Highway subgrade disease detection method and system based on improved probabilistic neural network

Abstract

The invention discloses a highway subgrade disease detection method and system based on an improved probabilistic neural network, and belongs to the application of machine learning in highway inspection and maintenance, and the method comprises the steps: constructing a training set and a test set according to historical disease data of a highway subgrade; a probabilistic neural network is constructed, the probabilistic neural network comprises an input layer, a mode layer, a summation layer and an output layer, and a locust optimization algorithm GOA is selected to optimize smoothing factors of the probabilistic neural network; and inputting the training set and the test set into the optimized probabilistic neural network for training, inputting a to-be-tested highway subgrade monitoring index into the trained probabilistic neural network in actual prediction, and outputting to obtain the prediction of the highway subgrade disease type. The expressway roadbed monitoring index condition is comprehensively and effectively evaluated by adopting the probabilistic neural network optimized by the locust optimization algorithm, and the expressway roadbed monitoring index condition evaluation method has the advantages of higher efficiency and analysis accuracy, strong classification capability, less parameter setting and high fault tolerance.

Description

Technical Field

[0001] The present invention discloses a method for detecting highway subgrade diseases based on an improved probabilistic neural network, belonging to the application of artificial intelligence in highway detection and maintenance technology. Background Technique

[0002] With the rapid growth of the economy, the construction of expressways in China has also developed rapidly. The subgrade is a key part of the expressway structure. In daily operation, various factors may cause different degrees of diseases in the expressway subgrade, such as cracks, subsidence, retaining wall offset, etc. These diseases greatly affect the expressway structure and endanger traffic safety. Doing a good job in the preventive assessment of the subgrade is of extremely important significance for expressway maintenance, which helps to timely discover and properly handle problems, prevent and repair them at the initial stage of the disease, and thus effectively reduce the occurrence of expressway accidents.

[0003] In view of the complexity and diversity of the causes of expressway subgrade diseases, the probabilistic neural network (PNN) method can be used to comprehensively and effectively evaluate the condition of the expressway subgrade. However, in different application backgrounds, the selection of the smoothing factor σ for the conventional probabilistic neural network often depends on experience or default values, resulting in a significant reduction in the performance of the model. Summary of the Invention

[0004] The technical problem solved by the present invention is: aiming at the problem that the traditional detection of highway subgrade diseases relies on manual intervention, the process is cumbersome and lacks accurate analysis, a method for detecting highway subgrade diseases based on an improved probabilistic neural network is provided.

[0005] The present invention is implemented by adopting the following technical solutions:

[0006] The present invention first discloses a method for detecting highway subgrade diseases based on an improved probabilistic neural network, including the following steps:

[0007] S1. According to the historical disease data of the highway subgrade, collect multiple groups of highway subgrade detection indexes corresponding to the historical disease types according to the historical disease types;

[0008] S2. Based on the historical disease types and the collected highway subgrade detection indexes, construct a training set and a test set, and perform normalization preprocessing on the data in the training set and the test set;

[0009] S3. Input the training set and the test set into the optimized probabilistic neural network for model training. The input samples of the training set and the test set include highway subgrade detection indexes, and the output result is the predicted disease type; during the model training process, construct a fitness function according to the error rate of the output results of the training set and the test set, and optimize the smoothing factor of the probabilistic neural network through the grasshopper optimization algorithm GOA;

[0010] S4. When actually detecting highway subgrade diseases, input the monitoring indicators of the highway subgrade to be measured into the trained probabilistic neural network, and output the prediction of the highway subgrade disease type obtained.

[0011] In a method for detecting highway subgrade diseases based on an improved probabilistic neural network according to the present invention, further, the historical highway subgrade disease types include subgrade pumping and mud gushing, subgrade cracking, uneven subgrade settlement, subgrade frost heave, subgrade deformation, and subgrade landslide.

[0012] In a method for detecting highway subgrade diseases based on an improved probabilistic neural network according to the present invention, further, the highway subgrade monitoring indicators include subgrade temperature, subgrade moisture content, subgrade stress, subgrade strain, disease depth, and crack span. All the above highway subgrade monitoring indicators are collected for each disease type.

[0013] In a method for detecting highway subgrade diseases based on an improved probabilistic neural network according to the present invention, further, the training set accounts for 80% of the overall historical disease data, and the test set accounts for 20% of the overall historical disease data.

