A method for adaptive adjustment of hydraulic support posture based on digital twin model

Abstract

The present invention discloses a method for predicting the posture of hydraulic supports in mines and adaptively adjusting support. The method uses digital twin technology to establish a three-dimensional model of the hydraulic support and adopts a hydraulic support posture prediction model that integrates an improved LSTM-GRU two-layer neural network and a genetic algorithm to achieve adaptive adjustment of the hydraulic support posture. The method comprises the following steps: (1) establishing a three-dimensional model of the hydraulic support; (2) obtaining measurement information of the relative posture of the hydraulic support; (3) training the network model to obtain the optimal solution of the prediction model; (4) constructing display information in a UGUI component in Unity3D; (5) building a decision control platform, constructing a classification model using a decision tree algorithm and a training set, and verifying the decision control effect of the classification model using a test set. The method uses digital twin technology and a deep learning neural network algorithm based on an attention mechanism to achieve adaptive adjustment of the hydraulic support posture.

Description

Technical Field

[0001] The present invention relates to deep learning adaptive control technology in the field of digital twin and artificial intelligence, and specifically to a method for adaptively adjusting the posture of a hydraulic support based on a digital twin model. Background Art

[0002] As an important support equipment for fully mechanized mining working faces, hydraulic supports effectively support the surrounding rock of coal seams by adjusting the support posture, providing support and protection for the roof, floor, coal wall and goaf of the fully mechanized mining working face. Currently, the control of hydraulic supports at home and abroad mainly forms automated support for the surrounding rock through electro-hydraulic control, which cannot achieve autonomous decision-making and control, and cannot meet the actual production needs of intelligent fully mechanized mining working faces in terms of posture monitoring and adjustment, and adaptability to complex working conditions. In order to improve the intelligence level and intelligent support level of hydraulic supports, it is necessary to accurately perceive the support posture of the hydraulic supports during operation, and to be able to autonomously control the support posture of the hydraulic supports to adapt to the complex and changeable coal seam surrounding rock roof. However, how to accurately and quickly perceive the current working posture of the hydraulic supports and autonomously adjust their support posture remains a key technical problem that needs to be solved in intelligent fully mechanized mining working faces.

[0003] During the hydraulic support support work, it constantly interacts with the surrounding rock of the mine tunnel in which it is located. In the intelligent fully mechanized mining working face, the hydraulic support needs to accurately sense the environment and make autonomous decisions and adjust its own posture to achieve the optimal support state. The present invention is based on the electro-hydraulic control theory and posture monitoring method of the hydraulic support. Through in-depth analysis of the characteristics of the hydraulic support support posture change, the support posture accurate perception and autonomous control algorithm are studied. On this basis, through the real-time interactive information between the hydraulic support and the surrounding rock environment, an improved hydraulic support posture prediction model that integrates the LSTM-GRU two-layer neural network and the genetic algorithm is proposed. A decision-making control platform is built to achieve adaptive adjustment of the hydraulic support posture and improve the intelligent support level of the hydraulic support. Summary of the Invention

[0004] This paper primarily addresses the problem of how to enable hydraulic supports to accurately perceive their environment, make autonomous decisions, and adjust their posture to achieve optimal support conditions. It proposes a method for autonomous decision-making in intelligent mine hydraulic supports. This method, based on digital twin technology and artificial intelligence deep learning algorithms, employs a hydraulic support posture prediction model that integrates an improved LSTM-GRU two-layer neural network with a genetic algorithm to achieve autonomous adjustment of the hydraulic support's support posture. The method includes the following steps:

[0005] Step 1: Create a 3D model of the hydraulic support: Use SolidWorks to create a 3D model of the hydraulic support, import it into Unity3D software in .fbx format, and add a parent-child relationship for the virtual hydraulic support in the 3D model; use C# scripts to write relative pose inversion and accuracy solution methods, and embed them into the bottom layer of the virtual 3D model system;

[0006] Step 2: Obtain measurement information of the relative posture of the hydraulic support: define the relative posture matrix and posture plane of the hydraulic support to describe the posture state of the support, arrange the monitoring points according to the rays of the posture matrix, and install measurement sensors on the hydraulic support to obtain measurement information of the relative posture of the monitoring points, and pre-process the measurement information;

[0007] Step 3, network model training: using an improved LSTM-GRU two-layer neural network and genetic algorithm fusion optimization method to establish a hydraulic support posture prediction model, the hydraulic support measurement parameters are input into the hydraulic support posture prediction model, an attention mechanism is added to the LSTM and GRU layers of the hydraulic support posture prediction model, and the genetic algorithm is used to tune the model parameters to obtain the optimal solution, which is used as the number of layers and the number of neurons in each layer of the hydraulic support posture prediction model to predict the posture state of the hydraulic support;

[0008] Step 4: Use the UGUI component in Unity3D to build an information panel, where the information panel includes test information and posture information, wherein the test information is used to display the measured sensor data, and the posture information is used to display the obtained posture information;

[0009] Step 5: Build a decision-making control platform based on Unity3D development software; the decision-making control platform consists of a three-dimensional status monitoring module, an aircraft parameter display module, and a remote intervention display module. A classification model is constructed using a decision tree algorithm and a training set, and the decision-making control effect of the classification model is verified using a test set.

