A method for predicting height of water flowing fractured zone in fully-mechanized caving mining based on AFSA-MLP neural network model

Abstract

The present application belongs to a kind of fully mechanized caving mining water conducting fractured zone height prediction method based on AFSA-MLP neural network model in coal mine safety mining technical field, including the data of water conducting fractured zone height influencing factor and measured data, classification after standardization processing obtains training set data and test set data, the weight matrix and bias matrix of MLP neural network model are initialized according to the data after classification, the optimal weight matrix and optimal bias matrix are obtained by using AFSA algorithm to optimize MLP neural network model to obtain AFSA-MLP neural network model, the test set data is substituted into AFSA-MLP neural network model to obtain output value, then the prediction value is obtained by anti-standardization processing and precision judgment;The present application comprehensively analyzes the influencing factors of fully mechanized caving mining overburden water conducting fractured zone height, and uses AFSA intelligent algorithm to optimize the weight and bias of MLP to make up for the defects of MLP neural network in error back propagation process, to build the water conducting fractured zone height prediction model based on AFSA-MLP, further improve the prediction accuracy.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of coal mine safety mining, and particularly relates to a fully mechanized caving mining water flowing fractured zone height prediction method based on an AFSA-MLP neural network model. BACKGROUND

[0002] Coal mining will cause overburden damage, forming a water flowing fractured zone composed of a caving zone and a fractured zone, and once the water flowing fractured zone is connected to the overlying water-rich aquifer, water inrush disaster is likely to occur. The water flowing fractured zone calculation formula in the traditional "three under" specification has been unable to meet the current needs of China's coal safety mining, and targeted research needs to be carried out to consider as many factors as possible that affect the development of the water flowing fractured zone, establish a more reliable prediction formula, and improve the prediction accuracy.

[0003] At present, many scientific researchers are engaged in research related thereto, and the research on overburden water flowing fractured zone height prediction has always been the focus of mine workers, and the related achievements play an important role in overburden water flowing fractured zone height prediction. However, due to the complexity of the mine geological environment and the uncertainty of the mining conditions in China, the existing research is still in the theoretical exploration stage, and so far a reliable fully mechanized overburden water flowing fractured zone height prediction method has not been formed. SUMMARY

[0004] The purpose of the present application is to provide a fully mechanized caving mining water flowing fractured zone height prediction method based on an AFSA-MLP neural network model to solve the problems raised in the background art.

[0005] The present application achieves the above-mentioned purposes through the following technical solutions:

[0006] A fully mechanized caving mining water flowing fractured zone height prediction method based on an AFSA-MLP neural network model, comprising the following steps:

[0007] S1: Collecting a plurality of sets of index data affecting the height of the water flowing fractured zone under fully mechanized coal mining conditions and a plurality of sets of measured data of the height of the water flowing fractured zone, and after standardizing the index data and the measured data, classifying the index data according to a set proportion to obtain training set data and test set data;

[0008] S2: Initializing the weight matrix and bias matrix of the MLP neural network model according to the training set data and the measured data, optimizing the MLP neural network model using the AFSA algorithm to obtain the optimal weight matrix and the optimal bias matrix to obtain the AFSA-MLP neural network model;

[0009] S3: Substitute the test set data into the AFSA-MLP neural network model to obtain the output value, standardize the output value to obtain the predicted value, and perform error analysis on the predicted value based on the measured data to obtain the prediction accuracy of the AFSA-MLP neural network model.

[0010] As a further optimization of the present invention, the index data includes at least the following factors affecting the height of the water-conducting fracture zone: mining thickness, mining depth, working face slope length, and hard rock lithology ratio coefficient.

[0011] As a further optimization of the present invention, the standardization process in step S1 is specifically as follows: z-score standardization is performed on several sets of the index data and several sets of the measured data to form a mean of 0 and a variance of 1 for both sets of the index data and the measured data.

[0012] As a further optimization of the present invention, the initialization of the weight matrix and bias matrix of the MLP neural network model in step S2 is specifically as follows:

[0013] The MLP neural network model is constructed by using the training set data as input and the standardized measured data as output. The weight matrix and bias matrix of the MLP neural network model are obtained by the following formula.

[0014] The formula is as follows:

[0015] ;

[0016] In the formula, z j It is the output of the i-th hidden layer node, x i For the input training set data, y k Let f be the measured data, f be the ReLU activation function, and b be the value of the measured data. j b is the bias of the j-th hidden layer node. k w is the bias of the k-th output node. ij For the first The connection weights between each input data point and the j-th hidden layer; w jk Let be the connection weight from the j-th hidden layer to the k-th output node, n be the n-th hidden layer node, and m be the m-th output node.

