric power customer demands and meteorological factors, quantitatively analyzing the correlation between the customer demands and meteorological factors, clarifying the degree of influence of each meteorological factor on the customer demands, and finding out the meteorological factors that have a greater impact on the customer demands. However, the present invention only calculates the correlation between the electric power customer demands and meteorological factors, and does not take into account the increased prediction difference between the high-frequency and low-frequency data of the electric power customer demands data, which need to be predicted separately, and the prediction accuracy is low. Summary of the invention

[0007] Aiming at the fact that the access of customers to electricity in the substation area is extremely challenging the flexibility of the grid dispatching, the present invention proposes a system and method for monitoring the demands of power customers affected by multi-dimensional factors.

[0008] The present invention adopts the following technical solution.

[0009] The first aspect of the present invention proposes a method for monitoring power customer demands influenced by multi-dimensional factors, which is characterized by comprising the following contents:

[0010] Collect historical electricity customer demands and corresponding weather types in the substation area;

[0011] For the missing data in the historical electricity customer demands of the substation, the weights are calculated according to the weather type, and the missing values ​​are filled based on weather weighting;

[0012] For the weather types in the station area, one-hot encoding is used to encode all non-numerical features such as weather types;

[0013] Based on the variational mode decomposition (VMD) optimized by ISCSO, the historical power customer demands are divided into high-frequency components and low-frequency components.

[0014] The MLR model and the TCN model optimized based on ISCSO were constructed. The high-frequency components of historical power customer demands and their corresponding weather types were used to train the TCN model, and the low-frequency components of historical power customer demands were used to train the MLR model.

[0015] The real-time electricity customer demands and their corresponding weather types are collected, and the high-frequency components, low-frequency components and unique-hot codes of the corresponding weather types of the real-time electricity customer demands are obtained. The high-frequency components of the real-time electricity customer demands and the unique-hot codes of the corresponding weather types are input into the TCN model to output the predicted high-frequency components. The low-frequency components of the real-time electricity customer demands are input into the MLR model to output the predicted low-frequency components. The predicted high-frequency components and the predicted low-frequency components are superimposed and reconstructed to obtain the predicted results of the electricity customer demands.

[0016] Preferably, the missing data in the historical electricity customer demand volume of the substation area is weighted by calculating the weight according to the weather type, and the missing value is supplemented based on weather weighting, specifically:

[0017] The weather types are divided into clear weather, partially clear weather, poor weather and severe weather, and a corresponding value is set for each weather type as the sunshine condition of the weather, and different weights are given to different types of weather according to the sunshine condition;

[0018] Get the winter sunrise and sunset times of the area, divide the interpolation area according to the time, and complete the missing values ​​based on weather weighting. The formula is as follows:

[0019]

[0020] Where d is the dth day, t is the time t, L(d, t), L(d-1, t), and L(d+1, t) are the power customer demand values ​​at time t on the dth day, time t on the d-1th day, and time t on the d+1th day, respectively. Q d , Q d-1 , Q d+1 are the weather weights of the dth day, d-1th day and d+1th day respectively.

[0021] Preferably, the corresponding value is set for each weather as the sunshine condition of the weather, and different weights are taken for different types of weather according to the sunshine condition, specifically:

[0022] When the weather type is sunny, the sunshine condition is set to -180; when the weather type is partly sunny, the sunshine condition is set to -140; when the weather type is poor weather, the sunshine condition is set to -90; when the weather type is severe weather, the sunshine condition is set to -20;

[0023] The weight of bad weather is set to 1, and the weights of other weather types are calculated according to the ratio of the weights of two different weather types and the ratio of their corresponding sunshine conditions. The calculated weight is 4.5 for worse weather, 7 for partly sunny weather, and 9 for sunny weather.

[0024] Preferably, the one-hot encoding is used to encode all non-numerical features such as weather types, specifically:

[0025] Clear weather is defined as state s1, whose unique hot code is 1000; partially clear weather is defined as state s2, whose unique hot code is 0100; poor weather is defined as state s3, whose unique hot code is 0010; severe weather is defined as state s4, whose unique hot code is 0001;

[0026] The four weather types are represented by a set as: {s1: sunny weather 1000, s2: partially sunny weather 0100, s3: poor weather 0010, s4: severe weather 0001}. 5. The method for monitoring power customer demands affected by multi-dimensional factors according to claim 1 is characterized by:

[0027] The ISCSO optimization is specifically as follows:

[0028] Use the ISCSO algorithm to optimize the hyperparameters of VMD and TCN;

[0029] The hyperparameters of VMD and TCN include bandwidth α, number of decomposed modes K and noise tolerance τ.

