CN114202112A

CN114202112A - Combination prediction method of service work order scale for power supply business hall considering regional mode

Info

Publication number: CN114202112A
Application number: CN202111428605.7A
Authority: CN
Inventors: 黄莉; 喻婧; 江薇; 林少红; 韩雅儒; 郑晨虹; 张诗鸣; 邹美华
Original assignee: State Grid Fujian Electric Power Co Ltd; Economic and Technological Research Institute of State Grid Fujian Electric Power Co Ltd
Current assignee: State Grid Fujian Electric Power Co Ltd; Economic and Technological Research Institute of State Grid Fujian Electric Power Co Ltd
Priority date: 2021-11-29
Filing date: 2021-11-29
Publication date: 2022-03-18
Anticipated expiration: 2041-11-29
Also published as: CN114202112B

Abstract

The invention relates to a power supply business hall service work order scale combined prediction method considering region modes. The method comprises the following steps: the service work order types are divided from two dimensions by combining the characteristics of the service work orders of the power supply business hall and considering the region modes, wherein the service work order types comprise work order service object types and work order emergency degree types; constructing the combination of the influence factors of the work order number of the power supply service business hall; collecting historical data, constructing a work order quantity prediction model by adopting a neural network model according to the classification of work order service objects, constructing the work order quantity prediction model by adopting a support vector machine method according to the type of emergency degree of the work order, and performing simulation training on the model by combining actual work order data as an input variable; and calculating the weights of the two models by using a CRITIC method and on the basis of the prediction result deviations of different prediction models, and finally realizing the construction of a combined prediction model. The invention can improve the prediction accuracy and provide support and reference for the planning and arrangement of the subsequent business service hall.

Description

Regional mode-considered power supply business hall service work order scale combined prediction method

Technical Field

The invention relates to the field of power supply business hall service work order scale prediction and the like, mainly acts on the power supply business hall service work order scale prediction, improves the service quality through accurate prediction, actively changes to an internet and power marketing service mode, provides business hall data reference for the transformation of marketing service from customer demand guidance to customer experience guidance, and provides a direction for business hall transformation construction, in particular to a power supply business hall service work order scale combined prediction method considering region modes.

Background

With the continuous emergence of new technologies such as the internet plus and the like, the demand of people on good life is increasing day by day, the short service mode of the traditional customer service in the power supply business hall is continuously obvious, and the demand and service of the power customer face unprecedented new challenges which are mainly reflected in the lack of an active service system for guiding the customer experience, the lack of service operation type operation capability, the lack of closed-loop and dynamic internal control means and mechanisms and the lack of service support capability of comprehensive business analysis. With the rise of new retail, the value of the entity channel gradually returns; younger people are more and more accustomed to mobile APP services, and the market segmentation and aggregation trend of customers is obvious. In the future, the power supply business hall needs to fully play the role of peripheral nerves, more accurately position customers, provide more comprehensive, comprehensive and professional services for the customers, fully develop various marketing activities such as new business promotion, new product sales, online drainage and the like, improve the profitability and become a place for creating the interactive value of the two parties. The invention combines different types of work order division and realizes the combined prediction of the work order quantity by different methods, thereby being capable of providing reference for subsequent improvement of service quality and planning arrangement.

Disclosure of Invention

The invention aims to provide a power supply business hall service work order scale combination prediction method considering region modes, which aims to solve the problems that the statistical analysis is not carried out on the number of work orders and the like in the prior art. The fine management level of the service work order of the power supply business hall is improved, more efficient, convenient and accurate high-quality service and brand new feeling of online and offline interaction combination are provided for customers, the customer satisfaction is comprehensively improved, the stickiness of the customers is increased, and the service market is further developed.

In order to achieve the purpose, the technical scheme of the invention is as follows: a combined prediction method for the service work order scale of a power supply business hall considering region modes comprises the following steps:

step S1, dividing service work order types from two dimensions by combining the characteristics of the service work orders of the power supply business hall and considering the region modes, wherein the service work order types comprise work order service object types and work order emergency degree types;

s2, constructing a power supply service business hall work order number influence factor combination, wherein the influence factor combination comprises the factors of population number, area, gdp, a three-product structure, power supply reliability of a regional power grid, voltage qualification rate of the regional power grid, asset scale of the regional power grid and average service life of the regional power grid;

step S3, collecting historical data, constructing a work order quantity prediction model by adopting a neural network model according to the type of a work order service object, constructing a work order quantity prediction model by adopting a support vector machine method according to the type of the emergency degree of the work order, and performing simulation training on the model by combining actual work order data as an input variable;

and step S4, calculating the weights of the two models by using a CRITIC method and on the basis of the prediction result deviations of different prediction models, and finally realizing the construction of a combined prediction model.

