CN117350146A - A health evaluation method of drainage pipe network based on GA-BP neural network - Google Patents

Abstract

The invention discloses a drainage pipe network health evaluation method based on a GA-BP neural network, and belongs to the technical field of drainage pipe network health evaluation. Determining a drainage pipe network health evaluation index system by analyzing various factors affecting the drainage pipe network health; establishing a BP neural network hierarchical structure, and determining the input and output of the BP neural network; optimizing and screening the weight and the threshold value of the BP neural network structure by adopting a genetic algorithm, and evaluating the effect of the GA-BP neural network model; the finally determined GA-BP model is used for evaluating the health of the drainage pipe network. The GA-BP model used by the invention has strong interpretability, solves the problem of high evaluation subjectivity in the previous evaluation process of the health of the drainage pipe network, is convenient for determining the pipe network detection priority, is beneficial to the safety transportation and disaster risk active management of pipe network facilities, and simultaneously provides decision support for the optimization and transformation of the pipe network.

Description

A health evaluation method of drainage pipe network based on GA-BP neural network

Technical field

The present invention relates to the technical field of drainage pipe network health assessment, and in particular to a drainage pipe network health evaluation method based on GA-BP neural network.

Background technique

my country's drainage pipe network has huge assets and is an indispensable part of municipal facilities. It is related to urban industrial construction and the normal operation of residents' lives. The management of the renovation, planning, design, construction and project acceptance process of drainage pipe network facilities is one of the important business contents of the urban drainage management department. Due to the long age of some pipelines and the aging of the pipe network, the safe operation of the drainage pipe network faces severe challenges, and the risk assessment of the drainage pipe network has received more and more attention.

At present, the repair and maintenance of drainage pipe networks are relatively haphazard, and the evaluation process is relatively simple. Some even replace pipes directly based on their age without evaluation, resulting in a waste of resources. To change this phenomenon, a systematic risk assessment of the pipeline network must be carried out before repairing or maintaining the pipeline network. However, the current risk assessment of drainage pipe networks mostly uses the analytic hierarchy process for weight analysis, and then uses the fuzzy comprehensive evaluation method to determine the risk level. The evaluation is too subjective, so a more accurate and reliable evaluation method is needed.

Contents of the invention

The purpose of this invention is to provide a method for evaluating the health of drainage pipe networks based on GA-BP neural network, optimizing the weights and thresholds of the BP neural network through genetic algorithms, and overcoming the problem that the BP neural network algorithm is prone to falling into local minimums during the calculation process. , shortcomings of unstable generalization ability, the evaluation model and parameters are highly interpretable, determine the priority of pipe network detection, contribute to the safe operation and maintenance of pipe network facilities and active management of disaster risks, and evaluate the health risk of the drainage pipe network More accurate and reliable.

In order to achieve the above objectives, the present invention provides a drainage pipe network health evaluation method based on GA-BP neural network. The steps are as follows:

S1. Analyze various factors that affect the health of the drainage pipe network, clarify the principles for constructing the indicator system based on the actual drainage conditions of the region, combined with literature analysis and national standard requirements, and determine the health evaluation index system of the drainage pipe network;

S2. Determine the input and output variables of the BP neural network structure, perform data preprocessing on the number of input layer nodes, the number of output layer nodes, and the number of hidden layer nodes, establish the BP neural network hierarchical structure, and train the BP neural network structure. ;

S3. Set up the genetic algorithm to improve the BP neural network parameters, including coding settings, determining the population size, selecting fitness functions, and genetic operation settings. According to the genetic algorithm, optimize and screen the determined BP neural network structure weights and thresholds to obtain GA -BP neural network model;

S4. Evaluate the health of the drainage pipe network based on the finalized GA-BP neural network structure. Divide the health of the drainage pipeline into four levels: r1 higher risk, r2 high risk, r3 medium risk, and r4 low risk. Determine the drainage level. The priority of pipeline network operation and maintenance provides decision-making support for pipeline network optimization and transformation, and scientifically formulates pipeline transformation and optimization plans.