[0030] Preferably, the specific content of optimizing the hyper parameters of VMD and TCN by the ISCSO algorithm is:

[0031] A sand cat population φ with initialization of hyperparameters α, K, and τ using Tent mapping i , and set the upper and lower limits of the hyperparameters;

[0032] Use the roulette wheel selection algorithm to select the sand cat population φ in the ISCSO algorithm i The individuals in are assigned a random angle θ, where 0°≤θ≤360°;

[0033] Use the following formula to update the search position of individuals in the ISCSO algorithm;

[0034] P(t+1)=r·(P bc (t)-rand(0,1)·P c (t))

[0035] Among them, P(t+1) is the next position of the optimization algorithm, r is the sensitivity range of the sand cat, and P bc (t) is the best candidate position, P c (t) is the current position;

[0036] And according to the state switching formula, switch the search and attack states of the individuals in the population;

[0037] The state switching formula is:

[0038]

[0039] I(X, Y) is the mutual information between X and Y, X is the search state, Y is the attack state, P(x, y) is the joint distribution probability of X and Y, P(x) is the probability of the search state, and P(y) is the probability of the attack state.

[0040] Repeatedly update the search position of individuals in the ISCSO algorithm and switch the search and attack states of individuals in the population until the set objective function reaches the minimum.

[0041] Preferably, the set objective function is the permutation entropy of the power customer demand quantity sequence; the process of calculating the permutation entropy of the power customer demand quantity sequence is:

[0042] The sequence of electricity customer demands [x1 x2…x j …x N ] T Reconstruct the phase space and obtain the matrix Q, as shown below:

[0043]

[0044] Among them, x j is the jth power customer demand; n is the number of embedded bits; t is the delay; each row vector in the matrix Q is a reconstruction component, K is the number of reconstruction components; N is the length of the power customer demand sequence, and the relationship between K and N satisfies K=N-(n-1)×t;

[0045] Arrange the elements in each reconstructed component in ascending order according to the element value, and record the arrangement pattern to obtain the following formula:

[0046] S(p)=(j p1 ,j p2 , …, j po , …, j pn ) p=1,2,…,K;

[0047] S′(p)=(q p1 ,q p2 , …, q po , …, q pn ) p=1,2,…,K;

[0048] Among them, S(p) is the result of arranging the p-th reconstruction component in ascending order, j po is the oth element after the reconstruction components are arranged in ascending order; S′(p) is the arrangement pattern of the pth reconstruction component; q po Represents the oth element j after the reconstruction components are arranged in ascending order po is the qth before the reconstruction components are arranged in ascending orderpo elements;

[0049] Calculate the number of times the arrangement pattern of each reconstructed component appears in all arrangement patterns as the probability of the arrangement pattern of the reconstructed component. All probabilities are {P1, P2, ..., P u , ..., P K};

[0050] Calculate the permutation entropy H of the power customer demand sequence p The formula is:

[0051]

[0052] Among them, P u is the probability of occurrence of the arrangement pattern of the u-th reconstruction component.

[0053] Preferably, the tent mapping is specifically: the tent mapping is to replace the pseudo-random number initial population generator in the sand cat population algorithm SCSO, and the formula is:

[0054]

[0055] Where: φ i is the initial solution space of the sand cat population, φi +1 is the solution space at time t+1, and γ is the hyperparameter of the Tent mapping.

[0056] Preferably, the variational mode decomposition VMD based on ISCSO optimization divides the historical power customer demand into high-frequency components and low-frequency components; specifically:

[0057] After decomposing the historical power customer demand data using variational mode decomposition (VMD) optimized based on ISCSO, multiple components are output; the zero-crossing rate of each component is calculated. If the adjacent sampled power customer demand quantities have different algebraic signs, a zero crossing occurs. The zero-crossing rate is the ratio of the number of zero crossings to the number of sampling times. When the zero-crossing rate of a component is greater than or equal to the set threshold, it is a high-frequency component, and if it is less than the threshold, it is a low-frequency component.

