CN107748936A - Based on the improved BP neural network life of storage battery prediction algorithm of genetic algorithm - Google Patents

Abstract

The invention discloses BP neural network is optimized based on the improved BP neural network life of storage battery prediction algorithm of genetic algorithm, including with genetic algorithm；Existing life of storage battery forecast reason is used for reference, establishes index system；Using corresponding achievement data as training sample, the BP neural network after training；The life of storage battery is predicted with the BP neural network after training.The present invention combines genetic algorithm and BP neural network, genetic algorithm is a kind of global search optimized algorithm based on biomechanism, utilize the initial weight and threshold values of genetic algorithm optimization BP neural network, recycle error back propagation method to find its optimal solution, avoid the situation for being absorbed in Local Extremum.

Description

Improved BP Neural Network Battery Life Prediction Algorithm Based on Genetic Algorithm

technical field

The invention relates to a storage battery life prediction algorithm, in particular to an improved BP neural network storage battery life prediction algorithm based on a genetic algorithm.

Background technique

With the development of information technology, more and more electrical equipment causes serious load to the power grid, so there will be power outages from time to time. In order to ensure the normal operation of electrical equipment, the battery of electrical equipment plays an important role. Therefore, how to manage and maintain power communication transmission equipment and batteries in a scientific, effective and practical way has become a major research topic for the communication management department.

The storage battery is the emergency backup power supply for many devices. In theory, the service life of the battery can reach more than 10 years, but the actual service life of the battery is only 4-6 years. By predicting the battery life of different brands under different temperature conditions, the effectiveness of power communication equipment can be effectively guaranteed. Using BP neural network to predict battery life has some defects, such as easy to fall into local extreme points.

Contents of the invention

In order to solve the above technical problems, the present invention provides an improved BP neural network battery life prediction algorithm based on genetic algorithm.

In order to achieve the above object, the technical scheme adopted in the present invention is:

Improved BP neural network battery life prediction algorithm based on genetic algorithm, including,

Optimize the BP neural network with genetic algorithm;

Learn from the existing battery life prediction ideas to establish an index system;

Use the corresponding index data as training samples, and the trained BP neural network;

Use the trained BP neural network to predict battery life.

The process of optimizing the BP neural network with genetic algorithm is as follows:

S1) population initialization;

S2) determine the fitness function;

S3) selection, crossover and mutation;

S4) Step S2 and step S3 are repeated until the evolution algebra is reached or the error requirement is met. At this time, the initial weight and threshold of the BP neural network optimized by the genetic algorithm are obtained.

Population initialization, for a 3-layer BP neural network with N input layers, L hidden layers, and M output layers, the length of the chromosome is S, and S=(N+1)L+(L+1)M.

The norm of the error matrix between the predicted value and the expected value is used as the output of the objective function, namely:

Among them, F is the norm of the error matrix of the predicted value and the expected value of the predicted sample, Yk is the expected value matrix, and Ok _s is the output matrix of the predicted value.

The fitness function takes the sorted fitness function.

The selection operation is performed by random traversal sampling.

The crossover operator adopts a single-point crossover operator.

The number of mutated genes is generated with a certain probability, and the mutated genes are selected randomly.

Beneficial effects achieved by the present invention: the present invention organically combines genetic algorithm and BP neural network, genetic algorithm is a kind of global search optimization algorithm based on biological mechanism, utilizes genetic algorithm to optimize the initial weight and threshold value of BP neural network, Then use the error backpropagation method to find its optimal solution, avoiding the situation of falling into local extreme points.

Description of drawings

Fig. 1 is the flow chart that optimizes BP neural network with genetic algorithm;

Fig. 2 is to adopt the prediction figure of the present invention in 12～15 degrees;

Fig. 3 adopts the prediction figure of the present invention in 15～18 degrees;

Fig. 4 adopts the prediction figure of the present invention in 18～21 degree;

Fig. 5 adopts the prediction figure of the present invention in 21～24 degrees;

Fig. 6 adopts the prediction figure of the present invention in 24～27 degrees;

Fig. 7 is a prediction diagram of the present invention within 27-30 degrees.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings. The following examples are only used to illustrate the technical solution of the present invention more clearly, but not to limit the protection scope of the present invention.

The improved BP neural network battery life prediction algorithm based on genetic algorithm includes the following steps:

Step 1, optimize the BP neural network with genetic algorithm.

As shown in Figure 1, the process of optimizing the BP neural network with the genetic algorithm is as follows:

S1) Population initialization:

For a 3-layer BP neural network with N input layers, L hidden layers, and M output layers, the length of the chromosome is S, and S=(N+1)L+(L+1)M.

S2) Determine the fitness function:

The fitness function adopts the sorted fitness function;

The norm of the error matrix between the predicted value and the expected value is used as the output of the objective function, namely:

Among them, F is the norm of the error matrix of the predicted value and the expected value of the predicted sample, Yk is the expected value matrix, and Ok _s is the output matrix of the predicted value.

S3) selection, crossover and mutation:

The selection operation is performed by random traversal sampling;

The crossover operator adopts a single-point crossover operator;

The number of mutated genes is generated with a certain probability, and the mutated genes are selected randomly.

S4) Step S2 and step S3 are repeated until the evolution algebra is reached or the error requirement is met. At this time, the initial weight and threshold of the BP neural network optimized by the genetic algorithm are obtained.