[0017] As a further optimization of the present invention, in step S2, the weight matrix and bias matrix of the MLP neural network model are optimized using the AFSA algorithm to obtain the AFSA-MLP neural network model, including:

[0018] S2.1: according to the connection weight matrix and the bias quantity matrix, a weight vector and a bias vector are formed, and an artificial fish population is constructed based on the weight vector and the bias vector, and presetting food concentration of the artificial fish, distance between artificial fish individuals, perception distance of the artificial fish, maximum step length of the artificial fish movement and crowding factor, and after the presetting is completed, step S2.2 is executed;

[0019] S2.2: according to a target function of a position where the artificial fish is located, a target value of each artificial fish in the initial fish population of the artificial fish population is calculated, and after the artificial fish is executed in sequence to perform the behaviors of grouping, foraging and tail chasing, the state of the artificial fish is updated, and the optimal value is taken in the target value, and after the iteration number is counted, step S2.3 is executed.

[0020] S2.3: whether the iteration number reaches the total number of artificial fish and the total number of iterations is judged, if both are reached, the optimal vector matrix and the optimal bias vector matrix of the MLP neural network model are output according to the optimal value, and step S2.4 is executed, otherwise step S2.2 is continuously executed.

[0021] S2.4: the MLP neural network model is constructed according to the optimal vector matrix and the optimal bias vector matrix.

[0022] As a further optimization scheme of the application, in step S2.3, the total number of iterations is a random preset value, and the total number of iterations must be greater than the total number of artificial fish.

[0023] As a further optimization scheme of the application, in step S3, the error analysis includes:

[0024] Constructing a plurality of groups of absolute value error arrays of measured data and a plurality of groups of the predicted values of the height of the water fractured zone;

[0025] According to the absolute value error array, the maximum absolute value error, the average absolute value error and the average relative error of the measured data and the predicted values are calculated;

[0026] The maximum absolute value error, the average absolute value error and the average relative error are taken as the judgment standard of the prediction accuracy of the AFSA-MLP model.

[0027] The application has the following beneficial effects:

[0028] The application is based on comprehensive analysis of influencing factors of overburden water flowing fractured zone in fully mechanized caving mining, selects mining thickness, mining depth, working face inclined length and hard rock lithology proportion coefficient as important influencing factors of water flowing fractured zone calculation, adopts AFSA intelligent algorithm to optimize weight and bias of MLP to make up defects of MLP neural network in error back propagation process, constructs water flowing fractured zone height prediction model based on AFSA-MLP, and verifies practicability of the model through test set and engineering examples, so as to provide certain reference significance for water flowing fractured zone prediction in fully mechanized caving mining. BRIEF DESCRIPTION OF DRAWINGS

[0029] Figure 1 is a whole process schematic diagram of the application;

[0030] Figure 2 is a fitting graph of measured values and water flowing fractured zone height predicted by the MLP neural network model in the application;

[0031] Figure 3 is a fitting graph of measured values and water flowing fractured zone height predicted by the AFSA-MLP neural network model in the application. DETAILED DESCRIPTION

[0032] It is necessary to point out that the following detailed description is only used for further description of the application and cannot be understood as limitation on the protection scope of the application, and the person skilled in the art can make some non-essential improvements and adjustments to the application according to the above description.

[0033] Example 1

[0034] As shown in the accompanying drawings, the application provides a water flowing fractured zone height prediction method for fully mechanized caving mining based on an AFSA-MLP neural network model, which comprises the following steps: Figure 1 S1: collect a plurality of sets of index data influencing water flowing fractured zone height and a plurality of sets of measured data of water flowing fractured zone height under fully mechanized mining conditions of coal seams, and after standardizing the index data and the measured data, classify the index data according to a set proportion to obtain training set data and test set data;

[0035] S2: initialize weight matrix and bias matrix of the MLP neural network model according to the training set data and the measured data, optimize the MLP neural network model by using the AFSA algorithm to obtain optimal weight matrix and optimal bias matrix to obtain the AFSA-MLP neural network model;

[0036]

[0037] ​S3: the test set data is substituted into the AFSA-MLP neural network model to obtain output values, the output values are subjected to inverse standardization processing to obtain predicted values, error analysis is performed on the predicted values according to the measured data, and the prediction accuracy of the AFSA-MLP neural network model is obtained.