[0058] The second aspect of the present invention proposes a multi-dimensional factor-influenced power customer demand monitoring system using the method described in the first aspect of the present invention, comprising a smart meter and a meteorological sensor, a missing value completion module, an encoding module, a decomposition module, a prediction model training module and a prediction module; characterized in that:

[0059] The smart meter and weather sensor collect the demand of electricity customers and the corresponding weather type;

[0060] Missing value completion module: used to calculate the weights according to the weather type for missing data in the power customer demand volume in the substation area, and perform missing value completion based on weather weighting;

[0061] Coding module: It is used to encode all weather types such as non-numerical features using one-hot encoding according to the weather types in the station area;

[0062] Decomposition module: used to divide the power customer demand into high-frequency components and low-frequency components based on the variational mode decomposition VMD optimized by ISCSO;

[0063] Prediction model training module: used to build the MLR model and the TCN model optimized based on ISCSO. The TCN model is used to predict the high-frequency components of the power customer's demands, and the MLR model is used to predict the low-frequency components of the power customer's demands. The TCN model is trained using the unique code of the high-frequency components of the historical power customer's demands and the corresponding weather types, and the prediction model is trained using the low-frequency components of the historical power customer's demands.

[0064] Prediction module: The high-frequency component of the real-time power customer demand and the unique-hot code of the corresponding weather type are input into the TCN model, and the predicted high-frequency component is output. The low-frequency component of the real-time power customer demand is input into the MLR model, and the predicted low-frequency component is output. The predicted high-frequency component and the predicted low-frequency component are superimposed and reconstructed to obtain the predicted result of the power customer demand.

[0065] The beneficial effect of the present invention is that, compared with the prior art, weather-weighted missing value completion is performed, and weather characteristics are input into the neural network as non-numerical features, and the weather type is encoded by one-hot encoding, thereby enhancing the adaptability of the model to weather changes.

[0066] The VMD decomposition method optimized based on the improved SCSO (ISCSO) algorithm effectively separates the high-frequency and low-frequency components of power customer demand data, and optimizes the hyperparameters of VMD and TCN.

[0067] For high-frequency components, the TCN model is used to capture long-term time series characteristics; for low-frequency components, the MLR model with a simple structure and fast calculation speed is used; through this combination method, the advantages of each model can be fully utilized, while making up for the shortcomings of the other model, thereby achieving accurate prediction of the demand volume data of power customers in the substation. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 A framework diagram of a method for monitoring electric power customer demands in an embodiment of the present invention;

[0069] Figure 2 A flowchart of a method for monitoring electric power customer demands in an embodiment of the present invention;

[0070] Figure 3 It is an ISCSO optimization VMD iteration diagram in an embodiment of the present invention;

[0071] Figure 4 It is a development exploration diagram of ISCSO optimal VMD in an embodiment of the present invention;

[0072] Figure 5 This is a result diagram of VMD decomposition of power customer demands after ISCSO optimization in an embodiment of the present invention;

[0073] Figure 6 This is a comparison chart of the decomposed power customer demand volume and the original power customer demand volume in an embodiment of the present invention. DETAILED DESCRIPTION

[0074] In order to make the purpose, technical scheme and advantages of the present invention clearer, the technical scheme of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The embodiments described in this application are only embodiments of a part of the present invention, rather than all embodiments. Based on the spirit of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the protection scope of the present invention.

[0075] like Figure 1 and Figure 2 As shown, Embodiment 1 of the present invention proposes a method for monitoring power customer demands influenced by multi-dimensional factors, which is characterized by including the following contents:

[0076] Collect historical electricity customer demands and corresponding weather types in the substation area;

[0077] For the missing data in the historical electricity customer demands of the substation, the weights are calculated according to the weather type, and the missing values ​​are filled based on weather weighting;

[0078] For the weather types in the station area, one-hot encoding is used to encode all non-numerical features such as weather types;

[0079] Based on the variational mode decomposition (VMD) optimized by ISCSO, the historical power customer demands are divided into high-frequency components and low-frequency components.