Step 2, learn from the existing battery life prediction ideas, and establish an index system, as shown in Table 1.

Table 1 Index System

Step 3, use the corresponding index data as the training sample, the trained BP neural network.

There are a lot of brand categories in the battery data, but some brand data correspond to a small amount of data and the time of operation is single, which cannot contribute to the model. On the contrary, it may affect the accuracy of the model, so the operation in the brand data is cleaned. A value with a single time and a small amount of data.

Ambient temperature is numerical data. However, for business requirements for ambient temperature, what we want more is the life of the battery in the ambient temperature range. Therefore, we need to first divide the ambient temperature into classified data of several ambient temperature ranges.

Step 4, use the trained BP neural network to predict the battery life.

In order to further illustrate the above algorithm, 2500 pieces of data were randomly selected from 3515 data sets as the training set, and the remaining 1015 pieces were used as the test set. The BP neural network algorithm and the BP neural network algorithm optimized based on the genetic algorithm were used to establish the prediction model respectively. Through the brand , ambient temperature and operation time to predict the remaining capacity.

The test results show that the accuracy of BP neural network algorithm is 87.1%, while the accuracy of BP neural network algorithm optimized based on genetic algorithm is 92.3%. The prediction results show that the BP neural network algorithm based on genetic algorithm optimization is superior to the BP neural network algorithm, and can be well applied to the battery life prediction scenario.

According to the predicted results, it is obvious that the remaining capacity curve based on the operation time under a given brand and temperature can be obtained, as shown in Figures 2-7. It can be seen that the remaining capacity curves of different brands change with the operation time within a given temperature range. Observing the experimental results, we can draw the following conclusions: the overall trend is a downward process, and the rate of decline slows down over time. For the same brand, the decline trend of remaining capacity is different under different temperature conditions. It can be seen from the figure that in the range of 18-21 degrees, the battery life is longer.

Generally speaking, when the remaining capacity of the battery drops below 80%, it should be replaced, otherwise the business will be affected. Therefore, setting the remaining capacity threshold can predict the time when batteries of different brands need to be replaced under different temperature environments, so as to guide users in battery operation management and maintenance.

The above algorithm builds a communication battery life prediction model through mining and modeling of communication battery-related data. By successfully estimating the service life of batteries of different brands in different temperature ranges, the battery operation status monitoring and life prediction are realized. System security operation monitoring and health status management provide support.

The above is only a preferred embodiment of the present invention, it should be pointed out that for those of ordinary skill in the art, without departing from the technical principle of the present invention, some improvements and modifications can also be made. It should also be regarded as the protection scope of the present invention.

Claims (8)

1. it is based on the improved BP neural network life of storage battery prediction algorithm of genetic algorithm, it is characterised in that：Including,

BP neural network is optimized with genetic algorithm；

Existing life of storage battery forecast reason is used for reference, establishes index system；

Using corresponding achievement data as training sample, the BP neural network after training；

The life of storage battery is predicted with the BP neural network after training.

2. according to claim 1 be based on the improved BP neural network life of storage battery prediction algorithm of genetic algorithm, it is special Sign is：It is to the process that BP neural network optimizes with genetic algorithm,

S1) initialization of population；

S2 fitness function) is determined；

S3) select, intersect and make a variation；

S4) repeat step S2 and step S3, until reaching evolutionary generation or meeting error requirements, now, obtain passing through heredity The initial weight and threshold values for the BP neural network that algorithm optimization is crossed.

3. according to claim 2 be based on the improved BP neural network life of storage battery prediction algorithm of genetic algorithm, it is special Sign is：Initialization of population, there is N number of input layer, L hidden layer, 3 layers of BP neural network of M output layer, dye for one The length of colour solid is S, S=(N+1) L+ (L+1) M.

4. according to claim 2 be based on the improved BP neural network life of storage battery prediction algorithm of genetic algorithm, it is special Sign is：Output using predicted value and the norm of the error matrix of desired value as object function, i.e.,：

<mrow> <mi>F</mi> <mo>=</mo> <mfrac> <mn>1</mn> <mn>2</mn> </mfrac> <msubsup> <mi>&amp;Sigma;</mi> <mrow> <mi>t</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>S</mi> </msubsup> <msubsup> <mi>&amp;Sigma;</mi> <mrow> <mi>k</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>M</mi> </msubsup> <msup> <mrow> <mo>(</mo> <mi>Y</mi> <mi>k</mi> <mo>-</mo> <msub> <mi>Ok</mi> <mi>s</mi> </msub> <mo>)</mo> </mrow> <mn>2</mn> </msup> </mrow>

Wherein, F be pre- test sample predicted value and desired value error matrix norm, Yk for it is expected value matrix, Ok_sFor predicted value Output matrix.

5. according to claim 2 be based on the improved BP neural network life of storage battery prediction algorithm of genetic algorithm, it is special Sign is：Fitness function is using the fitness function to sort.

6. according to claim 2 be based on the improved BP neural network life of storage battery prediction algorithm of genetic algorithm, it is special Sign is：Selection operation is carried out by the way of random ergodic sampling.

7. according to claim 2 be based on the improved BP neural network life of storage battery prediction algorithm of genetic algorithm, it is special Sign is：Crossover operator uses single-point crossover operator.

8. according to claim 2 be based on the improved BP neural network life of storage battery prediction algorithm of genetic algorithm, it is special Sign is：Mutant gene number is produced with certain probability, the gene to morph is selected with random device.