Power utilization address and electric quantity coupled fine identification method and device for power utilization function area, storage medium and computer system

Abstract

The invention relates to the technical field of electric power data analysis, in particular to a power utilization function area fine identification method and device coupling a power utilization address and electric quantity, a storage medium and a computer system. Performing multi-source data standardization processing; splicing the text semantic representation vector and the electricity consumption sequence representation vector to obtain a joint representation vector Xi of the electricity consumption address; training to obtain an electricity utilization function area fine identification depth model by utilizing labeling data containing urban electricity utilization, industrial electricity utilization, commercial electricity utilization and agricultural electricity utilization labels; distinguishing the type of the functional area; the method specifically comprises the following steps: inputting joint representation vectors X1, X2,..., Xn into a trained depth model, and obtaining prediction probability distribution of each power utilization address; and taking the category with the maximum average prediction probability as the power utilization function area type of the area. According to the invention, a reliable basis is provided for reasonable configuration and optimal management of electric power resources.

Description

Technical Field

[0001] The present invention relates to the technical field of electrical data processing, and in particular to a method, device, storage medium and computer system for finely identifying electrical functional areas by coupling electrical address and electrical quantity. Background Art

[0002] In the process of power resource management and optimal allocation, accurate identification of power consumption functional zones is of great significance for rationally planning power grids, formulating power consumption policies, and improving energy efficiency. Traditional power consumption functional zone identification methods often rely solely on single power consumption characteristics or simple statistical analysis. For example, based solely on the type of power equipment and the approximate power consumption range, it is difficult to fully and accurately reflect the actual situation of the power consumption functional zone. These methods ignore the rich semantic information contained in the power consumption address and the variation pattern of power consumption in the time dimension, resulting in low recognition accuracy and unable to meet the needs of refined management of modern power systems. With the advent of the era of power big data, how to effectively integrate power consumption address information and power consumption data to achieve refined identification of power consumption functional zones has become an urgent problem to be solved.

[0003] In order to solve the above problems, there is an urgent need for a method for fine identification of power consumption functional areas that couples power consumption addresses and power consumption. It can construct a joint representation of power consumption data and improve the accuracy and precision of power consumption functional area identification through computer large model technology, providing a reliable basis for the rational allocation and optimal management of power resources. Summary of the Invention

[0004] The purpose of the present invention is to provide a method, device, storage medium and computer system for fine identification of power consumption functional areas by coupling power consumption addresses and power quantities, so as to improve the accuracy and fineness of power consumption functional area identification.

[0005] In a first aspect, the present invention provides a method for finely identifying power usage functional areas by coupling power usage addresses and power consumption, comprising the following steps:

[0006] S100: Data acquisition; there are n electricity addresses in the specified area, and the electricity address set is represented as A = {a1, a2, ..., a n}, for each power address a i (i=1,2,...,n), and the hourly electricity consumption sequence in the time interval [t1,t2] is expressed as E i ={e i,1 ,e i,2 ,...,e i,m}, where m is the total number of hours in the time interval, e i,j Indicates the power address a i Electricity consumption at hour j;

[0007] S200: Multi-source data standardization processing: standardize the multi-source data and standardize the electricity consumption addresses of the electricity consumption set A. The standardized conversion formula is:

[0008]

[0009] is the converted address set, is the power-based GIS database conversion function, δ represents the format agreement function based on regular expressions;

[0010] S300: Construct a joint representation; construct a joint representation of the electricity usage address by splicing, including the text semantic representation of the electricity usage address and the hourly electricity usage sequence representation within a week, and splice the text semantic representation vector and the electricity usage sequence representation vector to obtain the joint representation vector X of the electricity usage address. i ;

[0011] S400: Training a deep model; using labeled data containing urban electricity, industrial electricity, commercial electricity, and agricultural electricity labels, a deep model for fine-grained identification of electricity functional areas is trained;

[0012] S500: Determine the type of functional area; specifically, the process includes the following:

[0013] S510: The area to be identified contains n electricity addresses, and their joint representation vectors X1, X2, ..., X n Input into the deep model trained in S400 to obtain the predicted probability distribution of each electricity consumption address:

[0014] P i ={p i,城市 ,p i,工业 ,p i,商业 ,p i,农业}(i=1,2,...,n);

[0015] S520: Calculate the average predicted probability of each power consumption functional area type in the region. Taking urban power consumption as an example, the calculation formula is:

[0016]

[0017] The same logic applies

[0018] S530: The category with the largest average predicted probability is used as the electricity consumption functional area type of the area. The calculation formula is:

[0019]

[0020] Where V is the output ribbon type.

