A method of power customer segmentation based on hybrid neural network

Abstract

The present invention proposes a method for segmenting power customers based on a hybrid neural network, which uses a hybrid neural network method to segment power customers, and classifies power customers by their characteristic indicators and the connection between time. In this way, targeted marketing strategies can be formulated to meet the power needs of different customers, achieve personalized and differentiated services, improve the core competitiveness of power grid companies, and expand the share of electric energy in social consumption terminals. This has become an urgent task for power companies. The power customer segmentation method proposed by the present invention helps to improve the accuracy of the power customer segmentation method.

Description

Technical Field

[0001] The invention relates to the research field of power customer data object segmentation, and in particular to a power customer segmentation method based on a hybrid neural network. Background Art

[0002] With the gradual deepening of my country's power market reform, the diversified characteristics of power market demand have gradually become prominent. Formulating targeted marketing strategies to meet the electricity needs of different customers, realizing personalized and differentiated services, improving the core competitiveness of power grid companies, and expanding the share of electricity in social consumer terminals have become an urgent task for power companies. In order to formulate electricity services for power customers, power customer segmentation is particularly important. The present invention proposes a hybrid neural network method of H-LSTM (Hybrid-Long Short-Term Memory) to perform power customer segmentation, and classifies power customers according to the characteristic indicators of power customers and the connection between time. Summary of the invention

[0003] In view of this, the purpose of the first aspect of the present invention is to provide a power customer segmentation method based on a hybrid neural network to meet the electricity needs of different customers and achieve personalized and differentiated services.

[0004] The purpose of the first aspect of the present invention is achieved through the following technical solutions:

[0005] The electric power customer segmentation method based on hybrid neural network includes the following steps:

[0006] Improved H-LSTM model based on power user evaluation indicators;

[0007] Extract electricity consumption characteristic indicators from the original electricity consumption data;

[0008] The time correlation of feature indicators is used for analysis, and a hybrid LSTM neural network is set up to solve the problem based on the H-LSTM model.

[0009] Finally, the result value is obtained and customers are classified according to the result value.

[0010] Furthermore, the steps for constructing the improved H-LSTM model based on the power user evaluation index are as follows:

[0011] Reconstruct a subsequence data set of n evaluation indicators based on the power consumption data;

[0012] Calculate the mean and variance of power consumption and electricity cost for 12 months each year;

[0013] The reconstructed subsequence data set and the mean and variance of power consumption and electricity cost are used as the input of the LSTM neural network, that is, the input dimension is 12 dimensions;

[0014] The dimension of the output layer is 1, and the loss function is the L2 error function;

[0015] The activation function is ReLU, and batch gradient descent is used for learning. ReLU(x)=max(0,x) is the input and output of the H-LSTM neural network.

[0016] Furthermore, the n is 8.

[0017] Furthermore, the framework of the H-LSTM power customer segmentation network, the attribute-based classification method is based on a local sliding window, assuming that the label of an attribute greatly depends on the attributes of its adjacent positions, given an attribute sequence c of length n (1:n) , a window of size k is drawn from the first attribute c of the attribute sequence (1) Slide to the last attribute c (n) , for each word c in the sequence (t) , when the window size is 5, the attribute c (t-2) , c (t-1) , c (t) , c (t+1) , c (t+2) will be sent to the lookup table; then, the attribute vectors extracted from the lookup table are concatenated into a vector X (t) .

[0018] Furthermore, when the range of an attribute exceeds the sequence boundary, it will be supplemented with a special marker.

[0019] Further, in the next layer of the neural network, X (t) After linear transformation, the Sigmoid function σ(x) = (1+E -x ) -1 Or tanh function activation.

[0020] Furthermore, the H-LSTM neural network stores information in gated cells outside the normal information flow of the recurrent network.