Preferably, the evaluation index system in step S1 is divided into three levels:

1) Target layer, drainage pipe network health U;

2) The influence of three aspects: the ontology factor U1, external factor U2, and environmental factor U3 of the pipeline network;

3) The respective sub-influencing factors of the three influencing factors of ontology factor U1, external factor U2 and environmental factor U3.

Preferably, step S2 specifically includes:

S21. Determine the number of hidden layers and neurons in the hidden layer. The transfer function of the hidden layer is the sigmoid function, and the transfer function of the output layer is a linear function;

S22. Through the function in step S21, obtain the predicted value and the actual measured value and calculate the root mean square error and regression R value. Compare and select the number of hidden layer nodes with the smallest RMSE and the largest correlation coefficient R value as the best parameter of the model;

S23. Design of input layer and output layer: Determine the assigned values of 10 evaluation indicators in the input layer as input parameters, use 10 evaluation indicators in the pipe network health evaluation indicators as the input layer, and use the health level of the drainage pipeline as the target output. value;

S24. Data preprocessing: Independent variables include continuous variables and categorical variables. Binomial categorical variables are coded from 0 to 1. 0 represents a pipeline that has not been damaged, 1 represents a pipeline that has been damaged, and multiple categorical variables are coded with dummy variables. .

Preferably, in step S3, before performing BP neural network training, the population individuals are used as BP neural network parameters, and the fitness of the population individuals is calculated through the error of the final result given by the BP neural network, and genetic operations are performed on individuals with good fitness. Set, according to the genetic algorithm, the optimal individual is used as the initial weight and threshold of the BP neural network structure, and then the BP algorithm quickly adjusts the weight and threshold according to the negative gradient direction;

The confusion matrix method is used to evaluate the effect of the GA-BP neural network model, and the ROC curve is used to sort the samples through the prediction results.

Preferably, in step S3,

Genetic algorithm coding: the coding method of coding length L is real number coding

L＝n×h+h×m+h+m

Among them, n is the number of neurons in the input layer, h is the number of nodes in the hidden layer, and m is the number of neurons in the output layer;

Determine the population size: set between 10-20;

Determine the fitness function: The calculation formula for the fitness value F of each individual in the population is:

F＝A∑|P _K -t _K |

Among them, F is the fitness value, A is the coefficient, and t _k is the corresponding actual output value.

Preferably, the genetic operation settings include selection operation, crossover operation, and mutation operation;

Selection operation: Using roulette and a selection strategy based on fitness ratio, the selection probability P _i of each individual i is

Among them, f _i is the fitness value of individual i, and n is the number of individuals in the population;

Crossover operation: the crossover between the k-th chromosome b _k and the l-th chromosome b _l at the j position. The specific intersection formula is:

b _Kj ＝b _Kj (1-a)+b _Ij a

b _Ij ＝b _Ij (1-a)+b _Kj a

Mutation operation: mutation of the j-th gene of the i-th individual. The specific mutation operation is:

Among them, b _min is the lower bound of gene b _ij , b _max is the previous limit of gene b _ij , r ₁ is a random number, g is the number of iterations, and G _max is the maximum number of evolutions.

Therefore, the present invention adopts a drainage pipe network health evaluation method based on the GA-BP neural network with the above structure, and achieves the following beneficial effects:

This invention selects three angle index types: pipe network equipment body factors, external factors and environment, selects relative influencing factors as sub-indexes, and uses genetic algorithms to optimize the weights and thresholds of the BP neural network, overcoming the problem of the BP neural network algorithm in calculation The process is prone to fall into local minima and the generalization ability is unstable, which makes the performance of the model better and the health risk assessment of the drainage pipe network more accurate and reliable.

The technical solution of the present invention will be further described in detail below through the accompanying drawings and examples.

Description of drawings

Figure 1 is a flow chart of the present invention;

Figure 2 is a diagram of the health index system of the drainage pipe network of the present invention;

Figure 3 is a diagram of data types and units of various indicators of the health of the drainage pipe network of the present invention;

Figure 4 is a structural diagram of the three-layer BP neural network of the present invention;

Figure 5 is a flow chart of the GA-BP neural network of the present invention.

Detailed ways

The technical solution of the present invention will be further described below through the drawings and examples.