[0021] Preferably, the S300 specifically includes the following process:

[0022] S310: The word vector model maps each word into a d1-dimensional vector, using address a i Contains k i words, then the text semantic representation vector S i These word vectors can be aggregated and calculated using the following formula:

[0023]

[0024] Among them, ω i,l The electricity address a i The word vector of the first word in ;

[0025] S320: Use the hourly electricity consumption sequence within an average week to represent the hourly electricity consumption, sort out the hourly electricity consumption within the time period, and form a time series vector; calculate the electricity address a i The vector representing the hourly electricity consumption sequence during an average week is T i ;

[0026] A week has h = 168 hours, and the time interval [t1, t2] contains q complete weeks. The calculation formula is:

[0027]

[0028] S330: Text semantic representation vector S i and the power consumption sequence representation vector T i Splice and get the power address a i The joint representation vector X i , the calculation formula is:

[0029] X i =[S i ;T i ];

[0030] Among them, “[;]” represents the vector concatenation operation.

[0031] Preferably, the step S400 specifically includes the following process:

[0032] S410: Using the joint representation vector as the input of the model and the label of the power consumption functional area as the output; training the model with a large amount of labeled data and optimizing the model parameters so that the model can learn the mapping relationship between the joint representation of the power consumption address and the power consumption functional area type;

[0033] The labeled data set is D = {(X1,y1),(X2,y2),...,(X n ,yn )}, where n is the number of samples of labeled data, X i is the joint representation vector of the i-th sample, y i is the corresponding power consumption functional area label, y i ∈(y 城市 ,y 工业 ,y 商业 ,y 农业 );

[0034] S420: A deep learning model is used for training. The loss function L of the model may adopt a cross entropy loss function, and the calculation formula is:

[0035]

[0036] Among them, y i,c Label y for sample i i One-hot encoded value of category c, p i,c The model predicts the probability that sample i belongs to category c;

[0037] S430: Continuously update the model parameters θ through the back-propagation algorithm and optimizer to minimize the loss function L, that is:

[0038]

[0039] Among them, α is the learning rate, is the gradient of the loss function L with respect to the model parameters θ.

[0040] In a second aspect, the present invention provides a device for finely identifying power usage functional areas by coupling power usage addresses and power quantities, comprising:

[0041] The data acquisition module is used to obtain the electricity consumption addresses and hourly electricity consumption of the addresses in the specified area, establish data connections with external data sources, and realize real-time data collection and updating;

[0042] The joint representation construction module combines the textual semantic representation of the electricity usage address with the hourly sequence representation of the average electricity usage over a week to construct a joint representation of the electricity usage address. This module integrates algorithms and tools related to natural language processing and time series analysis.

[0043] The model training module uses labeled data to train a deep model for fine-grained identification of power consumption functional areas. It is responsible for selecting the deep learning model architecture, setting model parameters, and performing model training and optimization operations.

[0044] The functional area discrimination module discriminates the type of power consumption functional area based on the average prediction results of the deep model, performs statistics and analysis on the prediction results of the model, and outputs the final recognition results.

[0045] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the processor executes a method for finely identifying power usage functional areas by coupling power usage addresses and power quantities as described in claims 1-3.

[0046] In a fourth aspect, the present invention provides a computer system comprising:

[0047] A processor, configured to execute instructions stored in a storage medium to implement the method for finely identifying power usage functional areas by coupling power usage addresses and power quantities as described in claims 1-3;

[0048] A memory for storing a data set and intermediate results generated during a calculation process for a method for finely identifying power consumption functional areas by coupling power consumption addresses and power quantities;

[0049] The input / output interface is used to receive the specified text or recognition information and output the recognition results.

[0050] The present invention combines electricity usage address information with electricity consumption data to construct a joint representation, and uses a deep learning model for training and prediction to improve the accuracy and precision of electricity usage functional area identification, providing a reliable basis for the rational allocation and optimized management of power resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 The present invention is an overall flow chart of a method for finely identifying power consumption functional areas by coupling power consumption addresses and power consumption.

[0052] Figure 2 An example diagram for vector calculation. DETAILED DESCRIPTION

[0053] The specific embodiments of the present invention are described below to facilitate understanding of the present invention by those skilled in the art. However, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the appended claims, these changes are obvious, and all inventions and creations utilizing the concepts of the present invention are protected.

[0054] Taking "a method for finely identifying electricity-using functional areas in a small commercial area" as an example, the specific implementation process of the present invention is described in detail.

[0055] S100: Acquire standardized text content.