[0014] In a method for detecting highway subgrade diseases based on an improved probabilistic neural network according to the present invention, further, the training set and the test set perform normalization preprocessing on the data according to the Z-score normalization method, and the formula is as follows:

[0015]

[0016] where y is the original data of the highway subgrade monitoring indicators in the training set and the test set; μ is the mean value of the highway subgrade monitoring indicator data in the training set and the test set; δ is the standard deviation of the highway subgrade monitoring indicator data in the training set and the test set; y std is the data after normalization of the training set and the test set.

[0017] In a method for detecting highway subgrade diseases based on an improved probabilistic neural network according to the present invention, further, the probabilistic neural network includes an input layer, a pattern layer, a summation layer, and an output layer;

[0018] The number of neurons in the input layer of the probabilistic neural network corresponds to the dimension of the input samples of the training set and the test set;

[0019] The number of neuron nodes in the pattern layer is equal to the number of input samples of the training set and the test set. The input-output relationship of the neurons in the pattern layer is as follows:

[0020]

[0021] where, y std is the data after normalization of the input sample, is the output of the pattern layer, y j is the feature vector of the j-th sample in the training set, σ is the smoothing factor, d is the dimension of the feature vector, ||y std -y j || is the Euclidean distance between y j and y std ;

[0022] The total number of neurons in the summation layer is equal to the number of disease types of the highway subgrade, and the classification results of each input sample are output and weighted averaged. The probability density that the input sample belongs to each disease type is calculated by the following formula:

[0023]

[0024] Among them, P j represents the probability density that the input sample belongs to the j-th type of disease, φ jk(y) represents the output of the k-th neuron in the j-th type of disease, L j represents the number of samples of the j-th type of disease;

[0025] The output layer selects the neuron with the maximum probability density and outputs 1, and the rest output 0;

[0026] The probability neural network outputs the predicted disease type according to the highway subgrade disease type corresponding to the neuron that outputs 1.

[0027] In an improved probability neural network-based highway subgrade disease detection method of the present invention, further, the optimization process of the grasshopper optimization algorithm GOA for the probability neural network is as follows:

[0028] S31. Initialize the grasshopper population, and use the grasshopper position information to represent the value of the smoothing factor in the probability neural network. Considering the influence of the social role of the grasshoppers on the grasshopper position, the position information of the grasshoppers is determined by the following formula:

[0029]

[0030] Among them, X i is the global position set position of the i-th grasshopper, N is the total number of individuals in the grasshopper population, s() is the social function, x i and x j are the individual positions of the i-th and j-th grasshoppers in the d-dimensional space, d ij is the distance between the i-th grasshopper and the j-th grasshopper, and this distance is normalized to the interval [1, 4] through standardization processing, ub d and lb d are the upper and lower bounds of the social function s() in the d-dimensional space respectively, T d is the best solution of the grasshopper position in the d-dimensional space so far, and c is the grasshopper position update iteration parameter;

[0031] S32. Calculate the fitness of each locust through the following fitness function:

[0032] f(X i ) = argmin{R train +R predict},

[0033] f(X i ) is the fitness of the current locust position X i , R train is the error rate of the output result of the probabilistic neural network for the training set, and R predict is the error rate of the output result of the probabilistic neural network for the test set;

[0034] S33. Repeat the above steps for iteration, and update the iterative attenuation parameter c by combining the fitness and the locust position,

[0035]

[0036] where c max is the initial maximum attenuation parameter, f(X i ) is the fitness of the current locust, f best is the global optimal fitness so far, d(t) is the distance from the current locust position to the current best position, and D max is the maximum boundary of the solution space;

[0037] S34. Stop the optimization when the set maximum number of iterations is reached. Select the minimum value from the iterative fitness as the best fitness, and obtain the locust position information corresponding to the best fitness as the optimized value of the smoothing factor σ of the probabilistic neural network.

