Sonar detection efficiency evaluation method based on GA-BP neural network

Abstract

The invention relates to a sonar detection efficiency evaluation method based on a GA-BP neural network, and belongs to the technical field of equipment efficiency evaluation. Comprising the following steps: 1, setting a BP neural network structure; 2, optimizing and adjusting the initial threshold value and the weight in the BP neural network by using a genetic algorithm to obtain the optimal initial threshold value and the weight value of the BP neural network; 3, training the GA-BP neural network to enable the GA-BP neural network to have the capability of classifying data influencing the detection efficiency, and dividing the GA-BP neural network into five grades; and 4, inputting test data into the GA-BP neural network, obtaining a classification result by predicting the network, and evaluating the detection efficiency of the sonar in the current marine environment. According to the method, the stricter requirement of sonar detection efficiency evaluation in the marine environment is met, a currently available evaluation method tool library can be further expanded by introducing the method, and the accuracy and robustness of evaluation are greatly improved.

Description

Technical Field

[0001] This invention relates to a sonar detection performance evaluation method based on GA-BP neural network, belonging to the field of equipment performance evaluation technology. Background Technology

[0002] Sonar is a key piece of equipment for underwater acoustic detection. A scientific, objective, and reasonable evaluation method can provide valuable reference for the maintenance and support of sonar equipment, and is crucial for improving naval combat capabilities. The premise and foundation of sonar detection effectiveness evaluation is to determine scientifically reasonable evaluation indicators and effective evaluation methods. When selecting effectiveness evaluation indicators, they should be as concise as possible to avoid information redundancy, while ensuring comprehensive evaluation content and no omissions in information coverage. To accurately evaluate sonar detection effectiveness, various indicators need to be considered comprehensively. A scientific and reasonable evaluation can not only comprehensively reflect sonar performance, but also provide valuable reference and guidance for the demonstration, development, and improvement of sonar equipment. However, traditional sonar detection effectiveness evaluation methods rely excessively on the personal experience of experts, leading to overly subjective evaluation results and a lack of objectivity and scientific rigor. This, to some extent, affects the accuracy and reliability of the evaluation. Therefore, an objective and reasonable evaluation method is crucial. Scientific and reasonable evaluation results can provide effective value for long-term sonar design and development. Therefore, evaluating the detection effectiveness of sonar systems can not only save research and development and production costs, but also promote the improvement and long-term development of sonar equipment. Due to the complexity and variability of the underwater environment, the detection effectiveness of sonar is affected by a variety of factors, such as water temperature and salinity. Changes in these factors can lead to changes in the propagation characteristics of sound waves, thereby affecting the detection effectiveness of sonar.

[0003] Neural networks, as mathematical models designed to simulate the complex functions of the human brain, excel in solving nonlinear problems, thus becoming a powerful tool for performance evaluation. Therefore, to achieve accurate "classification" of sonar detection performance under the influence of marine hydrological environments, this invention applies neural networks to the evaluation of sonar detection performance. Backpropagation (BP) neural networks hold a core position in the field of neural networks and are widely used. This network structure not only represents the essence of neural network technology but also reflects its perfection and maturity. Although BP neural networks dominate artificial neural network algorithms, their application still faces a series of challenges, such as slow convergence speed, difficulty in ensuring the acquisition of the global optimum, and the difficulty in determining the network architecture. Furthermore, the network construction process, the setting of initial weights and thresholds are crucial to performance, but these factors are often difficult to control precisely. To address these challenges, genetic algorithms can be used to optimize and adjust BP neural networks, thereby improving their performance and effectiveness.

[0004] Genetic algorithms are search heuristic algorithms based on the principles of natural selection and genetics. They find optimal solutions to problems by simulating biological evolution. The core of this algorithm lies in representing potential solutions to a problem as individuals in a population, where each individual is a candidate solution, and these are converted into genotype sequence structures through encoding.

[0005] Genetic algorithms start with a randomly generated initial set of solutions (initial population) and use a fitness function to evaluate the quality of each solution. The fitness function is designed based on the characteristics of the problem and measures how close each individual is to the optimal solution. Individuals with higher fitness are preferentially retained during the evolutionary process, while individuals with lower fitness are eliminated.

[0006] Selection is a crucial step in genetic algorithms, determining which individuals will be chosen for subsequent genetic operations. Common selection methods include roulette wheel selection, which uses an individual's fitness score to determine its probability of selection.

[0007] Crossover is one of the most important genetic operations in genetic algorithms, simulating the gene recombination process in biological evolution. Through crossover, genetic algorithms can generate new individuals that combine the superior genes of their parents, potentially exhibiting better fitness. The specific crossover operation can be flexibly designed according to the characteristics of the problem.