In an embodiment of the present invention, in step S1, the work order service object types are industry, general industry and business, residents, and agriculture, respectively; the emergency degree types of the work orders are emergency, emergency and general emergency respectively.

In an embodiment of the invention, in step S3, the neural network model is a BP neural network model.

In an embodiment of the present invention, in step S4, the concrete implementation manner of calculating the weights of the two models based on the prediction result deviations of different prediction models by using the CRITIC method is as follows:

supposing that n samples to be evaluated are provided, p evaluation indexes form an original index data matrix:

in order to eliminate the influence on the evaluation result caused by different dimensions, each index needs to be subjected to dimensionless treatment;

CRITIC weight methods generally use either forward or reverse processing;

forward or reverse processing:

as the index used is larger, the better (positive index:)

The smaller the value of the index used, the better (reverse index:)

Expressed in terms of standard deviation

Sj represents the standard deviation of the jth index;

the standard deviation is used in the CRITIC method to represent the difference fluctuation condition of the internal values of each index, the larger the standard deviation is, the larger the numerical difference of the index is, more information can be reflected, the stronger the evaluation intensity of the index is, and more weight should be assigned to the index;

expressed by correlation coefficient

r_ijRepresenting a correlation coefficient between the evaluation indexes i and j;

the correlation coefficient is used for representing the correlation between the indexes, the stronger the correlation with other indexes, the smaller the conflict between the index and other indexes, the more the same information is reflected, the more repeated evaluation contents can be embodied, the evaluation strength of the index is weakened, and the weight distributed to the index is reduced;

the larger the Cj is, the larger the effect of the jth evaluation index in the whole evaluation index system is, more weight should be distributed to the jth evaluation index;

so the objective weight Wj of the jth index is

Compared with the prior art, the invention has the following beneficial effects: the method scientifically and reasonably divides the work order types on the basis that the system comprehensively investigates the current situation of the service work orders of the power supply business hall, and respectively constructs an intelligent prediction method by combining the division dimensions, respectively realizes scientific prediction of the work order scale, and realizes weight determination of a prediction model by applying a scientific objective weighting method and combining the prediction precision of different prediction models.

Drawings

Fig. 1 is a flowchart of a power supply business hall service work order scale combination prediction method considering region modalities according to the present invention.

FIG. 2 is a topology diagram of the BP neural network of the present invention.

Fig. 3 is an optimal hyperplane view of the SVM of the present invention.

FIG. 4 is a diagram of an intermediate node of an SVM classification function output according to the present invention.

Detailed Description

The technical scheme of the invention is specifically explained below with reference to the accompanying drawings.

The invention relates to a power supply business hall service work order scale combined prediction method considering region modes, which comprises the following steps:

The following is a specific implementation process of the present invention.

And S1, dividing service work order types from two dimensions by combining the characteristics of the service work orders of the power supply business hall and considering the region modes. Including work order service object types, respectively large industry, general industry and commerce, residents, agriculture, and the like. The emergency degree types of the work orders are emergency, emergency and general emergency respectively.

And S2, establishing influence factor combination of the work order number of the power supply service business hall, wherein the influence factor combination comprises the factors of population number, area, gdp, a three-product structure, power supply reliability of a regional power grid, voltage qualification rate of the regional power grid, asset scale of the regional power grid, average service life of the regional power grid asset and the like.

S3, collecting historical data, constructing a work order quantity prediction model by adopting a neural network model according to the classification of work order service objects, constructing the work order quantity prediction model by adopting a support vector machine method according to the type of emergency degree of the work order, and performing simulation training on the model by combining actual work order data as an input variable;

in step S3, the topological graph of the constructed BP neural network is shown in fig. 2.

(1) Forward propagation process

The input layer has n neurons x ∈ Rn, x ═ x [ (x) ]₁，x₂，…，x_n) (ii) a The hidden layer has d neurons h ∈ Rd, h ═ h [ (-h ∈ Rd)₁，h₂，…，h_d) (ii) a The output layer has m neurons y ∈ Rm, and F ═₁，F₂，…，F_m) The weight and threshold between the input layer and the hidden layer are respectively W_ijAnd b_jThe weight and the threshold between the hidden layer and the output layer are respectively W_jkAnd b_k；

Hidden layer node:

output layer node:

in the formula, transfer function

The error at the output node for one sample is:

in the formula, t_k、F_kThe expected output and the calculated output of the network, respectively; error of all samples

Wherein p is the number of samples;

(2) process of back propagation

Between the output layer node and the hidden layer node, the error calculation formula is as follows:

δ_k＝(t_k-F_k)F_k(1-F_k)

weight correction:

threshold value correction:

in the above formula, n₀Is the iteration number; between the input layer node and the hidden layer node, the error calculation formula is as follows:

weight correction:

threshold value correction:

constructing a prediction model of a support vector machine,

the support vector machine is originally used for solving the classification problem, the prediction technology is the development on the basis of the classification problem, and the classification theory and the application popularization of the prediction technology are mainly described below. Let the training sample set be { (x)_i,y_i) I ═ 1,2, ·, l }, where x is_i∈RⁿIs the ith example of the input pattern, y_iE { +1, -1} is a target output corresponding to the two classes of classification problems; the support vector classifier aims to construct a decision function, namely a classification hyperplane, and classify test data as correctly as possible;

if there is a classification hyperplane:

<ω,x>+b＝0

and the constraint conditions are met:

y_i(<ω,x>+b)-1≥0,i＝1,2,···,l

then the training set is said to be linearly separable, where<·,·>Denotes the inner product of the vector, ω ∈ RⁿIs an adjustable weight vector, and b belongs to R as a bias;

from the theory of statistical learning, if the training sample set is not separated by the hyperplane error and the distance between the sample data closest to the hyperplane and the classification hyperplane is the largest, the hyperplane is the optimal hyperplane (OptimalHyperplane), as shown in fig. 3, and the decision function obtained thereby:

f(x)＝sgn(<ω,x>+b)

wherein sgn (·) is a sign function;

by utilizing a Lagrange optimization method and a Wofle dual theory, two classification problems can be converted into dual problems, namely a maximized functional:

wherein alpha is_iLagrange multipliers corresponding to the samples i;

the above formula is a quadratic programming problem with inequality constraints, and has a unique solution, and only a small part of alpha exists in the solution_iThe value is not zero, the corresponding sample is the support vector, and the optimal classification function obtained by solving the problems is as follows:

wherein nsv is the number of support vectors, b is a classification threshold, and can be obtained by taking the median value from any pair of support vectors in two classes;

cortex in 1995&Vapnild is the relaxation term xi introduced by_iAnd (3) realizing a generalized classification surface, and solving the linear inseparable condition of the training samples, namely compromising the minimum misclassified sample and the maximum classification interval. For non-linear problems, it can be converted into a linear problem in some high-dimensional space by a non-linear transformation, and then an optimal classification surface is sought in this high-dimensional space. Only inner product operation (x) between samples_i·x_j) Is involved, so only the inner product operation needs to be performed in the high dimensional space, and the inner product operation can be realized by the function in the original space. According to the Hilbert-schmidt principle, as long as the kernel function K (x)_i·x_j) If the Mercer condition is satisfied, it corresponds to the inner product in a certain swap space.

Thus, with a kernel function K (x) satisfying the Mercer condition_i·x_j) Instead of the inner product in the formula, some kind of linear classification after nonlinear transformation can be realized, and then the optimal classification function becomes:

in summary, SVM maps the input vector to a high-dimensional feature space through some kind of pre-selected non-linear mapping, and constructs an optimal classification hyperplane in this feature space. In form, the SVM classification function is similar to a neural network, with the output being a linear combination of intermediate nodes, one for each support vector, as shown in FIG. 4. The number of intermediate nodes of the neural network is selected by experience or contrast experiments, and the difference of logics can greatly influence the network performance; and the number of intermediate nodes of the SVM is automatically determined by calculation.

By introducing an epsilon insensitive loss function, Vapnik generalizes the result obtained in the SVM classification theory, so that the result can be used for function fitting. Introducing a relaxation variable xi_iAnd

the following optimization problem is constructed:

wherein the constant C is a penalty coefficient;

the dual space optimization problem is as follows:

wherein alpha is_i,

Is a Lagrange multiplier;

solving the problem to obtain the optimal Lagrange multiplier alpha_i,

Thus, a fitting function is obtained:

The CRITIC weight method is an objective weighting method. The idea is to use two indexes, namely contrast strength and conflict index. The contrast intensity is expressed by standard deviation, and if the standard deviation of the data is larger, the fluctuation is larger, and the weight is higher; the conflict is expressed by using a correlation coefficient, and if the value of the correlation coefficient between the indexes is larger, the conflict is smaller, and the weight is lower. And during weight calculation, multiplying the contrast intensity by the conflict index, and performing normalization processing to obtain the final weight. When the SPSSAU is used, the contrast strength and the conflict index are automatically established, and the weight value is calculated.

The CRITIC weighting method is suitable for data, namely the stability of the data can be regarded as information, and certain incidence relation exists among analyzed indexes or factors, so that the CRITIC weighting method just uses the volatility (contrast strength) and the relevance (conflict) of the data to carry out weight calculation.