[0038] In the present application, the index data at least includes the following water flowing fractured zone height influencing factors:

[0039] Mining thickness: the mining thickness plays an important role in the development of the water flowing fractured zone, which will directly affect the stress redistribution and the fracture range of the roof stratum, and the greater the mining thickness, the greater the plastic range zone of the roof and the space for the falling rock stratum, thereby resulting in the increase of the height of the caving zone;

[0040] Mining depth: when the coal seam is mined, the original rock stress in the surrounding rock increases with the increase of the mining depth, according to the coal mine geological principle, the initial ground stress condition near the mining space has a great relationship with the damage range and damage strength of the surrounding rock after excavation, in general, the deeper the mine is mined, the greater the mine pressure, the more serious the damage of the overlying rock, and the more developed the water flowing fractured zone height;

[0041] Working face oblique length: the water flowing fractured zone has a greater relationship with the smaller one of the inclination length and the length along the strike of the working face size. For the fully mechanized caving face, the inclination length of the working face is smaller than the length along the strike. According to the principle of material mechanics, the bending degree of the roof rock beam increases with the span of the rock beam on the mining working face, and the probability of rock beam fracture also increases. The higher the rock beam fracture probability, the greater the development height of the water flowing fractured zone. Therefore, the working face oblique length is an important factor in the development process of the water flowing fractured zone;

[0042] Hard rock lithology proportion coefficient: when the overlying rock is hard rock, the development of the water flowing fractured zone height is greater in the mining process, on the contrary, when the overlying rock is soft rock, the development of the water flowing fractured zone height is smaller; in order to prevent the unclear value of the compressive strength of the multi-layer rock and the difficulty in quantifying the rock mass structure type, the hard rock lithology proportion coefficient is used to replace the two indexes of the compressive strength of the combined rock and the rock structure type, and the hard rock lithology proportion coefficient refers to the ratio of the hard rock to the statistical height within the statistical height above the coal seam roof, which is determined according to the following formula:

[0043] ;

[0044] In the formula: is the mining thickness; is the cumulative thickness of the hard rock within the estimated water flowing fractured zone height (generally 15 times of the mining height) of the roof; is the extremely soft roof type, is the soft roof type, It was a medium-hard roof type. It is a type of rigid top plate.

[0045] In this invention, due to the complexity and diversity of geological conditions in different mining areas of China, the collected measured values ​​of water-conducting fracture zones and influencing factors vary significantly. Directly using the collected data for model training would result in substantial training errors. Step S1, the standardization process, specifically involves z-score standardization of several sets of index data and several sets of measured data, resulting in sets with a mean of 0 and a variance of 1. Data standardization aims to eliminate errors caused by different units, inherent variations, or large numerical differences. This embodiment selects z-score standardization, processing each column of data into data with a mean of 0 and a variance of 1. Its advantage is that it is less affected by outliers. The formula is as follows:

[0046] ;

[0047] In the formula: These are sample values. It is the average value of the influencing factor data in each column. It is the standard deviation of the data for each influencing factor.

[0048] In this invention, the initialization of the weight matrix and bias matrix of the MLP neural network model in step S2 is specifically as follows:

[0049] The MLP neural network model is constructed by taking the training set data as input and the standardized measured data as output. The weight matrix and bias matrix of the MLP neural network model are obtained by the following formula.

[0050] The formula is as follows:

[0051] ;

[0052] In the formula, z j It is the output of the i-th hidden layer node, x i Given the input training set data, y k The data are from actual measurements, f is the ReLU activation function, and b is the actual data. j b is the bias of the j-th hidden layer node. k w is the bias of the k-th output node. ij For the first The connection weights between each input data point and the j-th hidden layer; w jk Let be the connection weight from the j-th hidden layer to the k-th output node, n be the n-th hidden layer node, and m be the m-th output node.

[0053] The ReLU activation function formula in this invention is as follows: .