[0038] In a method for detecting highway subgrade diseases based on an improved probabilistic neural network according to the present invention, further, an F1-score model is selected to evaluate the prediction effect of the output result of the probabilistic neural network. The specific steps are as follows:

[0039] Input the test set data into the probabilistic neural network to obtain the predicted disease type. Count the output predicted disease type results as an A×A confusion matrix. The rows of the confusion matrix represent the actual disease types, the columns of the confusion matrix represent the predicted disease types, and each element in the confusion matrix represents the number of predicted types output by the probabilistic neural network under the data of the disease type in that row;

[0040] Select one of the disease types as a specific category as the true positive (TP), which represents the number of samples that actually belong to the selected disease type and are correctly predicted. The number of samples that do not actually belong to this category but are misjudged as this specific category is used as the false positive (FP), and the number of samples that actually belong to this category but are misjudged as other categories is used as the false negative (FN). Extract the values of the true positive (TP), false positive (FP), and false negative (FN) from the confusion matrix respectively;

[0041] Calculate the value of F1 from the recall rate and precision through the following formula,

[0042]

[0043] where R represents the recall rate, P represents the precision, and the closer the value of F1 is to 1, the better the prediction effect of the model.

[0044] The present invention also discloses a highway subgrade disease detection system based on an improved probabilistic neural network, including a monitoring module and a prediction module; the monitoring module includes a sensing device for collecting highway subgrade monitoring indicators, and the highway subgrade detection indicators include subgrade temperature, subgrade moisture content, subgrade stress, subgrade strain, disease depth, and crack span; the prediction module includes a memory, a processor, and a computer program stored on the memory and executable on the processor. The processor receives the subgrade detection indicators collected by the monitoring module and implements the steps of the above-mentioned highway subgrade disease detection method of the present invention when executing the computer program.

[0045] In a highway subgrade disease detection system based on an improved probabilistic neural network of the present invention, further, the sensing device includes a temperature sensor, a humidity sensor, a stress sensor, a strain sensor, a subgrade pressure sensor, a data acquisition board, and a single-point settlement gauge.

[0046] In view of the characteristics of multiple highway subgrade disease types and complex influencing factors in the present invention, a probabilistic neural network model (PNN) is selected to realize the function of a non-linear learning algorithm through linear learning, and predict and evaluate highway subgrade diseases under multiple influencing factors. The probabilistic neural network model has better stability compared with other neural network models.

[0047] Considering that the smoothing factor σ in the Probabilistic Neural Network (PNN) has a great influence on the output results of the PNN, and there are problems such as inappropriate selection of the smoothing factor, the model being vulnerable to outliers, and the error between the prediction result and the actual result being too large when directly using the PNN for evaluation, the present invention uses the Grasshopper Optimization Algorithm (GOA) to improve the PNN. By automatically optimizing the smoothing factor σ, the optimal model parameters are found in the application of predicting the types of highway subgrade diseases, solving problems such as overfitting, underfitting, low computational efficiency, and inaccurate prediction results, increasing the convergence of the model, improving the accuracy of the predicted evaluation results of the PNN after training by adjusting the value of the smoothing factor σ, and the optimized PNN provides stronger adaptability and optimization ability.

[0048] When the present invention optimizes the smoothing factor σ of the PNN using the GOA, it adjusts the positions of the grasshoppers according to the error rate of the output results of the training set and the test set, so that the population moves in the direction of a lower error rate. Using the sum of the error rates as the fitness function reduces the influence caused by one of the training set or the test set being too low or too high. If the error rates of both the training set and the test set are relatively high, the GOA will increase the complexity of the model during the optimization process to better fit the data. This makes the optimization process not only target a specific data set, but optimize the overall performance of the model, making it more adaptable to unknown data in applications.

[0049] In summary, the method and system for highway subgrade disease detection based on the improved Probabilistic Neural Network provided by the present invention, based on the complexity and diversity of the causes of highway subgrade diseases, uses the PNN optimized by the GOA to comprehensively and effectively evaluate the status of highway subgrade monitoring indicators. It has higher efficiency and analysis accuracy compared with traditional manual experience judgment and prediction. Compared with traditional machine learning methods, the improved PNN of the present invention only requires a small amount of data to complete training and has the advantages of strong classification ability, few parameter settings, and high fault tolerance.

[0050] The following further describes the present invention in conjunction with the accompanying drawings and specific embodiments. Description of the Drawings

[0051] Figure 1 It is a flowchart of the method for highway subgrade disease detection based on the improved Probabilistic Neural Network of the present invention.

[0052] Figure 2 It is a schematic diagram of the Probabilistic Neural Network model adopted by the present invention.