Unless otherwise defined, technical terms or scientific terms used in the present invention shall have the usual meaning understood by a person with ordinary skill in the field to which the present invention belongs. "First", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. Words such as "include" or "comprising" mean that the elements or things appearing before the word include the elements or things listed after the word and their equivalents, without excluding other elements or things. The terms "setting", "installation" and "connection" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection. Connected, it can also be connected indirectly through an intermediary, or it can be an internal connection between two components. "Up", "down", "left", "right", etc. are only used to express relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

Example

As shown in Figures 1-5, the present invention provides a drainage pipe network health evaluation method based on GA-BP neural network. The steps are as follows:

1. (Step S1) Analyze various factors that affect the health of the drainage pipe network. Based on the actual drainage conditions in the region, combined with the literature analysis method and national standard requirements, clarify the principles for constructing the index system (the principle of completeness, the principle of representativeness, and operable quantification). principle), determine the health evaluation index system of the drainage pipe network.

As shown in Figure 2-3, the evaluation index system is divided into three levels:

1) Target layer, drainage pipe network health U;

2) The influence of three aspects: the ontology factor U1, external factor U2, and environmental factor U3 of the pipeline network;

3) The respective sub-influencing factors of the three influencing factors of ontology factor U1, external factor U2 and environmental factor U3.

Indicators related to pipeline network health assessment mainly include: pipeline structural characteristics, pipeline loads and effects, boundary conditions and durability. This invention combines previous research results, and refers to the "Urban Drainage Pipe Structure Grade Assessment" (DB11/T1492-2017) and "Urban Drainage Pipe Detection and Evaluation Technical Regulations" (CJJ181-2012) and other standards and specifications, combined with regional temperature, geology and drainage conditions to determine the health insurance evaluation system for the drainage pipe network.

From three aspects: pipeline structural factors, external factors and environmental factors, pipe age, pipe material, pipe diameter, pipeline length, pipeline burial depth, pipeline flow rate, operating pressure, road type, weather temperature, and construction degree around the pipeline are selected as evaluation indicators.

BP (Back Propagation) neural network model, that is, feedback neural network model, is one of the most mature and extensive neural network models at present. It relies on the complex connections of a large number of neurons to form a nonlinear dynamic system; it generally consists of an input layer, It consists of hidden layer and output layer.

When the network is running, the input data is propagated from front to back between layers, and the output of each layer only affects the next layer; in addition, neurons in the same layer are neither directly connected nor have any interference; there is no interference between neurons. The error will inevitably propagate from back to front. Until the error reaches the set accuracy, the network will stop adjusting the weights and thresholds based on the error between the actual output and the expected output.

2. (Step S2) Determine the input and output variables of the BP neural network structure, perform data preprocessing on the number of input layer nodes, the number of output layer nodes, and the number of hidden layer nodes, establish a BP neural network hierarchical structure, and train BP neural network structure.

As shown in Figure 4, the specific steps are:

S21. Determine the number of hidden layers: BP neural network mainly relies on hidden layers to exert its learning ability. The selection of the number of hidden layers needs to be determined according to the characteristics of the problem.

S22. Determine the number of hidden layer neurons: If there are too many hidden layer neurons, it will increase the amount of network calculations and prone to over-fitting problems; if there are too few hidden neurons, it will affect the network performance and fail to achieve the expected results. The number of hidden layer neurons in the network is directly related to the complexity of the actual problem, the number of input and output layer neurons, and the setting of the expected error;

The transfer function of the hidden layer is the sigmoid function, also known as the hyperbolic tangent S-shaped function, and the formula is

The output layer uses a linear function as the transfer function, generally using the tansig function, the formula is

Through the above function, the root mean square error (RMSE) and regression R value of the predicted value and actual measurement are obtained. By comparison, the number of hidden layer nodes when the RMSE is the smallest and the correlation coefficient R value is the largest is selected, which is the optimal parameter of the model. The formula is:

S23. Design of the input layer and output layer: Determine the 10 evaluation index assignment values of the input layer as input parameters, and use the 10 evaluation indicators {U11, U12, U13, U14, U21, U22, U23, U31, U32, U33} as input layer;

The health level of the drainage pipe is used as the target output value; the health of the drainage pipe is divided into four levels, so the number of neurons in the output layer is 4.