[0056] Select a small commercial area, where there are 10 electricity addresses, and the electricity address set A = {a1, a2, ..., a 10}; Get the hourly electricity consumption sequence of these 10 electricity consumption addresses within a total of 44 hours from January 1, 2024 to January 31, 2024; Taking electricity consumption address a1 as an example, its hourly electricity consumption sequence E1 = {e 1,1 ,e 1,2 ,...,e 1,744}, where e 1,j represents the power consumption of a1 in the jth hour.

[0057] S200: Standardize the electrical address.

[0058] The 10 electricity consumption addresses of electricity consumption set A are standardized. Taking electricity consumption address a1 as an example, "No. 12, Beijing West Road, Gulou District, Nanjing City, Jiangsu Province" is standardized to "No. 12, Beijing West, Gulou District, Nanjing City, Jiangsu Province".

[0059] S300: Constructing a joint representation; specifically including:

[0060] S310: Text semantic representation of the electricity address. The text information of the electricity address a1 is "No. 12, Beijing West, Nanjing, Jiangsu Province". The BERT model is used to convert it into a text semantic representation vector. The word vector dimension d1 = 768 output by the BERT model is obtained after word segmentation. K1 = 6 words are obtained. After average pooling, the following is calculated:

[0061]

[0062] The text semantic representation vector S1 is obtained, and its dimension is 768;

[0063] S320: Hourly electricity consumption sequence representation within an average week; the 31 days in January include q = 4 complete weeks, 3 days remainder, and a week has h = 168 hours; calculate the hourly electricity consumption sequence representation vector T1 within the average week of a1:

[0064]

[0065] The representation vector T1 is obtained, whose dimension is 168;

[0066] S330: Joint representation vector: Concatenate S1 and T1 to obtain the joint representation vector X1 of a1:

[0067] X1=[S1;T1];

[0068] At this time, the dimension of X1 is 768+168=936;

[0069] Similarly, the joint representation vectors X2, X3, ..., X of the other 9 power addresses can be calculated. 10 .

[0070] S400: Training a deep model; specifically including:

[0071] S410: Collect a dataset D containing 1000 labeled samples.

[0072] D={(X'1,y'1),(X'2,y'2),...,(X' 1000 ,y' 1000 )}, where X' i is the joint representation vector of the sample, y' i ∈{y 城市 ,y 工业 ,y 商业 ,y 农业} is the corresponding power consumption functional area label;

[0073] S420: LSTM neural network is selected for training. The model input dimension is 936, the hidden layer dimension is set to 128, and the output layer dimension is 4, corresponding to the four types of power consumption functional areas; the loss function uses the cross entropy loss function:

[0074]

[0075] Among them, y' i,c Label y' for sample i i One-hot encoded value of category c, p' i,c The model predicts the probability that sample i belongs to category c;

[0076] S430: Using the Adam optimizer with a learning rate of α = 0.001, the model parameters θ are updated through the back-propagation algorithm: After 50 rounds of iterative training, the model achieved good recognition performance.

[0077] S500: Determine the type of functional area; specifically including:

[0078] S510: Jointly represent the 10 electricity addresses of the small business district with vectors X1, X3, ..., X 10 Input into the trained LSTM model to obtain the predicted probability distribution of each electricity consumption address; taking a1 as an example, its predicted probability distribution P1 = {p 1,城市 ,p 1,工业 ,p 1,商业 ,p 1,农业};

[0079] S520: Calculate the average predicted probability of each power consumption functional area type in the region, taking urban power consumption as an example:

[0080]

[0081] The same logic applies The final calculation result is

[0082] S530: Because If the maximum value is 0, the small commercial area is determined to be a commercial electricity functional area.

[0083] The above complete example calculation process demonstrates the application and effectiveness of the method of the present invention in actual scenarios.

