CN112348259A

CN112348259A - Method for predicting purchase price of power equipment

Info

Publication number: CN112348259A
Application number: CN202011243022.2A
Authority: CN
Inventors: 周程; 高正平; 李阿勇; 丁一; 张盛; 尤伟
Original assignee: State Grid Jiangsu Electric Power Co Ltd
Current assignee: State Grid Jiangsu Electric Power Co Ltd
Priority date: 2020-11-09
Filing date: 2020-11-09
Publication date: 2021-02-09

Abstract

The invention relates to the field of big data, in particular to the field of prediction calculation by utilizing big data in the power industry, and more particularly relates to a method for predicting the purchase price of power equipment. The model prediction result trained by the method is closer to the actual price range, the prediction error is within the acceptable range, the model has higher reliability, and the prediction result has certain reference value for differentiated purchasing strategies such as calibration packet division, and the rating principle.

Description

Method for predicting purchase price of power equipment

Technical Field

The invention relates to the field of big data, in particular to the field of prediction calculation by utilizing big data in the power industry, and more particularly relates to a method for predicting the purchase price of power equipment.

Background

Purchasing as a third profit source of an enterprise has a crucial influence on the benefit growth of the enterprise. The reasonable purchase price can reduce unnecessary expenditure and waste for the enterprise, help the enterprise to carry out purchasing activities more efficiently, and realize the goals of cost reduction and efficiency improvement.

In the field, the analysis of big data is deepened, the resource value of mass data is fully mined, pain points, difficulty points and promotion points existing in material management services are analyzed and summarized, data is used for driving management change and transformation upgrading, and guidance on the quality of purchasing equipment and the promotion of purchasing supply timeliness is the key point and the hot point of research. The establishment of data thinking, the innovation of technical methods and the development of multi-dimensional big data research and analysis by combining with business requirements become the key for enhancing competitiveness of power enterprises.

At present, a price trend prediction model based on equipment material information price, a raw material price linkage model and an equipment purchase price analysis model based on internal and external price information are established by focusing on a purchase business symptom node of an electric power enterprise, but the price trend prediction model, the raw material price linkage model and the equipment purchase price analysis model are analyzed and researched from a single purchase price influence factor or an expansion research of a single business direction.

The purchase price of the equipment of the power enterprise is influenced by various factors, and has the characteristics of nonlinearity and non-stationarity, irregular distribution of data nodes, high difficulty in predicting the purchase price, low prediction precision and the like.

Therefore, according to the purchasing characteristics of the electric power equipment, a plurality of factors influencing the purchasing price of the electric power equipment are considered, and the method for accurately predicting the purchasing price interval is a problem to be solved by technical personnel in the field and is a hot research area for application of big data in the electric power industry.

Disclosure of Invention

The technical problem to be solved by the invention is to combine the purchasing price variation characteristics of the power equipment, analyze and comb the acquired and quantifiable equipment purchasing price influence factors, thereby seeking an algorithm model which is suitable for multi-factor analysis and has strong nonlinear conversion capability, analyzing and predicting a reasonable purchasing price interval of the equipment, effectively improving the accuracy of purchasing price prediction and providing a basis for making a differentiated bidding purchasing strategy.

In order to solve the technical problem, the invention discloses a method for predicting the purchase price of electric power equipment, which comprises the following steps:

firstly, constructing a data index system according to different equipment types; when the equipment is non-universal equipment, selecting historical bidding prices, raw material prices, labor costs, macroscopic economic indexes, bidding package quantity and bidder quantity to construct a data index system; when the equipment is general equipment, selecting historical bidding prices, raw material prices, labor costs, macroscopic economic indexes, bidding package quantity, bidder quantity and historical external purchase price to construct a data index system;

secondly, preprocessing data in the constructed data index system according to modeling requirements;

thirdly, determining the number of nodes of an output layer, selecting 'tender price benchmark max' and 'tender price benchmark min' as reasonable ranges of purchase prices, and setting the reasonable ranges as prediction targets, wherein the tender price benchmark is the average value of the rest tender prices with deviation exceeding 20% of tender price average values in a removed package, the tender price benchmark max is the maximum tender price after the removal deviation exceeds 20%, and the tender price benchmark min is the minimum tender price after the removal deviation exceeds 20%;

the fourth step, determine the number of nodes in hidden layer, wherein the common formula is used

(m is hiddenThe number of nodes of a layer is included, n is the number of nodes of an input layer, l is the number of nodes of an output layer, and t is a constant between 1 and 10) to calculate an initial value, and the optimal number of nodes is determined by a trial and error method;

fifthly, setting an activation function and a training function by taking a BP neural network algorithm as a basic model;

sixthly, setting a training parameter of the prediction model;

and seventhly, selecting the data in the data index system constructed through preprocessing in the second step as a training set and a prediction set to obtain a prediction model meeting the requirements.