S24. Data preprocessing: independent variables include continuous variables and categorical variables; binomial categorical variables are coded from 0 to 1, with 0 representing pipelines that have not been damaged and 1 representing pipelines that have been damaged; attribute values of multiple categorical variables It is a parallel relationship, and its meaning cannot be defined by its numerical value, so dummy variable coding is required. That is, in the index system, pipe material, road type, weather temperature, and week-to-week construction level are multi-parallel classification variables, using dummy variable coding; pipe age, pipe diameter, pipeline length, pipeline burial depth, pipeline flow rate, and operating pressure are selected for maximum and minimum method to standardize the data.

There are large differences in the dimensions and magnitudes of other evaluation index data. It is necessary to use the max-min method to standardize the data. Unify the input and output data dimensions and classify them into the range of 0 to 1 to avoid data samples with different orders of magnitude and dimensions. The impact brought by it, and reducing the error speeds up the convergence speed of the neural network. The formula is:

Among them, y is the new sample data, x is the original data, x _max is the maximum value of the original data sequence; x _min is the minimum value of the original data sequence.

80% of the sample set is used for the training set and 20% is used for the test set. The former is used to train the drainage pipe network health prediction model, and the latter is used to verify the drainage pipe network health prediction model.

Genetic Algorithm evolved from natural selection and natural inheritance in Darwin's theory of evolution. It can simulate genetic mechanisms and species evolution rules, thereby forming a parallel random search optimization method; it has global search capabilities, simple operation, and robustness. and strong ability to learn independently;

In order to solve the problems that the BP neural network is easy to fall into local optimum, the hidden layer structure is difficult to determine and the real-time performance is poor, it is combined with the genetic algorithm to form the GA-BP neural network optimization algorithm; by using the powerful global search capability of the genetic algorithm, Optimize so that the optimized weights and thresholds meet the built fitness requirements, and then assign the optimized weights and thresholds to the BP network.

3. (Step S3) As shown in Figure 5, set the genetic algorithm to improve the BP neural network parameters, including coding settings, determining the population size, selecting fitness functions, and genetic operation settings. According to the genetic algorithm, the determined BP neural network structure is The weights and thresholds are optimized and screened to obtain the GA-BP neural network model.

Before training the BP neural network, the population individuals are used as BP neural network parameters. The BP neural network finally gives the error of the result, calculates the fitness of the population individuals, and performs genetic operation settings on individuals with good fitness. According to the genetic algorithm, the The optimal individual serves as the initial weight and threshold of the BP neural network structure, and then the BP algorithm quickly adjusts the weight and threshold according to the negative gradient direction;

The confusion matrix method is used to evaluate the effect of the GA-BP neural network model, and the ROC curve is used to sort the samples through the prediction results.

1) Genetic algorithm coding: There are two main types of coding: binary coding and real number coding. Although binary coding is relatively simple, it has the disadvantage of too long composition length. Therefore, real number coding is used as the coding method. The calculation formula of coding length L is as follows

L＝n×h+h×m+h+m

Among them, n is the number of neurons in the input layer, h is the number of nodes in the hidden layer, and m is the number of neurons in the output layer.

2) Determine the population size: There is no unified standard for the population size, which is usually set between 10-20.

3) Determine the fitness function: The fitness function is an evaluation function for judging the quality of individuals in the population. It is used as the basis for optimization search. The calculation formula for the fitness value F of each individual in the population is:

F＝A∑|P _K -t _K |

Among them, F is the fitness value, A is the coefficient, and t _k is the corresponding actual output value.