[0021] The second aspect of the present invention aims to provide a computer system for electricity customer segmentation based on a hybrid neural network, comprising a memory, a processor and a computer program stored in the memory and capable of running on the processor, wherein the processor implements the method described above when executing the computer program.

[0022] The third aspect of the present invention aims to provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the above-mentioned method when executed by a processor.

[0023] The beneficial effects of the present invention are:

[0024] The present invention proposes a method and system for segmenting electricity customers based on a hybrid neural network, which can improve the accuracy of electricity customer segmentation. The present invention uses a hybrid neural network method to segment electricity customers, and classifies electricity customers according to the characteristic indicators of electricity customers and the connection between time, so as to formulate targeted plans to meet the electricity needs of different customers, realize personalized and differentiated services, improve the core competitiveness of power grid companies, and expand the share of electricity in social consumption terminals.

[0025] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description and the preceding claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings, in which:

[0027] Figure 1 It is a schematic diagram of the process of the present invention;

[0028] Figure 2 This is the structural diagram of the H-LSTM neural network unit;

[0029] Figure 3 Schematic diagram of the input and output of the H-LSTM neural network;

[0030] Figure 4 It is a schematic diagram of the power segmentation framework based on H-LSTM neural network;

[0031] Figure 5 This is the structure diagram of the improved H-LSTM neural network. DETAILED DESCRIPTION

[0032] The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the preferred embodiments are only for illustrating the present invention, rather than for limiting the protection scope of the present invention.

[0033] As shown in the figure, a method for segmenting electric power customers based on a hybrid neural network of the present invention comprises the following steps:

[0034] S1: Improved H-LSTM model based on power user evaluation indicators;

[0035] S2: extracting electricity consumption characteristic indicators from the original electricity consumption data;

[0036] S3: Analyze the temporal correlation of feature indicators and set a hybrid LSTM neural network to solve the problem based on the H-LSTM model;

[0037] S4: Finally, the result value is obtained and customers are classified according to the result value.

[0038] The present invention applies the long short-term memory of the improved H-LSTM neural network to customer segmentation.

[0039] LSTM network is an extension of RNN, specially designed to avoid long-term dependency problems. The repetitive neural network module of LSTM has a different structure, which is different from the naive RNN. There are four neural network layers that interact with each other in a special way. The key to LSTM network is the cell state, which is a bit like a conveyor belt. In LSTM, the gate structure is used to add or delete information to the cell state, and the gate structure is a way to selectively let information pass through, usually consisting of a Sigmoid neural network layer and a point-by-point product operation (the output of the Sigmoid layer is between 0 and 1, which defines the degree of information passing through. 0 means nothing is allowed to pass, and 1 means everything is allowed to pass). The H-LSTM network has three gate structures, namely input gates, forget gates, and output gates, to maintain and update the cell state. In the following formula, it, ft, ot, and Ct respectively represent the three gate structures and cell states corresponding to time t.

[0040] Forgetting information from the cell state is determined by the Sigmoid layer of the forget gate. The input of the current layer Xt and the output of the previous layer ht-1 are used as inputs. The cell state output at time t-1 is f t =σ(W f ·(h t-1 ,x t )+b f )

[0041] The information stored in the cell state mainly consists of two parts: (1) the result of the Sigmoid layer of the input gate i t As The updated information; (2) the vector c newly created by the tanh layer t , will be added to the cell state. The old cell state c t-1 Multiply by f t , used to forget information and new candidate information The sum of , generates the update of the cell state:

[0042] The output information is determined by the output gate. First, the Sigmoid layer is used to determine the part of the cell state information to be output, and then the tanh layer is used to process the cell state. The product of the two parts of information is the output value: t =σ(W i ·(h t-1 ,x t )+b i )

[0043] In this embodiment, the improved H-LSTM model based on the power user evaluation index is constructed as follows:

[0044] (1) Reconstruct a subsequence data set of eight evaluation indicators based on power consumption data;

[0045] (2) Calculate the mean and variance of power consumption and electricity costs for 12 months each year

[0046] (3) The reconstructed subsequence data set and the mean and variance of power consumption and electricity charges are used as the input of the LSTM neural network, that is, the input dimension is 12 dimensions.