[0054] The MLP neural network model in the application determines the weight vector and the bias vector in the training process, which learns and feedback adjusts in the training process, so as to correct the weight and bias of each neuron. In this paper, the weight and bias of the MLP are optimized by the artificial fish school algorithm (AFSA) to make up for the defects of the MLP neural network model in the error backpropagation process, and a water fractured zone height prediction model based on AFSA-MLP is constructed. In step S2, the weight matrix and the bias matrix of the MLP neural network model are optimized by the AFSA algorithm to obtain the AFSA-MLP neural network model, which includes:

[0055] S2.1: According to the connection weight matrix and the bias matrix, the weight vector and the bias vector are constructed, and the artificial fish population is constructed, and the food concentration of the artificial fish, the distance between the artificial fish individuals, the perception distance of the artificial fish, the maximum step length of the artificial fish movement and the crowding factor are preset. After the presetting is completed, step S2.2 is executed;

[0056] S2.2: According to the objective function of the position of the artificial fish, the objective value of each artificial fish in the initial fish population of the artificial fish population is calculated, and then the school, foraging and tail chasing behaviors of the artificial fish are executed in turn. The state of the artificial fish is updated and the optimal value is taken in the target value, and after the iteration number is counted, step S2.3 is executed;

[0057] Wherein, the artificial fish population can be described as (the connection weight w and the bias b), Indicates The total number of Indicates The individual state of (connection weight w and bias b), The variable to be optimized (optimal weight vector and optimal bias vector), the food concentration of the current position of the artificial fish is (target function), wherein The target function value is , The perception distance of the artificial fish is represented by step, which represents the maximum step length of the artificial fish movement, The crowding factor is .

[0058] S2.3: Determine whether the iteration number reaches the total number of artificial fish and the total number of iterations. If both are reached, the optimal vector matrix and the optimal bias vector matrix of the MLP neural network model are output according to the optimal value, and step S2.4 is executed, otherwise step S2.2 is continued.

[0059] S2.4: Construct an MLP neural network model based on the optimal vector matrix and the optimal bias vector matrix.

[0060] The behaviors observed in artificially bred fish, including swarming, foraging, and tail-chasing, are as follows:

[0061] (1) Foraging behavior. The current state of the artificially raised fish is as follows: Randomly select a state within its perception range as , , Represents any random number between 0 and 1. Compares two food concentration functions. ,when If the condition is met, then move one step in that direction; otherwise, choose another state. The comparison is performed. After repeatedly attempting the maximum number of trials, if the artificial fish's movement condition is still not met, it moves forward one step randomly. The formula is as follows:

[0062] ;

[0063] (2) Grouping behavior. Number of artificial fish within the current field of view. and the location of artificial fish in the swarm center Central artificial fish food concentration .when If it moves in that direction, it will take one step; otherwise, it will engage in foraging behavior.

[0064] ;

[0065] (3) Rear-end collision. Within the current field of vision. The smallest artificial fish is ,when If the animal moves in that direction, it will take a step forward; otherwise, it will engage in foraging behavior.

[0066] ;

[0067] Specific combination Figure 1 First, artificial fish are randomly generated within the parameter range, and the food concentration function (objective function) is calculated and the target value is recorded. Then, the state of each artificial fish after performing the above three behaviors is compared with the target value, and the optimal value is selected. After the experiment is carried out (total number of iterations) times, the state of the artificial fish is the optimal state. Then, the optimal vector matrix and optimal bias vector matrix of the MLP neural network model are determined based on the optimal value.

[0068] In this invention, in step S2.3, the total number of iterations is a random preset value, and the total number of iterations must be greater than the total number of artificial fish.

[0069] Furthermore, the error analysis in step S3 includes:

[0070] First, construct several groups of measured data of water flowing fractured zone height and several groups of absolute error arrays of predicted values; second, calculate the maximum absolute error, average absolute error and average relative error of measured data and predicted values according to the absolute error arrays; finally, take the maximum absolute error, average absolute error and average relative error as the judgment standard of prediction accuracy of the AFSA-MLP model.

[0071] The above prediction method is further described below in combination with an actual case:

[0072] Case 1: Water flowing fractured zone height prediction model based on MLP in fully mechanized caving mining

[0073] 83 groups of high-quality samples meeting the requirements of water flowing fractured zone height prediction model research under the conditions of gently inclined coal seam (dip angle < 25°) and fully mechanized mining are selected. Part of the experimental data is shown in Table 1:

[0074] Table 1. Measured values of water flowing fractured zone and influence factor data (part)

[0075] ;

[0076] In Table 1, is the mining thickness, is the hard rock lithology proportion coefficient, is the working face oblique length, is the mining depth, is the water flowing fractured zone height.

[0077] The measured values of water flowing fractured zone and influence factor data are standardized by z-score standardization, so that each column of data is processed into data with mean value of 0 and variance of 1, which has the advantage of being less affected by abnormal values. Details are shown in Table 2.