CRITIC weight methods generally use either forward or reverse processing; the normalization process is not recommended because if it is used, the standard deviations all become the number 1, i.e., the standard deviations of all indexes are completely identical, which results in no significance for the volatility index.

Forward or reverse processing:

as the index used is larger, the better (positive index:)

The smaller the value of the index used, the better (reverse index:)

Expressed in terms of standard deviation

Sj represents the standard deviation of the jth index;

expressed by correlation coefficient

r_ijIndicates the evaluation indexes i andthe correlation coefficient between j;

the correlation coefficient is used for representing the correlation among the indexes, the stronger the correlation with other indexes, the smaller the conflict between the index and other indexes, the more the same information is reflected, the more repeated evaluation contents can be embodied, the evaluation strength of the index is weakened to a certain extent, and the weight distributed to the index is reduced;

so the objective weight Wj of the jth index is

Those skilled in the art will appreciate that all or part of the flow of the method implementing the above embodiments may be implemented by a computer program, which is stored in a computer readable storage medium, to instruct related hardware. The computer readable storage medium is a magnetic disk, an optical disk, a read-only memory or a random access memory.

It is noted that relational terms such as "first" and "second," and the like, may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that an article or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.

The above description is merely exemplary of the present application and is presented to enable those skilled in the art to understand and practice the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

It will be understood that the present application is not limited to what has been described above and shown in the accompanying drawings, and that various modifications and changes can be made without departing from the scope thereof. The scope of the application is limited only by the appended claims.

The above are preferred embodiments of the present invention, and all changes made according to the technical scheme of the present invention that produce functional effects do not exceed the scope of the technical scheme of the present invention belong to the protection scope of the present invention.

Claims

1. a power supply business hall service work order scale combination prediction method considering regional modal, is characterized in that, comprises the steps:

Step S1, combining the characteristics of the service work order of the power supply business hall and considering the regional mode, the service work order types are divided from two dimensions, including the type of work order service object and the type of work order urgency;

Step S2, construct a combination of factors affecting the number of work orders in the power supply service business hall, including population, area, gdp, tertiary industry structure, regional power supply reliability, regional power grid voltage qualification rate, regional power grid asset scale, and regional power grid asset average. factors of age;

Step S3: Collect historical data, use a neural network model to build a work order quantity prediction model according to the type of work order service object, use a support vector machine method to build a work order quantity prediction model for the type of work order urgency, and use the actual work order data as input variables, and simulate the training of the model;

In step S4, the CRITIC method is used to calculate the weights of the two models based on the deviation of the prediction results of the different prediction models, and finally the construction of the combined prediction model is realized.

2. The method for predicting the scale of service work orders for power supply business halls considering regional modalities according to claim 1, wherein in step S1, the types of work order service objects are respectively large industry, general industry and commerce, residents, and agriculture. ; Work order urgency type, which are very urgent, urgent, and general urgency.

3 . The method for predicting the scale of service work orders for power supply business halls considering regional modalities according to claim 1 , wherein in step S3 , the neural network model is a BP neural network model. 4 .

4. The power supply business hall service work order scale combination prediction method considering regional modal according to claim 1, is characterized in that, in step S4, using the CRITIC method, based on the deviation of the prediction results of different prediction models, calculate the two. The specific implementation of the weights of each model is as follows:

Suppose there are n samples to be evaluated and p evaluation indicators to form the original indicator data matrix:

In order to eliminate the influence of different dimensions on the evaluation results, it is necessary to perform dimensionless processing on each index;

The CRITIC weight method generally uses forward or reverse processing;

Forward or reverse processing:

The larger the value of the indicator used, the better (positive indicator:)

The smaller the value of the indicator used, the better (inverse indicator:)

Expressed in the form of standard deviation

Sj represents the standard deviation of the jth indicator;

In the CRITIC method, the standard deviation is used to represent the difference and fluctuation of the internal values of each index. The larger the standard deviation, the greater the numerical difference of the index, the more information can be displayed, and the evaluation strength of the index itself is also The stronger it is, the more weight should be assigned to the indicator;

represented by the correlation coefficient

r _ij represents the correlation coefficient between evaluation indicators i and j;

The correlation coefficient is used to represent the correlation between indicators. The stronger the correlation with other indicators, the less conflict between the indicator and other indicators. The more the same information is reflected, the more the evaluation content can be reflected. The repetition will weaken the evaluation strength of the indicator, and the weight assigned to the indicator should be reduced;

The larger the Cj, the greater the role of the jth evaluation index in the entire evaluation index system, and more weights should be assigned to it;

So the objective weight Wj of the jth indicator is