[0080] The MLR model and the TCN model optimized based on ISCSO were constructed. The high-frequency components of historical power customer demands and their corresponding weather types were used to train the TCN model, and the low-frequency components of historical power customer demands were used to train the MLR model.

[0081] The real-time electricity customer demands and their corresponding weather types are collected, and the high-frequency components, low-frequency components and unique-hot codes of the corresponding weather types of the real-time electricity customer demands are obtained. The high-frequency components of the real-time electricity customer demands and the unique-hot codes of the corresponding weather types are input into the TCN model to output the predicted high-frequency components. The low-frequency components of the real-time electricity customer demands are input into the MLR model to output the predicted low-frequency components. The predicted high-frequency components and the predicted low-frequency components are superimposed and reconstructed to obtain the predicted results of the electricity customer demands.

[0082] It should be noted that, specifically, the number of convolutional layer filters nb_filters of the TCN model described in this embodiment is 61, the size of the kernel kernel_size used in each convolutional layer is 3, and the learning rate η is 0.00214.

[0083] Preferably, the missing data in the historical electricity customer demand volume of the substation area is weighted by calculating the weight according to the weather type, and the missing value is supplemented based on weather weighting, specifically:

[0084] The weather types are divided into clear weather, partially clear weather, poor weather and severe weather, and a corresponding value is set for each weather type as the sunshine condition of the weather, and different weights are given to different types of weather according to the sunshine condition;

[0085] Get the winter sunrise and sunset times of the area, divide the interpolation area according to the time, and complete the missing values ​​based on weather weighting. The formula is as follows:

[0086]

[0087] Where d is the dth day, t is the time t, L(d, t), L(d-1, t), and L(d+1, t) are the power customer demand values ​​at time t on the dth day, time t on the d-1th day, and time t on the d+1th day, respectively. Q d , Q d-1 , Q d+1 are the weather weights of the dth day, d-1th day and d+1th day respectively.

[0088] Preferably, the corresponding value is set for each weather as the sunshine condition of the weather, and different weights are taken for different types of weather according to the sunshine condition, specifically:

[0089] When the weather type is sunny, the sunshine condition is set to -180; when the weather type is partly sunny, the sunshine condition is set to -140; when the weather type is poor weather, the sunshine condition is set to -90; when the weather type is severe weather, the sunshine condition is set to -20;

[0090] The weight of bad weather is set to 1, and the weights of other weather types are calculated according to the ratio of the weights of two different weather types and the ratio of their corresponding sunshine conditions. The calculated weight is 4.5 for worse weather, 7 for partly sunny weather, and 9 for sunny weather.

[0091] Preferably, the one-hot encoding is used to encode all non-numerical features such as weather types, specifically:

[0092] Clear weather is defined as state s1, whose unique hot code is 1000; partially clear weather is defined as state s2, whose unique hot code is 0100; poor weather is defined as state s3, whose unique hot code is 0010; severe weather is defined as state s4, whose unique hot code is 0001;

[0093] The four weather types are represented by a set as: {s1: sunny weather 1000, s2: partially sunny weather 0100, s3: poor weather 0010, s4: severe weather 0001}. 5. The method for monitoring power customer demands affected by multi-dimensional factors according to claim 1 is characterized by:

[0094] The ISCSO optimization is specifically as follows:

[0095] Use the ISCSO algorithm to optimize the hyperparameters of VMD and TCN;

[0096] It should be noted that the hyperparameters are parameters that will not be optimized or changed during the model training process, but need to be manually set before training and will affect the model performance;

[0097] The optimized hyperparameters of VMD and TCN include bandwidth α, number of decomposed modes K and noise tolerance τ.

[0098] Preferably, the specific content of optimizing the hyper parameters of VMD and TCN by the ISCSO algorithm is:

[0099] A sand cat population φ with initialization of hyperparameters α, K, and τ using Tent mapping i , and set the upper and lower limits of the hyperparameters;

[0100] Specifically, in this embodiment, for the hyperparameter α, the upper and lower limits are 32000 and 29000 respectively; for K, the upper and lower limits are 7 and 1 respectively; for τ, the upper and lower limits are 1 and 0 respectively.