Genetic operation settings include selection operations, crossover operations, and mutation operations:

a) Selection operation: Using roulette and a selection strategy based on fitness ratio, the selection probability P _i of each individual i is

Among them, f _i is the fitness value of individual i, and n is the number of individuals in the population;

b) Crossover operation: The greater the crossover probability, the faster it converges to the optimal solution range. However, if the crossover probability is too large, the model will converge to a specific solution, and subsequent optimization operations cannot be performed;

The intersection of the k-th chromosome b _k and the l-th chromosome b _l at the j position, the specific intersection formula is:

b _Kj ＝b _Kj (1-a)+b _Ij a

b _Ij ＝b _Ij (1-a)+b _Kj a

c) Mutation operation: Mutation refers to changing a gene segment of an individual selected in advance according to a set probability. The mutation probability is generally between 0.001-0.1. If the mutation probability is too large, the entire process will oscillate; if the mutation probability is too small, it will be difficult to search for the optimal solution;

The mutation of the j-th gene of the i-th individual, the specific mutation operation is:

Among them, b _min is the lower bound of gene b _ij , b _max is the previous limit of gene b _ij , r ₁ is a random number, g is the number of iterations, and G _max is the maximum number of evolutions.

The number of termination evolutionary generations is usually selected between 50-100;

Use the initial weights and thresholds of the BP neural network optimized by the genetic algorithm to set the initial training parameters, and use training samples to train the BP neural network structure.

Use the confusion matrix method to evaluate the model effect: Assume that n _ij represents the number of samples of class i classified into class j, K is the sample type, and the recall rate R _i of the sample of class i is the following formula

The ROC curve is used to sort the samples by prediction results. The curve drawn with FPR as the horizontal axis and TPR as the vertical axis is the ROC curve. Calculate the true positive rate (TPR) and the false positive rate (FPR). The true positive rate indicates the proportion of actual positive examples to all positive examples among the samples currently assigned to positive examples; the false positive rate indicates the current samples that are incorrectly assigned to positive examples. The proportion of actual negative examples in the category to the total number of negative examples, the formula is

Among them, TP is a true example, which means that the sample is actually a positive example and the model prediction result is a positive example; FP is a false positive example, which means that the sample is actually a negative example but the model prediction result is a positive example; TN is a true negative example, which means The sample is actually a negative example and the model prediction result is a negative example; FP is a false counterexample, which means that the sample is actually a positive example but the model prediction result is a negative example.

4. (Step S4) Evaluate the health of the drainage pipe network based on the finalized GA-BP neural network structure, and divide the health of the drainage pipe network into four categories: r1 high risk, r2 high risk, r3 medium risk, and r4 low risk. Level R = (r1, r2, r3, r4), determine the priority of operation and maintenance of the drainage pipe network, provide decision support for the optimization and transformation of the pipe network, and scientifically formulate optimization plans for pipeline reconstruction.

r1 higher risk: refers to the fact that the pipe network cannot meet the expected functions and needs to be updated immediately. It has obvious signs of deterioration and the possibility of damage is extremely high and needs to be maintained immediately;

r2 High risk: Refers to the fact that the pipeline network is in an unreliable state and is lower than the design standard, which will directly affect its function and safety, and the maintenance plan of the pipeline network needs to be customized;

r3 medium risk: refers to that the pipeline network is in a generally normal state and has general signs of deterioration, and the pipeline network needs to be monitored regularly;

r4 low risk: refers to meeting the expected functional requirements, the pipeline is in good condition, and basically does not require management.

Therefore, the present invention adopts a drainage pipe network health evaluation method based on GA-BP neural network, selects three angle index types of pipe network equipment ontology factors, external factors and environment, selects relative influencing factors as sub-indexes, and uses The genetic algorithm optimizes the weights and thresholds of the BP neural network, overcoming the shortcomings of the BP neural network algorithm that is easy to fall into local minimums and unstable generalization capabilities during the calculation process, making the performance of the model better and reducing the health risks of the drainage pipe network. Evaluations are more accurate and reliable.

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 of ordinary skill in the art should understand that: The technical solution of the present invention may be modified or equivalently substituted, but these modifications or equivalent substitutions cannot cause the modified technical solution to depart from the spirit and scope of the technical solution of the present invention.