[0078] Table 2. Standardized measured values of water flowing fractured zone and influence factor data (part)

[0079] ;

[0080] In this case, the standardized data is divided into training set and test set in the ratio of 8:2, and the training set data and measured data are input into the MLP model for training and testing. Because the data in each column is processed into data with mean value of 0 and variance of 1 in the foregoing, the predicted water flowing fractured zone height needs to be standardized to restore the original size of the data. The formula is as follows, and the results are shown in Table 3: Figure 2

[0081] ;

[0082] ​It is calculated that the maximum absolute error of the MLP model without AFSA algorithm optimization for predicting the height of water flowing fractured zone and the measured value is 6.6 m, the minimum absolute error is 0.1 m, the average absolute error is 2.5 m, the maximum relative error is 16.3%, the minimum relative error is 0.3%, and the average relative error is 6.1%. As known from case 1, the MLP model for predicting the height of water flowing fractured zone has problems such as low accuracy of some samples and unstable model accuracy, and the accuracy of the height of water flowing fractured zone prediction model based on MLP needs to be further improved.

[0083] Case 2: AFSA-MLP based height prediction model of water flowing fractured zone in fully mechanized caving mining

[0084] Similarly, 83 groups of high-quality samples meeting the requirements of the height prediction model in the gently inclined coal seam (dip angle < 25°) under the condition of fully mechanized mining in case 1 are used, and the following prediction process is executed in turn:

[0085] (1) The measured value of the height of water flowing fractured zone and the influence factor data z-score are standardized, and the standardized data are classified, and the training data and the test data are divided according to the ratio of 8:2.

[0086] (2) The MLP neural network structure and the initialized MLP neural network parameters are determined, specifically, the training set data are used as input, the standardized measured value is used as output to construct the MLP neural network model, and the weight matrix and the bias matrix of the MLP neural network model are obtained according to the formula of the MLP neural network.

[0087] (3) The weight matrix and the bias matrix of the MLP neural network are optimized by using the AFSA algorithm; the trained MLP model is verified by using the test data, and the output value is calculated.

[0088] (4) The output value is de-standardized, the prediction result is output, the error analysis is carried out with the measured value of the water flowing fractured zone, and the model accuracy is calculated.

[0089] It should be noted that the measured data used in the error analysis process is the measured data without standardization.

[0090] In the present application, the weight matrix and the bias matrix of the MLP neural network model are optimized by using the AFSA algorithm in the experiments of case 1 and case 2 to obtain the AFSA-MLP neural network model, and the total number of iterations is preferably 100 times.

[0091] Among them, the output value is de-standardized and fitted with the measured value, and the reference Figure 3It can be obtained that the maximum absolute error of the predicted water flowing fractured zone height of the AFSA-MLP model and the measured value is 2.9 m, the average absolute error is 1.1 m, and the average relative error is 2.6%.

[0092] According to the error analysis results of case 1 and case 2, the AFSA-MLP neural network model has higher prediction accuracy and stronger stability than the MLP neural network model in predicting the water flowing fractured zone height in the fully-mechanized caving mining, because the AFSA intelligent algorithm is used to optimize the weight and bias of the MLP to compensate for the defects of the MLP neural network in the error back propagation process.

[0093] In order to further verify the engineering practical value of the AFSA-MLP model, the prediction method is applied to predict the water flowing fractured zone height of the overburden rock in the working face of the Yulin ShenShuPan coal mine as follows:

[0094] The actual measured average inclined length of a coal seam working face in ShenShuPan is 200 m, the average mining thickness is 11.18 m, the average mining depth is 673 m, and the hard rock lithology proportion coefficient is 0.95, and the actual measured water flowing fractured zone height of the overburden rock in the working face is 93.58 m.

[0095] The above parameters are input into the AFSA-MLP prediction model, the predicted water flowing fractured zone height of the overburden rock in the working face of ShenShuPan is 91.8 m, the absolute error between the predicted value and the measured value is 1.78 m, and the relative error is 1.9%.

[0096] In summary, the engineering example results show that the water flowing fractured zone prediction model based on AFSA-MLP has high precision and certain engineering practical value in fully-mechanized caving mining.