[0101] Use the roulette wheel selection algorithm to select the sand cat population φ in the ISCSO algorithm i The individuals in are assigned a random angle θ, where 0°≤θ≤360°;

[0102] It should be noted that the improved sand cat swarm optimization algorithm assumes that the perception range of sand cats is a circle, and the movement direction of sand cats is determined by random angles on the circumference. By assigning random angles, each sand cat in the sand cat swarm can explore along the boundary of the search space. In order to prevent the algorithm from falling into a local optimal solution, a roulette selection mechanism is used to determine a random direction of travel for each sand cat.

[0103] Use the following formula to update the search position of individuals in the ISCSO algorithm;

[0104] P(t+1)=r·(P bc (t)-rand(0,1)·P c (t))

[0105] Among them, P(t+1) is the next position of the optimization algorithm, r is the sensitivity range of the sand cat, and P bc (t) is the best candidate position, P c (t) is the current position;

[0106] And according to the state switching formula, switch the search and attack states of the individuals in the population;

[0107] The state switching formula is:

[0108]

[0109] I(X, Y) is the mutual information between X and Y, X is the search state, Y is the attack state, P(x, y) is the joint distribution probability of X and Y, P(x) is the probability of the search state, and P(y) is the probability of the attack state.

[0110] Repeatedly update the search position of individuals in the ISCSO algorithm and switch the search and attack states of individuals in the population until the set objective function reaches the minimum.

[0111] Preferably, the set objective function is the permutation entropy of the power customer demand quantity sequence; the process of calculating the permutation entropy of the power customer demand quantity sequence is:

[0112] The sequence of electricity customer demands [x1 x2…x j …x N ] T Reconstruct the phase space and obtain the matrix Q, as shown below:

[0113]

[0114] Among them, x jis the jth power customer demand; n is the number of embedded bits; t is the delay; each row vector in the matrix Q is a reconstruction component, K is the number of reconstruction components; N is the length of the power customer demand sequence, and the relationship between K and N satisfies K=N-(n-1)×t;

[0115] Arrange the elements in each reconstructed component in ascending order according to the element value, and record the arrangement pattern to obtain the following formula:

[0116] S(p)=(j p1 ,j p2 , …, j po , …, j pn ) p=1,2,…,K;

[0117] S′(p)=(q p1 ,q p2 , …, q po , …, q pn ) p=1,2,…,K;

[0118] Among them, S(p) is the result of arranging the p-th reconstruction component in ascending order, j po is the oth element after the reconstruction components are arranged in ascending order; S′(p) is the arrangement pattern of the pth reconstruction component; q po Represents the oth element j after the reconstruction components are arranged in ascending order po is the qth before the reconstruction components are arranged in ascending order po elements;

[0119] Calculate the number of times the arrangement pattern of each reconstructed component appears in all arrangement patterns as the probability of the arrangement pattern of the reconstructed component. All probabilities are {P1, P2, ..., P u , ..., P K};

[0120] Calculate the permutation entropy H of the power customer demand sequence p The formula is:

[0121]

[0122] Among them, P u is the probability of occurrence of the arrangement pattern of the u-th reconstruction component.

[0123] Preferably, the tent mapping is specifically: the tent mapping is to replace the pseudo-random number initial population generator in the sand cat population algorithm SCSO, and the formula is:

[0124]

[0125] Where: φ i is the initial solution space of the sand cat population, φi+1 is the solution space at time t+1, and γ is the hyperparameter of the Tent mapping.

[0126] It should be noted that the population size of the improved sand cat colony optimization algorithm in this embodiment is set to 50, the maximum number of iterations is 30, and the change of permutation entropy during the iteration process is as follows: Figure 3 As shown in the figure, the optimal solution of the hyperparameters after optimization is: α = 290921, K = 3, τ = 0.8426; the exploration progress and development progress of the improved sand cat swarm optimization algorithm in the iterative process are shown in Figure 4 shown.

[0127] Preferably, the variational mode decomposition VMD based on the improved sand cat swarm optimization algorithm ISCSO is used to divide the historical power customer demands into high-frequency components and low-frequency components; specifically:

[0128] After decomposing the historical power customer demand data using variational mode decomposition VMD optimized by the improved sand cat swarm optimization algorithm ISCSO, multiple components are output; the zero-crossing rate of each component is calculated. If the adjacent sampled power customer demand has different algebraic signs, a zero-crossing occurs. The zero-crossing rate is the ratio of the number of zero-crossings to the number of samplings. When the zero-crossing rate of a component is greater than or equal to the set threshold, it is a high-frequency component, and less than it is a low-frequency component. The decomposition results are as follows: Figure 5 shown.