[0047] (4) The dimension of the output layer is 1, and the loss function is the L2 error function

[0048] (5) The activation function is ReLU, and batch gradient descent is used for learning. ReLU(x) = max(0,x) is the input and output of the H-LSTM neural network.

[0049] The framework of the H-LSTM power customer segmentation network, the attribute-based classification method is based on a local sliding window, assuming that the label of an attribute greatly depends on the attributes in its adjacent position. Given an attribute sequence c(1: n) of length n, a window of size k slides from the first attribute c(1) of the attribute sequence to the last attribute c(n). For each word c(t) in the sequence, when the window size is 5, the attributes c(t-2), c(t-1), c(t), c(t+ 1), c(t+2)) will be sent to the lookup table. When the range of the attribute exceeds the sequence boundary, it will be supplemented with special tags such as "start" and "end"; then, the attribute vectors extracted from the lookup table are concatenated into a vector X(t). In the next layer of the neural network, X(t) is linearly transformed and activated by the Sigmoid function σ(x)= (1+Ex)-1 or the tanh function.

[0050] Given a set of annotations, a similar linear transformation will be performed, except that there is no nonlinear function, and the resulting y(t) is a score vector for each possible label. y(t) = W2 h(t) + b2

[0051] H-LSTM neural networks store information in gated cells outside the normal information flow of the recurrent network. These cells can store, write or read information. The cells decide which information to store and when to allow it to be read, written or cleared by opening and closing the gates. But unlike the digital memory in a computer, these gates are analog, consisting of element-by-element multiplications of sigmoid functions whose outputs all range from 0 to 1. The advantage of analog values ​​over digital storage is that they are differentiable and therefore suitable for back propagation. Therefore, it is possible to decide which values ​​of power user attributes to remember and which to forget. These gates open and close based on the signals they receive, and similar to the nodes of a neural network, they use their own set of weights to filter information and decide whether to allow it to pass based on its strength and the content of the input. These weights, like the weights that modulate the input and hidden states, are adjusted through the learning process of the recurrent network. That is, the memory cells learn when to allow data to enter, leave or be deleted through an iterative process of guessing, backpropagating errors, and adjusting weights using gradient descent.

[0052] The memory gate and forget gate of the improved H-LSTM neural network are used to realize the contextual association between attributes by preprocessing the original data.

[0053] Table 1: Training data examples

[0054] Ar A Dp Pc Ec Ac Gr Dc Pc-m Pc-v Ec-m Ec-v 0.8 1.2 0.2 100.23 1233 30 0.12 16 120.23 12.39 193.43 23.78 … … … … … … … … … … … …

[0055] In the power user classification task, the input of the H-LSTM memory unit comes from the attribute values ​​of the upper and lower attribute windows. For each attribute value c(t), the input of the H-LSTM memory unit is X(t), which is formed by connecting the upper and lower attribute value embeddings (c(tk), ..., c(t), ..., c(t+k)), where k represents the distance from the current attribute. The output of the H-LSTM unit is used for the label inference function after linear transformation to infer the label corresponding to the attribute, and finally the output of the mixed H-LSTM neural network is XORed.

[0056] It should be appreciated that embodiments of the present invention may be implemented or enforced by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer-readable memory. The method may be implemented in a computer program using standard programming techniques - including a non-transitory computer-readable storage medium configured with a computer program, wherein the storage medium so configured causes the computer to operate in a specific and predefined manner - according to the methods and drawings described in the specific embodiments. Each program may be implemented in a high-level procedural or object-oriented programming language to communicate with a computer system. However, if desired, the program may be implemented in an assembly or machine language. In any case, the language may be a compiled or interpreted language. In addition, the program may be run on a programmed ASIC for this purpose.