[0008] Mutation is an important mechanism for introducing new genes in genetic algorithms. It randomly changes certain genes of an individual with a certain probability, which helps to suppress local convergence.

[0009] By repeatedly performing selection, crossover, and mutation operations, genetic algorithms can gradually evolve a population, causing its individuals to approach the optimal solution to the problem. When a termination condition is met (such as reaching a preset number of iterations or finding a known optimal solution), the algorithm stops running and returns the best individual in the current population as the solution to the problem. In summary, genetic algorithms search for the optimal solution to a problem in the solution space. They have advantages such as strong global search capabilities and applicability to complex problems, and therefore have been widely used in many fields. Summary of the Invention

[0010] The purpose of this invention is to address the shortcomings of the existing technology and provide a sonar detection performance evaluation method based on GA-BP neural network. This method is based on a BP neural network optimized by genetic algorithm (GA-BP neural network) to meet the more stringent requirements of sonar detection performance evaluation in marine environments. By introducing this method, the currently available evaluation method tool library can be further expanded, and the accuracy and robustness of the evaluation can be greatly improved.

[0011] To achieve the above objectives, the present invention adopts the following technical solution: The present invention provides a sonar detection performance evaluation method based on GA-BP neural network, characterized by the following steps: Step 1: Set up the BP neural network structure; Step 2: Use a genetic algorithm to optimize and adjust the initial threshold and weights in the BP neural network to obtain the optimal initial threshold and weight values ​​for the BP neural network; Step 3: Train the GA-BP neural network to enable it to classify data that affects detection performance into five levels: "very good", "good", "average", "poor" and "very poor". Step 4: Input the test data into the GA-BP neural network, obtain the classification results through the network's predictions, and evaluate the detection performance of sonar in the current marine environment based on these classification results.

[0012] The advantages of the sonar detection performance evaluation method based on GA-BP neural network of the present invention are as follows: The complexity and variability of the marine environment significantly impact the operational effectiveness of underwater acoustic detection equipment, increasing the difficulty of assessment. Evaluating operational effectiveness is crucial for accurately assessing combat performance and determining optimal military action plans. Assessing the operational effectiveness of underwater acoustic detection equipment not only improves existing equipment but also provides a reference for the development of new equipment, thereby optimizing weapon and equipment configuration in national defense and assisting military planners and commanders in making optimal decisions. This invention proposes a sonar detection effectiveness evaluation method based on a GA-BP neural network. This scientific, objective, and reasonable evaluation method provides valuable reference for the maintenance and support of sonar equipment and is crucial for enhancing naval combat capabilities. Attached Figure Description

[0013] Figure 1 Flowchart of the genetic algorithm for finding the optimal BP network; Figure 2 This is a graph showing the fitness change curve of the genetic algorithm. Figure 3 The image shows the prediction results of the training set for optimizing the BP neural network based on the genetic algorithm. Figure 4 The image shows the test set prediction results based on the optimization of the BP neural network using a genetic algorithm. Figure 5 The confusion matrix diagram of the training set for optimizing a BP neural network based on a genetic algorithm; Figure 6 The confusion matrix diagram of the test set for optimizing a BP neural network based on a genetic algorithm; Figure 7 This is a graph showing the prediction results for the training set based on a BP neural network. Figure 8 This is a graph showing the prediction results for the test set based on a BP neural network. Figure 9 This is a confusion matrix diagram of the training set based on a BP neural network. Figure 10 This is the confusion matrix of the test set based on a BP neural network. Detailed Implementation

[0014] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0015] Reference Figure 1-10 This embodiment of a sonar detection performance evaluation method based on a GA-BP neural network includes the following steps: Step 1: Set up the BP neural network structure; Step 2: Use a genetic algorithm to optimize and adjust the initial threshold and weights in the BP neural network to obtain the optimal initial threshold and weight values ​​for the BP neural network; Step 3: Train the GA-BP neural network to enable it to classify data that affects detection performance into five levels: "very good", "good", "average", "poor" and "very poor". Step 4: Input the test data into the GA-BP neural network, obtain the classification results through the network's predictions, and evaluate the detection performance of sonar in the current marine environment based on these classification results.