[0129] In order to quantify the loss after decomposition, this embodiment uses R 2 The score is the evaluation index, and the R value of the power customer demand data after the three components are synthesized and the original power customer demand data is calculated. 2 The ratio of the values ​​is 0.9914, and Figure 6 The plot shown is a comparison of the partially decomposed synthetic data with the original electricity customer demand data.

[0130] In order to verify the effectiveness of the present invention, this embodiment also built LSTM, GRU, and TCN as comparison models of the hybrid prediction model of this embodiment. After the construction was completed, the prediction results of the LSTM, GRU, and TCN models were compared and it was found that the hybrid prediction model proposed in this embodiment achieved the best results in terms of the three evaluation indicators of MAE, RMSE, and MSE. It is the best fitting effect with the actual power customer demand data among the comparison models, and the prediction accuracy reached 98.260%.

[0131] Embodiment 2 of the present invention proposes a multi-dimensional factor-influenced power customer demand monitoring system using the method described in Embodiment 1 of the present invention, including a smart meter and a meteorological sensor, a missing value completion module, an encoding module, a decomposition module, a prediction model training module and a prediction module; characterized in that:

[0132] The smart meter and weather sensor collect the demand of electricity customers and the corresponding weather type;

[0133] Missing value completion module: used to calculate the weights according to the weather type for missing data in the power customer demand volume in the substation area, and perform missing value completion based on weather weighting;

[0134] Coding module: It is used to encode all weather types such as non-numerical features using one-hot encoding according to the weather types in the station area;

[0135] Decomposition module: used to divide the power customer demand into high-frequency components and low-frequency components based on the variational mode decomposition VMD optimized by ISCSO;

[0136] Prediction model training module: used to build the MLR model and the TCN model optimized based on ISCSO. The TCN model is used to predict the high-frequency components of the power customer's demands, and the MLR model is used to predict the low-frequency components of the power customer's demands. The TCN model is trained using the unique code of the high-frequency components of the historical power customer's demands and the corresponding weather types, and the prediction model is trained using the low-frequency components of the historical power customer's demands.

[0137] Prediction module: The high-frequency component of the real-time power customer demand and the unique-hot code of the corresponding weather type are input into the TCN model, and the predicted high-frequency component is output. The low-frequency component of the real-time power customer demand is input into the MLR model, and the predicted low-frequency component is output. The predicted high-frequency component and the predicted low-frequency component are superimposed and reconstructed to obtain the predicted result of the power customer demand.

[0138] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0139] A computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples of computer-readable storage media (a non-exhaustive list) include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination of the foregoing. As used herein, a computer-readable storage medium is not to be interpreted as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through a wire.

[0140] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0141] The computer program instructions for performing the operation of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. Computer-readable program instructions may be executed completely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be customized by utilizing the state information of the computer-readable program instructions, and the electronic circuit may execute the computer-readable program instructions, thereby realizing various aspects of the present disclosure.

[0142] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for monitoring power customer demands influenced by multi-dimensional factors, characterized in that: It includes the following: Collect historical electricity customer demands and corresponding weather types in the substation area; For the missing data in the historical electricity customer demands of the substation, the weights are calculated according to the weather type, and the missing values ​​are filled based on weather weighting; For the weather types in the station area, one-hot encoding is used to encode all non-numerical features such as weather types; Based on the variational mode decomposition (VMD) optimized by ISCSO, the historical power customer demands are divided into high-frequency components and low-frequency components. The MLR model and the TCN model optimized based on ISCSO were constructed. The high-frequency components of historical power customer demands and their corresponding weather types were used to train the TCN model, and the low-frequency components of historical power customer demands were used to train the MLR model. The real-time electricity customer demands and their corresponding weather types are collected, and the high-frequency components, low-frequency components and unique-hot codes of the corresponding weather types of the real-time electricity customer demands are obtained. The high-frequency components of the real-time electricity customer demands and the unique-hot codes of the corresponding weather types are input into the TCN model to output the predicted high-frequency components. The low-frequency components of the real-time electricity customer demands are input into the MLR model to output the predicted low-frequency components. The predicted high-frequency components and the predicted low-frequency components are superimposed and reconstructed to obtain the predicted results of the electricity customer demands.