[0057] Furthermore, the operations of the processes described herein may be performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by context. The processes described herein (or variations and / or combinations thereof) may be performed under the control of one or more computer systems configured with executable instructions, and may be implemented as code (e.g., executable instructions, one or more computer programs, or one or more applications) that is executed collectively on one or more processors, by hardware, or a combination thereof. The computer program includes a plurality of instructions that may be executed by one or more processors.

[0058] Further, the method can be implemented in any type of computing platform that is operably connected to a suitable computer, including but not limited to a personal computer, a minicomputer, a mainframe, a workstation, a network or distributed computing environment, a separate or integrated computer platform, or in communication with a charged particle tool or other imaging device, etc. Various aspects of the present invention can be implemented in machine-readable code stored on a non-transitory storage medium or device, whether removable or integrated into a computing platform, such as a hard disk, an optical read and / or write storage medium, a RAM, a ROM, etc., so that it can be read by a programmable computer, and when the storage medium or device is read by the computer, it can be used to configure and operate the computer to perform the process described herein. In addition, the machine-readable code, or portions thereof, can be transmitted via a wired or wireless network. When such media includes instructions or programs that implement the steps described above in conjunction with a microprocessor or other data processor, the invention described herein includes these and other different types of non-transitory computer-readable storage media. When programmed according to the methods and techniques of the present invention, the present invention also includes the computer itself.

[0059] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A method for segmenting power customers based on hybrid neural network, characterized by: The following steps are involved: Improved H-LSTM model based on power user evaluation indicators; Extract electricity consumption characteristic indicators from the original electricity consumption data; The time correlation of feature indicators is used for analysis, and a hybrid LSTM neural network is set up to solve the problem based on the H-LSTM model. Finally, the result value is obtained and the customers are classified according to the result value; The steps for constructing the improved H-LSTM model based on power user evaluation indicators are as follows: Reconstruct a subsequence data set of n evaluation indicators based on the power consumption data; Calculate the mean and variance of power consumption and electricity cost for 12 months each year; The reconstructed subsequence data set and the mean and variance of power consumption and electricity cost are used as the input of the LSTM neural network, that is, the input dimension is 12 dimensions; The dimension of the output layer is 1, and the loss function is the L2 error function; The activation function is ReLU, and batch gradient descent is used for learning. ReLU(x)=max(0,x) is the input and output of the H-LSTM neural network.

2. The method for segmenting power customers based on a hybrid neural network according to claim 1, characterized in that: The n is 8.

3. The method for segmenting electric power customers based on hybrid neural network according to claim 1, characterized in that: The framework of the H-LSTM power customer segmentation network, the attribute-based classification method is based on a local sliding window, assuming that the label of an attribute depends greatly on the attributes of its adjacent positions, given an attribute sequence c of length n ( 1: n) , a window of size k is drawn from the first attribute c of the attribute sequence (1) Slide to the last attribute c (n) , for each word c in the sequence ( t) , when the window size is 5, the attribute c ( t-2) , c ( t-1) , c ( t) , c ( t + 1) , c ( t + 2) will be sent to the lookup table; then, the attribute vectors extracted from the lookup table are concatenated into a vector X (t) .

4. The method for segmenting power customers based on a hybrid neural network according to claim 3 is characterized in that: When the range of an attribute exceeds the sequence boundary, it will be supplemented with special markers.

5. The method and system for electric power customer segmentation based on hybrid neural network according to claim 3 is characterized by: In the next layer of the neural network, X (t) After linear transformation, Sigmoid function σ(x) = (1+E -x ) -1 Or tanh function activation.

6. A method for segmenting power customers based on a hybrid neural network according to claim 3, 4 or 5, characterized in that: The H-LSTM neural network stores information in gated cells outside the normal information flow of the recurrent network.

7. A computer system for electric power customer segmentation based on a hybrid neural network, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 6 is implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images