[0053] Figure 3 It is a flowchart of the optimization process of the smoothing factor of the Probabilistic Neural Network model by the Grasshopper Optimization Algorithm in the present invention. Specific Embodiments

[0054] Embodiment

[0055] Refer to Figure 1 , the specific process of the highway subgrade disease detection method based on the improved probabilistic neural network of the present invention is as follows:

[0056] S1. According to the historical disease data of the highway subgrade, collect multiple groups of highway subgrade detection indexes corresponding to the disease types according to the historical disease types.

[0057] Specifically, the historical disease types of the highway subgrade include subgrade pumping and mud gushing, subgrade cracking, uneven settlement of subgrade, subgrade frost heave, subgrade deformation, and subgrade landslide. The highway subgrade monitoring indexes include subgrade temperature, subgrade moisture content, subgrade stress, subgrade strain, disease depth, and crack span. All of the above highway subgrade monitoring indexes are collected for each disease type, and all these indexes are used for evaluation of all disease types because it cannot be guaranteed that any one of the monitoring indexes is completely irrelevant to the disease type. Considering all highway subgrade monitoring indexes for each disease type will not affect the correctness of the final evaluation result.

[0058] S2. Construct a training set and a test set based on the historical disease types and the collected highway subgrade detection indexes, and perform normalization preprocessing on the data in the training set and the test set.

[0059] Specifically, the training set accounts for 80% of the overall historical disease data, and the test set accounts for 20% of the overall historical disease data. And all types in the historical disease types should be covered in the training set to ensure that the model can identify and train all disease types.

[0060] The training set and the test set perform normalization preprocessing on the data according to the Z-score standardization method, and the formula is as follows:

[0061]

[0062] where y is the original data of the highway subgrade monitoring indexes in the training set and the test set; μ is the mean value of the highway subgrade monitoring index data in the training set and the test set; δ is the standard deviation of the highway subgrade monitoring index data in the training set and the test set; y std is the data after standardization of the training set and the test set.

[0063] S3. Input the training set and the test set into the optimized probabilistic neural network for model training. The input samples X1, X2... X M of the training set and the test set are the highway subgrade detection indexes collected for each disease type, and the output results Y1, Y2... Y M are the predicted disease types.

[0064] Such as Figure 2As shown, the probabilistic neural network includes an input layer, a pattern layer, a summation layer, and an output layer. After the input samples enter the input layer of the probabilistic neural network, they are converted into input vectors, and the number of neurons in the input layer corresponds to the dimensions of the input samples in the input training set and the test set.

[0065] The number of neuron nodes in the pattern layer is equal to the number of input samples in the input training set and the test set. The input-output relationship of the nth neuron in the mth category corresponding to the input samples in the pattern layer is as follows:

[0066]

[0067] Among them, y std is the data feature vector after the input samples are standardized, is the output of the pattern layer, y j is the feature vector of the jth sample in the training set, σ is the smoothing factor, d is the dimension of the feature vector, ||y std -y j || is the Euclidean distance between y j and y std .

[0068] The total number of neurons in the summation layer is equal to the number of types of diseases of the highway subgrade. The classification results of each input sample are output and weighted averaged, and the probability density that the input sample belongs to each disease type is calculated by the following formula:

[0069]

[0070] Among them, P j represents the probability density that the input sample belongs to the jth type of disease, φ jk(y) represents the output of the kth neuron in the jth type of disease, and L j represents the number of samples of the jth type of disease.

[0071] The output layer selects the neuron with the maximum probability density and outputs 1, and the rest output 0. The probabilistic neural network outputs the predicted disease type according to the type of highway subgrade disease corresponding to the neuron that outputs 1.

[0072] Select the F1-score model to evaluate the prediction effect of the output result of the probabilistic neural network. The specific steps are as follows:

[0073] Input the samples of the test set into the probabilistic neural network to obtain the predicted disease type, and compare the label of the predicted disease type with the actual disease type label.

[0074] The test set data is input into the probabilistic neural network to obtain the predicted disease types. The output results of the predicted disease types are statistically counted as a confusion matrix of A×A, where A represents the total number of possible disease types output by the test set, which is a summary based on the output prediction results of several groups of test sets. The rows of the confusion matrix represent the actual disease types, the columns represent the predicted disease types, and each element in the confusion matrix represents the number of predicted types output by the probabilistic neural network under the data of the disease type in that row;

[0075] One of the disease types is selected as a specific category as the true positive TP, which represents the number of samples that actually belong to the selected disease type and are correctly predicted. The number of samples that do not actually belong to this category but are misjudged as this specific category is used as the false positive FP, and the number of samples that actually belong to this category but are misjudged as other categories is used as the false negative FN. The values of true positive TP, false positive FP, and false negative FN are extracted from the confusion matrix respectively.