2. The method for monitoring power customer demands influenced by multi-dimensional factors according to claim 1, characterized in that: The missing data in the historical electricity customer demand volume of the substation area is calculated by weight according to the weather type, and the missing value is filled based on weather weighting, specifically: The weather types are divided into clear weather, partially clear weather, poor weather and severe weather, and a corresponding value is set for each weather type as the sunshine condition of the weather, and different weights are given to different types of weather according to the sunshine condition; Get the winter sunrise and sunset times of the area, divide the interpolation area according to the time, and complete the missing values ​​based on weather weighting. The formula is as follows: Where d is the dth day, t is the time t, L(d,t), L(d-1,t), and L(d+1,t) are the power customer demand values ​​at time t on the dth day, time t on the d-1th day, and time t on the d+1th day, respectively. Q d , Q d-1 , Q d+1 are the weather weights of the dth day, d-1th day and d+1th day respectively.

3. The method for monitoring power customer demands influenced by multi-dimensional factors according to claim 2, characterized in that: The corresponding value is set for each weather type as the sunshine condition of the weather, and different weights are taken for different types of weather according to the sunshine condition, specifically: When the weather type is sunny, the sunshine condition is set to -180; when the weather type is partly sunny, the sunshine condition is set to -140; when the weather type is poor weather, the sunshine condition is set to -90; when the weather type is severe weather, the sunshine condition is set to -20; The weight of bad weather is set to 1, and the weights of other weather types are calculated according to the ratio of the weights of two different weather types and the ratio of their corresponding sunshine conditions. The calculated weight is 4.5 for worse weather, 7 for partly sunny weather, and 9 for sunny weather.

4. The method for monitoring power customer demands influenced by multi-dimensional factors according to claim 3 is characterized by: The one-hot encoding is used to encode all non-numerical features such as weather types, specifically: Clear weather is defined as state s1, whose unique hot code is 1000; partially clear weather is defined as state s2, whose unique hot code is 0100; poor weather is defined as state s3, whose unique hot code is 0010; severe weather is defined as state s4, whose unique hot code is 0001; The four weather types are represented by sets: {s1: sunny weather 1000, s2: partly sunny weather 0100, s3: poor weather 0010, s4: severe weather 0001}.

5. The method for monitoring power customer demands influenced by multi-dimensional factors according to claim 1, characterized in that: The ISCSO optimization is specifically as follows: Use the ISCSO algorithm to optimize the hyperparameters of VMD and TCN; The optimized hyperparameters of VMD and TCN include bandwidth α, number of decomposed modes K and noise tolerance τ.

6. The method for monitoring power customer demands influenced by multi-dimensional factors according to claim 2, characterized in that: The specific content of the ISCSO algorithm for optimizing the hyperparameters of VMD and TCN is: A sand cat population φ with initialization of hyperparameters α, K, and τ using Tent mapping i , and set the upper and lower limits of the hyperparameters; Use the roulette wheel selection algorithm to select the sand cat population φ in the ISCSO algorithm i The individuals in are assigned a random angle θ, where 0°≤θ≤360°; Use the following formula to update the search position of individuals in the ISCSO algorithm; P(t+1)=r·(P bc (t)-rand(0,1)·P c (t)) Among them, P(t+1) is the next position of the optimization algorithm, r is the sensitivity range of the sand cat, and P bc (t) is the best candidate position, P c (t) is the current position; And according to the state switching formula, switch the search and attack states of the individuals in the population; The state switching formula is: I(X,Y) is the mutual information between X and Y, X is the search state, Y is the attack state, P(x,y) is the joint distribution probability of X and Y, P(x) is the probability of the search state, and P(y) is the probability of the attack state. Repeatedly update the search position of individuals in the ISCSO algorithm and switch the search and attack states of individuals in the population until the set objective function reaches the minimum.