In a preferred embodiment, the activation function includes an activation function of a hidden layer and an activation function of an output layer.

In a preferred embodiment, the training function includes a training function of a network and an algorithm training function.

Further, preferably, the activation function of the hidden layer is a hyperbolic tangent sigmoid function tansig, and the activation function of the output layer is a linear activation function purelin.

Preferably, the training function of the network is Levenberg-Marquardt, an algorithmic training function, rainlm.

In a preferred embodiment, the training parameters include a learning rate, a target error, and a maximum training step number.

Preferably, the macro economic index includes regional economic growth index (GDP), resident Consumption Price Index (CPI), Producer Price Index (PPI).

Further, the preprocessing the data in the constructed data index system in the second step refers to cleaning and integrating the data, and specifically includes:

a. and (5) preprocessing labor cost. Acquiring the disposable income of the manual work in each province and quarter through a national data website so as to measure the labor cost of each stage of each region;

b. and (4) pretreating the raw material price. Acquiring a copper price index from a Shanghai colored net, acquiring a steel price index from a Chinese combined steel net, and measuring and calculating an average value of each item of data according to monthly degrees;

c. regional economic growth indicator (GDP) data preprocessing. Acquiring quarter economic growth indexes of each province, namely the total GDP value of regional quarters, through a national data website;

d. and preprocessing resident Consumption Price Index (CPI) data. Acquiring monthly resident Consumption Price Indexes (CPI) of various provinces through a national data website to measure the inflation condition of the goods in various regions;

e. procurement Manager Index (PMI) data preprocessing. Acquiring a manufacturing PMI, a non-manufacturing PMI and a comprehensive PMI index through a Chinese logistics and purchasing network;

f. producer Price Index (PPI) data is preprocessed. Factory price indexes (PPI) of monthly producers of each province are obtained through a national data website so as to measure the price variation trend of the produced products in each region.

Further preferably, the satisfactory prediction model means that the average error is within 5%.

The method takes the BP neural network algorithm as a basic model algorithm, does not need a priori formula, can automatically mine rules from data, accurately predicts the purchase price through self-learning, self-adaption and nonlinear conversion capabilities and comprehensive consideration of various factors, and provides reference for differential bidding purchase.

Meanwhile, the model prediction result trained by the method is closer to the actual price range, the prediction error is within the acceptable range, the model has higher reliability, and the prediction result has certain reference value for differentiated purchasing strategies such as calibration packet division, and the rating principle.

Drawings

Fig. 1 is a diagram illustrating an MSE training performance curve in example 2.

Fig. 2 is a diagram illustrating a training state curve in example 2.

Fig. 3 is a schematic diagram comparing the predicted medium bid trend curve obtained in example 2 with the actual medium bid trend curve.

Detailed Description

In order that the invention may be better understood, we now provide further explanation of the invention with reference to specific examples.

Example 1

The method for predicting the purchase price of the power equipment disclosed in the embodiment comprises the following steps:

wherein the macro economic index preferably comprises regional economic growth index (GDP), resident Consumption Price Index (CPI), and Producer Price Index (PPI).

The data is selected after comprehensive consideration of multidimensional influence factors of the material price of the power enterprise equipment. The multi-dimensional data mainly comprises factors such as self production cost, market currency expansion, government macro regulation and control policies, regional development level, industry monopoly behaviors and the like. The type, name, index, unit and other attributes of each index data refer to table 1,

table 1:

preferably, in the present embodiment, the processing manners are respectively,

(m is the number of hidden layer nodes, n is the number of input layer nodes, l is the number of output layer nodes, and t is a constant between 1 and 10) to calculate an initial value, and determining the optimal number of nodes by a trial and error method;

fifthly, setting an activation function and a training function by taking a BP neural network algorithm as a basic model; preferably, in this embodiment, the activation function includes an activation function of a hidden layer and an activation function of an output layer, and the activation function of the hidden layer is a hyperbolic tangent sigmoid function tansig, and the activation function of the output layer is a linear activation function purelin. The training functions preferred in this embodiment include a training function of a network and an algorithm training function. And the training function of the network is Levenberg-Marquardt, the algorithm training function, rainlm.