[0076] The value of F1 is calculated from the recall rate and precision rate by the following formula,

[0077]

[0078] where R represents the recall rate, P represents the precision rate, and the closer the value of F1 is to 1, the better the prediction effect of the model. This model is the harmonic mean of the recall rate R and the precision rate P, which can more effectively reduce the influence caused by small data quantity or uneven disease sample categories and improve the prediction accuracy. The prediction effect of the output results of the probabilistic neural network is evaluated through the above F1-score model. The model performance is improved by optimizing the smoothing factor or normalizing the data, etc. This model is the harmonic mean of the recall rate R and the precision rate P, which can more effectively reduce the influence caused by small data quantity or uneven disease sample categories and improve the prediction accuracy.

[0079] During the model training process, a fitness function is constructed based on the error rates of the output results of the training set and the test set, and the smoothing factor of the probabilistic neural network is optimized through the Grasshopper Optimization Algorithm (GOA).

[0080] Combined with reference to Figure 3 , the optimization process of the probabilistic neural network by the Grasshopper Optimization Algorithm (GOA) is as follows:

[0081] S31. Initialize the grasshopper population. The position information of the grasshoppers represents the value of the smoothing factor in the probabilistic neural network. Considering the influence of the social role of the grasshoppers on their positions, the position information of the grasshoppers is determined by the following formula:

[0082]

[0083] where X iis the set of global positions of the \(i\)-th locust, \(N\) is the total number of individuals in the locust population, and \(N = 60\) in this embodiment. \(s()\) is the social function, and \(x\) i and \(x\) j are the individual positions of the \(i\)-th and \(j\)-th locusts in the \(d\)-dimensional space, and \(d\) ij is the distance between the \(i\)-th locust and the \(j\)-th locust. This distance is normalized to the interval \([1, 4]\) through standardization processing. \(ub\) d and \(lb\) d are the upper and lower bounds of the social function \(s()\) in the \(d\)-dimensional space, respectively. \(T\) d is the best solution of the locust positions in the \(d\)-dimensional space so far, and \(c\) is the locust position update iteration parameter. The position \(X_i\) of each locust represents the value of the smoothing factor, and the smoothing factor affects the correctness of the output results of the training set and the test set. Therefore, the optimization of the PNN by GOA is carried out during the training process of the model, optimizing while training.

[0084] S32. To reduce the probability of calculation errors in the probabilistic neural network model, the error rates of the training set and the test set output results are selected as \(\epsilon\). The fitness of each locust is calculated through the following fitness function:

[0085] \(f(X\) i ) = argmin{R train + R predict},

[0086] \(f(X\) i ) is the fitness of the current locust position \(X\) i . \(R\) train is the error rate of the probabilistic neural network for the training set output result, and \(R\) predict is the error rate of the probabilistic neural network for the test set output result;

[0087] S33. Repeat the above steps for iteration, and combine the fitness and the locust position update iteration parameter \(c\),

[0088]

[0089] where \(c\) max is the initial maximum attenuation parameter, taking 2. \(f(X\) i ) is the fitness of the current locust, and \(f\) best is the global optimal fitness so far. \(d(t)\) is the distance from the current locust position to the best position so far, and \(D\) max is the maximum boundary of the solution space, taking 3.

[0090] S34. Stop the optimization when the iteration reaches the set maximum number of iterations. Select the minimum value from the iterative fitness as the best fitness, and obtain the locust position information corresponding to the best fitness as the optimized value of the smoothing factor \(\sigma\) of the probabilistic neural network.

[0091] S4. When actually detecting highway subgrade diseases, the monitored highway subgrade monitoring indicators obtained by real-time monitoring are input into the trained probabilistic neural network, and the prediction of the highway subgrade disease type is output.