7. The method for monitoring power customer demands influenced by multi-dimensional factors according to claim 6, characterized in that: The objective function set is the permutation entropy of the power customer demand quantity sequence; the process of calculating the permutation entropy of the power customer demand quantity sequence is: The sequence of electricity customer demands [x1 x2 … x j … x N ] T Reconstruct the phase space and obtain the matrix Q, as shown below: Among them, x j is the jth power customer demand; n is the number of embedded bits; t is the delay; each row vector in the matrix Q is a reconstruction component, K is the number of reconstruction components; N is the length of the power customer demand sequence, and the relationship between K and N satisfies K=N-(n-1)×t; Arrange the elements in each reconstructed component in ascending order according to the element value, and record the arrangement pattern to obtain the following formula: S(p)=(j p1 ,j p2 ,…,j po ,…,j p n) p=1,2,...,K; S'(p)=(q p1 ,q p2 ,…,q po ,…,q pn ) p=1,2,...,K; Among them, S(p) is the result of arranging the p-th reconstruction component in ascending order, j po is the oth element after the reconstruction components are arranged in ascending order; S'(p) is the arrangement pattern of the pth reconstruction component; q po Represents the oth element j after the reconstruction components are arranged in ascending order po is the qth before the reconstruction components are arranged in ascending order po elements; Calculate the number of times the permutation pattern of each reconstructed component appears in all permutation patterns as the probability of the permutation pattern of the reconstructed component. All probabilities are {P1, P2, …, P u ,…,P K }; Calculate the permutation entropy H of the power customer demand sequence p The formula is: Among them, P u is the probability of occurrence of the arrangement pattern of the u-th reconstruction component.

8. The method for monitoring power customer demands influenced by multi-dimensional factors according to claim 6, characterized in that: The tent mapping is specifically: the tent mapping is to replace the pseudo-random number initial population generator in the sand cat population algorithm SCSO, and the formula is: Where: φ i is the initial solution space of the sand cat population, φ i+1 is the solution space at time t+1, and γ is the hyperparameter of the Tent mapping.

9. The method for monitoring electric power customer demands influenced by multi-dimensional factors according to claim 1, characterized in that: The variational mode decomposition VMD based on ISCSO optimization divides the historical power customer demands into high-frequency components and low-frequency components; specifically: After decomposing the historical power customer demand data using variational mode decomposition (VMD) optimized based on ISCSO, multiple components are output; The zero-crossing rate of each component is calculated. If the adjacent sampled power customer demand quantities have different algebraic signs, a zero-crossing occurs. The zero-crossing rate is the ratio of the number of zero-crossings to the number of sampling times. When the zero-crossing rate of a component is greater than or equal to the set threshold, it is a high-frequency component, and if it is less than the threshold, it is a low-frequency component.

10. A multi-dimensional factor-influenced power customer demand monitoring system using the method according to any one of claims 1 to 9, comprising a smart meter and a meteorological sensor, a missing value completion module, an encoding module, a decomposition module, a prediction model training module and a prediction module; characterized in that: The smart meter and weather sensor collect the demand of electricity customers and the corresponding weather type; Missing value completion module: used to calculate the weights according to the weather type for missing data in the power customer demand volume in the substation area, and perform missing value completion based on weather weighting; Coding module: It is used to encode all weather types such as non-numerical features using one-hot encoding according to the weather types in the station area; Decomposition module: used to divide the power customer demand into high-frequency components and low-frequency components based on the variational mode decomposition VMD optimized by ISCSO; Prediction model training module: used to build the MLR model and the TCN model optimized based on ISCSO. The TCN model is used to predict the high-frequency components of the power customer's demands, and the MLR model is used to predict the low-frequency components of the power customer's demands. The TCN model is trained using the unique code of the high-frequency components of the historical power customer's demands and the corresponding weather types, and the prediction model is trained using the low-frequency components of the historical power customer's demands. Prediction module: The high-frequency component of the real-time power customer demand and the unique-hot code of the corresponding weather type are input into the TCN model, and the predicted high-frequency component is output. The low-frequency component of the real-time power customer demand is input into the MLR model, and the predicted low-frequency component is output. The predicted high-frequency component and the predicted low-frequency component are superimposed and reconstructed to obtain the predicted result of the power customer demand.

Images