[0016] This embodiment addresses the issue of marine environmental factors influencing the evaluation of sonar detection effectiveness. It abstracts this as a nonlinear problem and uses a backpropagation (BP) neural network to solve it. Evaluation data is mapped to five categories: "Very Good," "Good," "Average," "Poor," and "Very Poor." To address the problems of slow convergence speed, difficulty in ensuring convergence to the global optimum, and insufficient determinism in the network structure of BP neural networks, this paper employs a BP neural network evaluation method based on genetic algorithm optimization. By introducing a genetic algorithm, the aim is to improve the convergence efficiency of the neural network, enhance its global search capability, and optimize the network structure design, thereby achieving accurate evaluation of sonar detection effectiveness.

[0017] In this embodiment, after setting up the BP network structure, the genetic algorithm begins to adjust the thresholds and weights in the network. After adjusting to appropriate values, the network can be trained and used for prediction. The determination of individual encodings in the genetic algorithm population and the number of parameters to be adjusted are both influenced by the sample input and output features. During this process, the initial weights and thresholds of the BP network become the main targets of the genetic algorithm optimization. The number of weights and thresholds can be derived from the network construction. The weights and thresholds of the neural network are assigned initial values ​​within a specific range. These initial values ​​are crucial to the network's training effect, but setting them accurately is quite challenging. This paper uses a genetic algorithm for initial value optimization, as shown in the attached diagram. Figure 1 As shown.

[0018] To assess the impact of marine hydrological factors on sonar detection performance and provide valuable reference for the use of sonar equipment, this embodiment proposes a neural network-based evaluation method. The core idea of ​​this method is to evaluate the degree of influence of different marine hydrological conditions on sonar detection performance by analyzing the usage of sonar equipment in a marine environment. The trained neural network is used to classify detection performance into five levels: "very good," "good," "average," "poor," and "very poor." In this way, through the learning ability of the neural network, the actual impact of marine meteorological and hydrological factors on equipment use can be determined.

[0019] To achieve this goal, a GA-BP neural network was trained using data from the National Center for Environmental Information (NCEI) to enable it to classify data affecting detection performance. Then, test data was input into the BP neural network optimized by a genetic algorithm, and the classification results were obtained through the network's prediction function. Finally, based on these classification results, the detection performance of sonar in the current marine environment can be evaluated.

[0020] This method not only allows for a more accurate assessment of sonar detection effectiveness, but also provides a more scientific and practical reference for the use of acoustic detection equipment.

[0021] To verify the effectiveness of the neural network-based performance evaluation method, 247 sets of marine hydrological data were compiled from the ICOADS (International Comprehensive Ocean-Atmosphere Data Set) dataset of the National Center for Environmental Information (NCEI). Partial data is provided to validate the effectiveness of this evaluation method, and is detailed in Table 1 below. This approach more clearly demonstrates the practical application effects of the method.

[0022] (1) Setting up the neural network In this paper, the validation example sample contains five input parameters: WIND_SPEED, PRES_WX, SEA_SURF_TEMP, WAVE_HGT, and SWELL_HGT. These input parameters correspond to five output categories: {1, 2, 3, 4, 5}. Validating this example allows for a deeper investigation into the relationship between these input parameters and the output categories, leading to more meaningful conclusions.

[0023] The meanings of the above parameters are as follows: WIND_SPEED — Wind speed, in units of 0.1 m / s or knots, depending on user specifications (standard or metric options), decimals omitted.

[0024] PRES_WX — Current weather. Values ​​[00-49] indicate no precipitation at the time of observation, [50-59] indicate light drizzle, [60-69] indicate more rain than drizzle, [70-79] indicate solid precipitation that is not showers, and [80-99] indicate heavy rainfall or precipitation with thunderstorms currently or in the near future.

[0025] SEA_SURF_TEMP — Sea surface temperature, in units of 0.1℃ or 0.1℉, depending on user specifications (standard or metric options).

[0026] WAVE_HGT — Wave height, measured in half a meter (i.e., 1 = 0.5 meters, 2 = 1 meter, etc.) or feet.

[0027] SWELL_HGT — Wave height, swell height, in half a meter (i.e., 1=0.5m, 2=1m, etc.) or feet.

[0028] Category 1: "Very good", 2: "Good", 3: "Average", 4: "Poor", 5: "Very poor".

[0029] Table 1. Data on marine environmental meteorological and hydrological elements

[0030] In this embodiment, the number of hidden layer neurons was set to 6, thus constructing a BP neural network with a 5-6-5 structure. This means the network has 5 input layer nodes, 6 hidden layer nodes, and 5 output layer nodes. The complexity of the network determines the number of its parameters, i.e., it contains numerous weights and thresholds. To determine the number of parameters that the genetic algorithm needs to optimize, the total number of these weights and thresholds needs to be calculated in detail. Furthermore, 247 sets of marine hydrological element data were randomly divided; some data was used to train the network, and the remainder was used as prediction samples to test the network's performance. There were 11 thresholds and 60 weights, for a total of 71.