[0097] Based on the comprehensive analysis of the influencing factors of the water flowing fractured zone height in the fully-mechanized caving mining, the mining thickness, the mining depth, the working face inclined length and the hard rock lithology proportion coefficient are selected as the important influencing factors for the water flowing fractured zone calculation according to the collectability of the data and the importance of the influence, the AFSA intelligent algorithm is used to optimize the weight and bias of the MLP to compensate for the defects of the MLP neural network in the error back propagation process, the water flowing fractured zone height prediction model based on AFSA-MLP is constructed, the practicability of the model is verified through the test set and the engineering example, and certain reference significance is provided for the water flowing fractured zone prediction in the fully-mechanized caving mining.

[0098] The above-described embodiments only express several embodiments of the present application, and the description is relatively specific and detailed, but it cannot be understood as a limitation on the scope of the present application. It should be noted that, for ordinary skilled persons in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the protection scope of the present application.

Claims

1. A method for predicting the height of a water flowing fractured zone in a fully mechanized caving mining based on an AFSA-MLP neural network model, characterized in that, The method comprises the following steps: S1: collecting a plurality of sets of index data and a plurality of sets of measured data of the height of the water flowing fractured zone under the fully mechanized mining condition of the coal seam, and performing data classification on the index data in a set proportion after standardizing the index data and the measured data to obtain training set data and test set data; S2: initializing a weight matrix and a bias matrix of an MLP neural network model according to the training set data and the measured data, optimizing the MLP neural network model by using an AFSA algorithm to obtain an optimal weight matrix and an optimal bias matrix to obtain an AFSA-MLP neural network model, and the method comprises the following steps: S2.1: constructing a weight vector and a bias vector according to the connection weight matrix and the bias matrix, and constructing an artificial fish population based on the weight vector and the bias vector, and presetting the food concentration of the artificial fish, the distance between the artificial fish individuals, the perception distance of the artificial fish, the maximum step length of the artificial fish movement, and the crowding factor, and performing step S2.2 after the presetting is completed; S2.2: calculating the target value of each artificial fish in the initial fish population of the artificial fish population according to the objective function of the position of the artificial fish, and then performing the behaviors of flocking, foraging, and tail chasing on the artificial fish in turn, updating the state of the artificial fish, taking the optimal value in the target value, and performing step S2.3 after the number of iterations is counted; S2.3: judging whether the number of iterations reaches the total number of artificial fish and the total number of iterations, if both reach, outputting the optimal vector matrix and the optimal bias vector matrix of the MLP neural network model according to the optimal value, and performing step S2.4, otherwise, continuing to perform step S2.2; S3: substituting the test set data into the AFSA-MLP neural network model to obtain an output value, performing reverse standardization on the output value to obtain a predicted value, performing error analysis on the predicted value according to the measured data to obtain the prediction accuracy of the AFSA-MLP neural network model.

2. The method according to claim 1, wherein the method is characterized in that: The index data at least includes the following water flowing fractured zone height influencing factors: mining thickness, mining depth, working face inclined length, and hard rock lithology proportion coefficient.

3. The method according to claim 1, wherein the method is characterized in that: The standardization processing in step S1 is specifically: performing z-score standardization processing on the plurality of sets of index data and the plurality of sets of measured data to form the mean value of the plurality of sets of index data and the plurality of sets of measured data as 0 and the variance as 1.

4. The method according to claim 1, wherein the method is characterized in that: The initialization of the weight matrix and the bias matrix of the MLP neural network model in step S2 is specifically: constructing an MLP neural network model by taking the training set data as input and the standardized measured data as output, and obtaining the weight matrix and the bias matrix of the MLP neural network model by using the following formula; using the formula as follows: ； wherein, is the output of the first hidden layer node, is the input training set data, is the measured data, f is a ReLU activation function, is the bias of the jth hidden layer node, is the bias of the kth output node, is the bias of the jth hidden layer node, is the connection weight between the jth hidden layer and the kth output node; and is the connection weight between the jth hidden layer and the kth output node, n is the nth hidden layer node, and m is the mth output node.

5. The method according to claim 4, wherein the method is characterized in that: In step S2.3, the total number of iterations is a random preset value, and the total number of iterations must be greater than the total number of artificial fish.

6. The method according to claim 1, wherein the method is characterized in that: The error analysis in step S3 includes: (1) constructing a plurality of sets of absolute value error arrays of the measured data of the height of the water flowing fractured zone and a plurality of sets of the predicted values; (2) calculating the maximum absolute value error, the average absolute value error, and the average relative error of the measured data and the predicted values according to the absolute value error arrays; (3) The maximum absolute error, the average absolute error and the average relative error are taken as the judgment criteria of the prediction accuracy of the AFSA-MLP model.

Images