Claims

1. A method for finely identifying power consumption functional areas by coupling power consumption addresses and power consumption, characterized in that: The following steps are involved: S100: Data acquisition; there are n electricity addresses in the specified area, and the electricity address set is represented as A = {a1, a2, ..., a n }, for each power address a i (i=1,2,...,n), and the hourly electricity consumption sequence in the time interval [t1,t2] is expressed as E i ={e i,1 ,e i,2 ,...,e i,m }, where m is the total number of hours in the time interval, e i,j Indicates the power address a i Electricity consumption at hour j; S200: Multi-source data standardization processing: standardize the multi-source data and standardize the electricity consumption addresses of the electricity consumption set A. The standardized conversion formula is: is the converted address set, is the power-based GIS database conversion function, δ represents the format agreement function based on regular expressions; S300: Construct joint representation; The joint representation of the electricity usage address is constructed by splicing, including the text semantic representation of the electricity usage address and the hourly electricity usage sequence representation within an average week. The text semantic representation vector and the electricity usage sequence representation vector are spliced ​​to obtain the joint representation vector X of the electricity usage address. i ; S400: training deep model; Using labeled data containing urban, industrial, commercial, and agricultural electricity labels, a deep model for fine-grained identification of electricity functional areas is trained; S500: Determine the type of functional area; specifically, the process includes the following: S510: The area to be identified contains n electricity addresses, and their joint representation vectors X1, X2, ..., X n Input into the deep model trained in S400 to obtain the predicted probability distribution of each electricity consumption address: P i ={p i,城市 ,p i,工业 ,p i,商业 ,p i,农业 }(i=1,2,...,n); S520: Calculate the average predicted probability of each power consumption functional area type in the region. Taking urban power consumption as an example, the calculation formula is: The same logic applies S530: The category with the largest average predicted probability is used as the electricity consumption functional area type of the area. The calculation formula is: Where V is the output ribbon type.

2. According to claim 1, a method for finely identifying power usage functional areas by coupling power usage addresses and power consumption, characterized in that: The S300 specifically includes the following processes: S310: The word vector model maps each word into a d1-dimensional vector, using address a i Contains k i words, then the text semantic representation vector S i These word vectors can be aggregated and calculated using the following formula: Among them, ω i,l The electricity address a i The word vector of the first word in ; S320: Use the hourly electricity consumption sequence within an average week to represent the hourly electricity consumption, sort out the hourly electricity consumption within the time period, and form a time series vector; calculate the electricity address a i The vector representing the hourly electricity consumption sequence during an average week is T i ; A week has h = 168 hours, and the time interval [t1, t2] contains q complete weeks. The calculation formula is: S330: Text semantic representation vector S i and the power consumption sequence representation vector T i Splice and get the power address a i The joint representation vector X i , the calculation formula is: X i =[S i ;T i ]; Among them, "[;]" represents the vector concatenation operation.

3. The method for finely identifying power usage functional areas by coupling power usage addresses and power consumption according to claim 1, characterized in that: The S400 specifically includes the following processes: S410: Take the joint representation vector as the input of the model and the label of the power function area as the output; the labeled data set is D = {(X1, y1), (X2, y2), ..., (X n ,y n )}, where n is the number of samples of labeled data, X i is the joint representation vector of the i-th sample, y i is the corresponding power consumption functional area label, y i ∈(y 城市 ,y 工业 ,y 商业 ,y 农业 ); S420: A deep learning model is used for training. The loss function L of the model may adopt a cross entropy loss function, and the calculation formula is: Among them, y i,c Label y for sample i i One-hot encoded value of category c, p i,c The model predicts the probability that sample i belongs to category c; S430: Continuously update the model parameters θ through the back-propagation algorithm and optimizer to minimize the loss function L, that is: Among them, α is the learning rate, is the gradient of the loss function L with respect to the model parameters θ.

4. A device for finely identifying power usage functional areas by coupling power usage addresses and power consumption, characterized in that: include: The data acquisition module is used to obtain the electricity consumption addresses and hourly electricity consumption of the addresses in the specified area, establish data connections with external data sources, and realize real-time data collection and updating; The joint representation construction module combines the textual semantic representation of the electricity usage address with the hourly sequence representation of the average electricity usage over a week to construct a joint representation of the electricity usage address. This module integrates algorithms and tools related to natural language processing and time series analysis. The model training module uses labeled data to train a deep model for fine-grained identification of power consumption functional areas. It is responsible for selecting the deep learning model architecture, setting model parameters, and performing model training and optimization operations. The functional area discrimination module discriminates the type of power consumption functional area based on the average prediction results of the deep model, performs statistics and analysis on the prediction results of the model, and outputs the final recognition results.

5. A computer-readable storage medium, characterized in that A computer program is stored thereon, and when the program is executed by a processor, the processor executes a method for finely identifying power usage functional areas by coupling power usage addresses and power quantities as described in claims 1-3.

6. A computer system, characterized in that: include: A processor, configured to execute instructions stored in a storage medium to implement the method for finely identifying power usage functional areas by coupling power usage addresses and power quantities as described in claims 1-3; A memory for storing a data set and intermediate results generated during a calculation process for a method for finely identifying power consumption functional areas by coupling power consumption addresses and power quantities; The input / output interface is used to receive the specified text or recognition information and output the recognition results.

Images