Sixthly, setting a training parameter of the prediction model; preferred training parameters include learning rate, target error, maximum number of training steps.

And seventhly, selecting the data in the data index system constructed through preprocessing in the second step as a training set and a prediction sum to obtain a prediction model meeting the requirements, wherein the average error is preferably within 5% in the embodiment.

Example 2

In the embodiment, the invention is further explained and explained by taking the analysis and prediction of a reasonable purchase price interval of a certain 10kV transformer (rated capacity: 400kVA, installation mode: common, iron core material: silicon steel sheet, insulation mode: oil immersion) in Jiangsu province as an example.

The equipment purchase price prediction method disclosed in the embodiment comprises the following steps:

preferably, in the present embodiment, the processing manners are respectively,

a sample of the input data of a 10kV transformer in this example of the province of Jiangsu is shown in table 2,

table 2:

The BP neural network model structure established in this embodiment is [11,9,2], the activation function of the hidden layer is a hyperbolic tangent sigmoid function tan, the activation function of the output layer is a linear activation function purelin, and the training function of the network is a Levenberg-Marquardt algorithm training function trainlm.

In this embodiment, the learning rate is set to 0.001, the target error is set to 0.0001, and the maximum number of training steps is set to 300.

And seventhly, selecting the data in the data index system constructed through preprocessing in the second step as a training set and a prediction set to obtain a prediction model meeting the requirements, wherein the average error is preferably within 5% in the embodiment.

In the present embodiment, after the parameters are set, the normalized data is input into the network, and the training of the model is performed, and as a result, it is found that the evaluation error does not decrease any more after the number of steps reaches about 6, and even starts to increase. The maximum number of training steps is then changed to 8 to retrain the model. It can be seen from the diagram of the MSE training performance curve in FIG. 1 and the training state curve shown in FIG. 2 that the model achieves the optimal validation performance by the training step 5.

In the present embodiment, the total number of the selected 10kV transformers is 114 standard packet data, the first 94 standard packet data are used as training data, and the last 20 standard packet data are used for displaying the prediction effect. It can be known from the prediction curve and the error result statistical table shown in fig. 3 that the average error is within 5%, the overall error is small, and meanwhile, it can be seen from the graph that the upper and lower limits of the predicted reasonable purchase price basically follow the trend of the actual bid price, including the trend of the peak, and the actual bid price is basically within the prediction price interval. Accordingly, it can be considered that the BP neural network model obtained according to the parameter correction method disclosed in this embodiment has a strong prediction capability, and can meet the prediction requirement.

What has been described above is a specific embodiment of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principle of the present invention, and such improvements and modifications are also considered to be within the scope of the present invention.

Claims

1. The method for predicting the purchase price of the power equipment is characterized by comprising the following steps of:

sixthly, setting a training parameter of the prediction model;

2. The electric power equipment purchase price prediction method according to claim 1, characterized in that: the activation functions include an activation function of the hidden layer and an activation function of the output layer.

3. The electric power equipment purchase price prediction method according to claim 1, characterized in that: the training functions include a network training function and an algorithm training function.

4. The electric power equipment purchase price prediction method according to claim 2, characterized in that: the activation function of the hidden layer is a hyperbolic tangent sigmoid function tansig, and the activation function of the output layer is a linear activation function purelin.

5. The electric power equipment purchase price prediction method according to claim 2, characterized in that: the training function of the network is Levenberg-Marquardt, and the algorithm training function, rainlm.

6. The electric power equipment purchase price prediction method according to claim 3, characterized in that: the training parameters include learning rate, target error, maximum training step number.

7. The electric power equipment purchase price prediction method according to claim 1, characterized in that: the macroscopic economic index comprises a regional economic growth index (GDP), a resident Consumption Price Index (CPI) and a Producer Price Index (PPI).

8. The electric power equipment purchase price prediction method according to claim 1, characterized in that: the second step of preprocessing the data in the data index system is to clean and integrate the data, and specifically comprises the following steps:

9. The electric power equipment purchase price prediction method according to claim 1, characterized in that: the satisfactory prediction model means that the average error is within 5%.