Claims (6)

1. A method for evaluating the health of drainage pipe networks based on GA-BP neural network, which is characterized in that the steps are as follows: S1. Analyze various factors that affect the health of the drainage pipe network, clarify the principles for constructing the indicator system based on the actual drainage conditions of the region, combined with literature analysis and national standard requirements, and determine the health evaluation index system of the drainage pipe network; S2. Determine the input and output variables of the BP neural network structure, perform data preprocessing on the number of input layer nodes, the number of output layer nodes, and the number of hidden layer nodes, establish the BP neural network hierarchical structure, and train the BP neural network structure. ; S3. Set up the genetic algorithm to improve the BP neural network parameters, including coding settings, determining the population size, selecting fitness functions, and genetic operation settings. According to the genetic algorithm, optimize and screen the determined BP neural network structure weights and thresholds to obtain GA -BP neural network model; S4. Evaluate the health of the drainage pipe network based on the finalized GA-BP neural network structure. Divide the health of the drainage pipeline into four levels: r1 higher risk, r2 high risk, r3 medium risk, and r4 low risk. Determine the drainage level. Priority of pipeline network operation and maintenance. 2. A drainage pipe network health evaluation method based on GA-BP neural network according to claim 1, characterized in that: the evaluation index system in step S1 is divided into three levels: 1) Target layer, drainage pipe network health U; 2) The three influencing factors of the pipeline network are the ontology factor U1, the external factor U2, and the environmental factor U3; 3) The respective sub-influencing factors of the three influencing factors of ontology factor U1, external factor U2 and environmental factor U3. 3. A drainage pipe network health evaluation method based on GA-BP neural network according to claim 2, characterized in that step S2 specifically includes: S21. Determine the number of hidden layers and neurons in the hidden layer. The transfer function of the hidden layer is the sigmoid function, and the transfer function of the output layer is a linear function; S22. Through the function in step S21, obtain the predicted value and the actual measured value and calculate the root mean square error and regression R value. Compare and select the number of hidden layer nodes with the smallest RMSE and the largest correlation coefficient R value as the best parameter of the model; S23. Design of input layer and output layer: Determine the assigned values of 10 evaluation indicators in the input layer as input parameters, use 10 evaluation indicators in the pipe network health evaluation indicators as the input layer, and use the health level of the drainage pipeline as the target output. value; S24. Data preprocessing: Independent variables include continuous variables and categorical variables. Binomial categorical variables are coded from 0 to 1. 0 represents a pipeline that has not been damaged, 1 represents a pipeline that has been damaged, and multiple categorical variables are coded with dummy variables. . 4. A drainage pipe network health evaluation method based on GA-BP neural network according to claim 3, characterized in that: in step S3, before performing BP neural network training, the population individuals are used as BP neural network parameters, Through the error of the final result given by the BP neural network, the fitness of the population individuals is calculated, and genetic operation settings are performed on individuals with good fitness. According to the genetic algorithm, the optimal individual is used as the initial weight and threshold of the BP neural network structure, and then The BP algorithm quickly adjusts the weights and thresholds according to the negative gradient direction; The confusion matrix method is used to evaluate the effect of the GA-BP neural network model, and the ROC curve is used to sort the samples through the prediction results. 5. A drainage pipe network health evaluation method based on GA-BP neural network according to claim 4, characterized in that: in step S3, Genetic algorithm coding: the coding method of coding length L is real number coding L＝n×h+h×m+h+m Among them, n is the number of neurons in the input layer, h is the number of nodes in the hidden layer, and m is the number of neurons in the output layer; Determine the population size: set between 10-20; Determine the fitness function: The calculation formula for the fitness value F of each individual in the population is: F＝A∑|P _K -t _K | Among them, F is the fitness value, A is the coefficient, and t _k is the corresponding actual output value. 6. A drainage pipe network health evaluation method based on GA-BP neural network according to claim 5, characterized in that: genetic operation settings include selection operation, crossover operation, and mutation operation; Selection operation: Using roulette and a selection strategy based on fitness ratio, the selection probability P _i of each individual i is Among them, f _i is the fitness value of individual i, and n is the number of individuals in the population; Crossover operation: the crossover between the k-th chromosome b _k and the l-th chromosome b _l at the j position. The specific intersection formula is: b _Kj ＝b _Kj (1-a)+b _Ij a b _Ij ＝b _Ij (1-a)+b _Kj a Mutation operation: mutation of the j-th gene of the i-th individual. The specific mutation operation is: Among them, b _min is the lower bound of gene b _ij , b _max is the previous limit of gene b _ij , r ₁ is a random number, g is the number of iterations, and G _max is the maximum number of evolutions.