[0092] This embodiment also discloses a system applying the above highway subgrade disease detection method. The system includes a monitoring module and a prediction module; the monitoring module includes a sensing device for collecting highway subgrade monitoring indicators. The highway subgrade detection indicators include subgrade temperature, subgrade water content, subgrade stress, subgrade strain, disease depth, and crack span. The sensing devices used include a temperature sensor, a humidity sensor, a stress sensor, a strain sensor, a subgrade pressure sensor, a data acquisition board, and a single-point settlement gauge. The prediction module includes a memory, a processor, and a computer program stored on the memory and executable on the processor. The processor receives the subgrade detection indicators collected by the monitoring module and implements the steps of the above highway subgrade disease detection method of this embodiment when executing the computer program.

[0093] The memory is a computer-readable storage medium, which can be an internal storage unit of the above-mentioned software and hardware device, such as the hard disk or memory of the controller. The readable storage medium can also be an external storage device of the controller, such as a plug-in hard disk equipped on the controller, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. Further, the readable storage medium can also include both the internal storage unit of the controller and the external storage device. The readable storage medium is used to store the computer program and other programs and data required by the controller. The readable storage medium can also be used to temporarily store the data that has been output or will be output.

[0094] In this article, the orientation or positional relationship indicated by the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", "vertical", "horizontal", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of clearly expressing the technical solution and description, so it cannot be understood as a limitation to the present invention.

[0095] In this article, the term "comprising", "including" or any other variation thereof is intended to cover non-exclusive inclusion, and in addition to the listed elements, it may also include other elements not specifically listed.

[0096] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A highway subgrade disease detection method based on an improved probabilistic neural network, characterized in that: It includes the following steps: S1. According to the historical disease data of highway subgrades, collect multiple groups of highway subgrade detection indexes corresponding to the disease types by classifying the historical disease types; S2. Based on the historical disease types and the collected highway subgrade detection indexes, construct a training set and a test set, and perform normalization preprocessing on the data in the training set and the test set; S3. Input the training set and the test set into the optimized probabilistic neural network for model training. The input samples of the training set and the test set include highway subgrade detection indexes, and the output result is the predicted disease type; During the model training process, construct a fitness function according to the error rate of the output results of the training set and the test set, and optimize the smoothing factor of the probabilistic neural network through the Grasshopper Optimization Algorithm (GOA); S4. When actually detecting highway subgrade diseases, use the monitoring indexes of the highway subgrade to be measured to input the trained probabilistic neural network, and output the prediction of the highway subgrade disease type.

2. The method for detecting highway subgrade diseases based on an improved probabilistic neural network according to claim 1, characterized in that: The historical disease types of the highway subgrade include subgrade pumping and mud gushing, subgrade cracking, uneven settlement of subgrade, subgrade frost heave, subgrade deformation, and subgrade landslide.

3. The method for detecting highway subgrade diseases based on an improved probabilistic neural network according to claim 2, wherein: The highway subgrade monitoring indexes include subgrade temperature, subgrade moisture content, subgrade stress, subgrade strain, disease depth, and crack span. All the above highway subgrade monitoring indexes are collected for each disease type.

4. A highway subgrade disease detection method based on an improved probabilistic neural network according to claim 1, characterized in that: The training set accounts for 80% of the overall historical disease data, and the test set accounts for 20% of the overall historical disease data.

5. A highway subgrade disease detection method based on an improved probabilistic neural network according to claim 4, characterized in that: The training set and the test set perform normalization preprocessing on the data according to the Z-score standardization method. The formula is as follows: Among them, y is the original data of the highway subgrade monitoring index in the training set and the test set; μ is the mean value of the highway subgrade monitoring index data in the training set and the test set; δ is the standard deviation of the highway subgrade monitoring index data in the training set and the test set; y std is the data after standardization of the training set and the test set.

6. A highway subgrade disease detection method based on an improved probabilistic neural network according to claim 1, characterized in that: The probabilistic neural network includes an input layer, a pattern layer, a summation layer, and an output layer; The number of neurons in the input layer of the probabilistic neural network corresponds to the dimension of the input samples of the training set and the test set; The number of neuron nodes in the pattern layer is equal to the number of input samples of the training set and the test set. The input-output relationship of the neurons in the pattern layer is as follows: Among them, y std is the data after standardizing the input samples, is the output of the pattern layer, y j is the feature vector of the j-th sample in the training set, σ is the smoothing factor, d is the dimension of the feature vector, ||y std - y j || is the Euclidean distance between y j and y std ; The total number of neurons in the summation layer is equal to the number of disease types of the highway subgrade, output the classification results of each input sample and perform weighted averaging, and calculate the probability density of each input sample belonging to each disease type through the following formula: where, P j represents the probability density that the input sample belongs to the j-th type of disease, and φ jk(y) represents the output of the k-th neuron in the j-th type of disease, and L j represents the number of samples of the j-th type of disease; The output layer selects the neuron with the maximum probability density and outputs 1, and the others output 0; The probabilistic neural network outputs the predicted disease type according to the highway subgrade disease type corresponding to the neuron that outputs 1.