[0031] The network parameters for this example are set as follows: learning rate of 0.01, target error of 0.000001, and maximum number of iterations of 1000.

[0032] (2) Implementation In the process of optimizing neural networks using genetic algorithms, the core steps include population initialization, selection of selection operators, design of crossover operators, implementation of mutation operators, and setting of the fitness function. These elements together constitute the core process of optimizing neural networks using genetic algorithms, ensuring that the optimized neural network can classify samples more accurately.

[0033] During the population initialization phase, it should be noted that this embodiment actually uses real number encoding as the encoding method. Compared with binary encoding, real number encoding facilitates genetic search in a larger space. As for the neural network structure involved in this example, it is specifically a 5-6-5 type, and the number of weights and thresholds is detailed in the table below.

[0034] Table 2 Number of weights and thresholds

[0035] The fitness function uses the `gabpEval` function from the goat genetic algorithm toolbox to decode the fitness value. The selection operator employs a sorting selection function based on a normalized geometric distribution. The crossover operator uses arithmetic crossover. Offspring are generated by combining the genes of the parents using a weighted average formula. Under the action of the mutation operator, the number of genes will mutate with a certain probability; this process is achieved by randomly selecting specific genes. When the selected gene's code is 1, it is changed to 0; if its code is 0, it is changed to 1 accordingly. This mutation mechanism ensures the diversity and flexibility of the genetic algorithm during the search process.

[0036] In this example, the genetic algorithm is set to 50 generations and a population size of 10 individuals. These parameter settings aim to ensure the algorithm converges within a reasonable timeframe while maintaining population diversity to find the optimal solution.

[0037] The simulation results of the performance evaluation algorithm based on the GA-BP neural network are shown in the attached figure.

[0038] Appendix Figure 2 To be continued Figure 3 The diagram shows the performance evaluation results of the optimized BP neural network and its corresponding error confusion matrix. Figure 2 The fitness change curves when applying the genetic algorithm are shown, revealing that the number of iterations falls within the range of (0,5). (Appendix) Figure 3The prediction results based on the GA-BP neural network training set are shown, achieving an accuracy of 99.3976%. This indicates that the adopted GA-BP neural network structure, parameter settings, and training strategy are very suitable for the current dataset and can effectively capture and learn complex patterns and relationships in the data. (Appendix) Figure 4 The prediction results on the test set are shown, with an accuracy of 97.5309%, demonstrating the model's good generalization ability. (Attached) Figure 5 and 6 The confusion matrices of the training and test sets for a BP neural network optimized using a genetic algorithm are shown below. Figure 7 To be continued Figure 10 The graphs show the performance evaluation results of the BP neural network without optimization, along with its corresponding error confusion matrix. These graphs clearly reflect the performance of the neural network and the error distribution during the evaluation process.

[0039] from Figure 7 The results show that the prediction results of the training set based on the BP neural network are about 7% lower than those based on the GA-BP neural network, which demonstrates the effectiveness of the genetic algorithm in optimizing neural network parameters and structure.

[0040] from Figure 8 The results show that the BP neural network has a 7% lower accuracy rate than the GA-BP neural network on the test set, which means that the GA-BP neural network has stronger generalization ability and is better able to adapt to new data.

[0041] By comparing simulation results using two different algorithms for performance evaluation, the impact of optimization methods on evaluation effectiveness can be clearly observed. After optimization of the BP neural network using a genetic algorithm, its evaluation accuracy reached 97.5309%. However, when the BP neural network was evaluated without genetic algorithm optimization, its performance evaluation accuracy dropped to 90.1235%. This fully demonstrates the importance of the genetic algorithm in improving the performance of BP neural networks, and the optimized algorithm exhibits superior performance evaluation. Therefore, it is evident that the classification accuracy of the BP neural network is significantly improved after optimization using a genetic algorithm, and simulation comparison analysis shows that the GA-BP neural network performs better.

[0042] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for evaluating sonar detection effectiveness based on GA-BP neural network, characterized in that... Includes the following steps: Step 1: Set up the BP neural network structure; Step 2: Use a genetic algorithm to optimize and adjust the initial threshold and weights in the BP neural network to obtain the optimal initial threshold and weight values ​​for the BP neural network; Step 3: Train the GA-BP neural network to enable it to classify data that affects detection performance into five levels: "very good", "good", "average", "poor" and "very poor". Step 4: Input the test data into the GA-BP neural network, obtain the classification results through the network's predictions, and evaluate the detection performance of sonar in the current marine environment based on these classification results.

Images