7. A method for detecting highway subgrade diseases based on an improved probabilistic neural network according to claim 6, characterized in that: The optimization process of the probabilistic neural network by the Grasshopper Optimization Algorithm (GOA) is as follows: S31. Initialize the grasshopper population. Represent the value of the smoothing factor in the probabilistic neural network by the grasshopper position information. Considering the influence of the social role of the grasshoppers on the grasshopper position, determine the grasshopper position information through the following formula: Among them, X i is the global position set position of the i-th locust, N is the total number of individuals in the locust population, s() is the social function, x i and x j are the individual positions of the i-th and j-th locusts in the d-dimensional space, d ij is the distance between the i-th locust and the j-th locust, and this distance is normalized to the interval [1, 4] through standardization processing, ub d and lb d are the upper and lower bounds of the social function s() in the d-dimensional space respectively, T d is the best solution of the locust position in the d-dimensional space so far, and c is the iteration parameter for updating the locust position; S32. Calculate the fitness of each grasshopper through the following fitness function: f(X i ) = argmin{R train + R predict}, f(X i ) is the fitness of the current locust position X i , R train is the error rate of the output result of the probabilistic neural network for the training set, R predict is the error rate of the output result of the probabilistic neural network for the test set; S33. Repeat the above steps for iteration, and update the iteration decay parameter c in combination with the fitness and the grasshopper position; Among them, c max is the initial maximum attenuation parameter, f(X i ) is the fitness of the current locust, f best is the global optimal fitness so far, d(t) is the distance from the current locust position to the current best position, D max is the maximum boundary of the solution space; S34. Stop the optimization when iterating to the set maximum number of iterations. Select the minimum value from the iterated fitness as the best fitness, and obtain the optimized value of the smoothing factor σ of the probabilistic neural network as the grasshopper position information corresponding to the best fitness.

8. A highway subgrade disease detection method based on an improved probabilistic neural network according to claim 1, characterized in that: Select the F1-score model to evaluate the prediction effect of the output results of the probabilistic neural network. The specific steps are as follows: Input the test set data into the probabilistic neural network to obtain the predicted disease types, and count the output predicted disease type results as an A×A confusion matrix. The rows of the confusion matrix represent the actual disease types, the columns represent the predicted disease types, and each element in the confusion matrix represents the number of predicted types output by the probabilistic neural network under the data of the disease type in that row. Select one of the disease types as a specific category as the true positive TP, which represents the number of samples that actually belong to the selected disease type and are correctly predicted. The number of samples that do not actually belong to this category but are misjudged as this specific category is used as the false positive FP, and the number of samples that actually belong to this category but are misjudged as other categories is used as the false negative FN. Extract the values of the true positive TP, false positive FP, and false negative FN from the confusion matrix respectively. Calculate the value of F1 from the recall rate and precision through the following formula: where R represents the recall rate, P represents the precision, and the closer the value of F1 is to 1, the better the prediction effect of the model.

9. A highway subgrade disease detection system based on an improved probabilistic neural network, characterized in that: It includes a monitoring module and a prediction module; The monitoring module includes sensing devices for collecting highway subgrade monitoring indicators, and the highway subgrade detection indicators include subgrade temperature, subgrade moisture content, subgrade stress, subgrade strain, disease depth, and crack span; The prediction module includes a memory, a processor, and a computer program stored on the memory and executable on the processor. The processor receives the subgrade detection indicators collected by the monitoring module and implements the steps of any one of the highway subgrade disease detection methods in claims 1-8 when executing the computer program.

10. A highway subgrade disease detection system based on an improved probabilistic neural network according to claim 9, characterized in that: The sensing devices include temperature sensors, humidity sensors, stress sensors, strain sensors, subgrade pressure sensors, data acquisition boards, and single-point settlement gauges.

Images