[0010] The relative posture matrix measurement method of the hydraulic support is as follows:

[0011] Determine the position and attitude plane of the moving hydraulic support based on the coordinates of the three known monitoring points in the hydraulic support reference coordinate system O-XYZ;

[0012] For any point P in space, its coordinates can be solved by the following formula:

[0013]

[0014] where x p ,y p , z pis the coordinate of point P in the rectangular coordinate system O-XYZ, l is the length of vector OP, θ is the angle between the projection of vector OP on the YOZ plane and the positive direction of the z axis, is the angle between the projection of vector OP on the XOY plane and the positive direction of the Z axis;

[0015] The coordinates of the three monitoring points in the reference hydraulic support coordinate system can be solved by the following formula:

[0016]

[0017]

[0018]

[0019] Among them, θ1, θ2, and θ3 are the angles between the projections of the measurement points A, B, and C on the YOZ plane and the positive direction of the Z axis, respectively. are the angles between the projections of the measuring points A, B, and C on the XOZ plane and the positive direction of the Z axis, and l1, l2, and l3 are the distances from the measuring points A, B, and C to point O respectively;

[0020] By solving the coordinates of monitoring points A, B, and C, the posture plane of the hydraulic support is determined.

[0021] The measurement information preprocessing process of the method is as follows:

[0022] Step 2-1: Remove invalid data: remove the data containing a large number of null values ​​and data that exceeds the range of the hydraulic support;

[0023] Step 2-2: Missing value processing: For data with individual missing data, use the average value for interpolation;

[0024] Step 2-3: Normalize the measurement information using the deviation method:

[0025]

[0026] Among them, X min is the minimum value in the experimental data, X max is the maximum value, X is the original data value, and X′ is the normalized data value;

[0027] The method uses an LSTM-GRU two-layer neural network to enhance key information in the input sequence and uses an attention mechanism to extract internal features to enhance the learning ability and output performance of the model.

[0028] The method uses the attention mechanism to optimize the hydraulic support posture prediction model. The specific algorithm process is as follows:

[0029] 1) Get the output of the network layer of the LSTM-GRU two-layer neural network as the input x of the attention mechanism layer, that is, {x1,x2,…,x n};

[0030] 2) Use Linear linear transformation function and ReLU() activation function to build a fully connected layer and get new samples {e1,e2,…,e n} and then pass the dimension compression function squeeze({e1,e2,…,e n})convert;

[0031] 3) Input the samples into the sofamax() function according to the time series parameter column to obtain the weight vector W. The number of column vector elements in W is equal to the input time step, and each element represents the weight of a time step. Then W is transformed into a different dimension by the dimension expansion function unsqueeze(W);

[0032] 4) Multiply the initial sample x and the weight vector W to obtain the sample {x1W, x2W, ..., x n W};

[0033] 5) Sum the samples according to the time series parameter column to obtain the final output of the attention mechanism layer.

[0034] The method adopts genetic algorithm to tune the parameters of LSTM layer, GRU layer, fully connected layer and the number of neurons in each layer of the hydraulic support posture prediction model, and uses the root mean square error of test data as the loss function to solve the optimal solution.

[0035] The training and optimization process of the hydraulic support posture prediction model of the method is as follows:

[0036] Step 3-1: Data acquisition, processing the hydraulic support posture information and filtering the data;

[0037] Step 3-2: Input the monitoring data of the hydraulic support, using the monitoring data of the hydraulic support as the input variable;

[0038] Step 3-3: Initialize the genetic algorithm parameter settings, set the DNA length, chromosome length, population size, crossover rate, mutation rate and number of iterations;

[0039] Step 3-4: Randomly generate N populations, perform crossover and mutation, calculate fitness, and select the solution with the best fitness value in each generation;

[0040] Step 3-5: Generate a new population and perform the next iteration until the number of iterations reaches the maximum value. The optimal solution obtained is used as the number of layers and the number of neurons in each layer of the hydraulic support posture prediction model;

[0041] Step 3-6: Train and obtain the hydraulic support posture prediction model.

[0042] The hydraulic support posture prediction model adopts an LSTM-GRU two-layer neural network model and an attention mechanism, and uses the mean square error as a loss function to train the prediction model. Then, a genetic algorithm is used for repeated iteration to select the optimal parameters with the best fitness.

[0043] The hydraulic support posture prediction model uses a genetic algorithm to optimize the hyperparameters of the hydraulic support posture neural network to obtain the optimal parameter combination of learning rate and the number of hidden layer neurons.

[0044] The hydraulic support posture prediction model uses an optimal parameter combination to construct a hydraulic support posture prediction model as a nonlinear transformation function between input samples and output samples at the current moment.

[0045] The hydraulic support posture prediction model applies a nonlinear transformation function to obtain the posture information of the hydraulic support at the next moment.

[0046] The construction process of the hydraulic support posture prediction model further includes the following steps:

[0047] 1) Select the training data set. The input data of the hydraulic support posture prediction model consists of the posture information of the three monitoring points A, B, and C of the hydraulic support at different times. The input vector m of the monitoring point is i and the target output vector n i Expressed as:

[0048] m i =[A i-2 ,A i-1 ,A i ,B i-2 ,B i-1 ,B i ,C i-2 ,C i-1 ,C i ,P i-2 ,P i-1 ,P i ]

[0049]

[0050] Where A i-2 , A i-1 , A i , B i-2 , B i-1 , B i , C i-2 , C i-1 , C i , Pi-2 , P i-1 , P i The coordinates of points A, B, C, and P at time i-2, i-1, and i respectively. i+1 , B i+1 , C i+1 are the coordinates of points A, B, and C at time i+1;

[0051] 2) Using genetic algorithm to optimize the parameters of the neural network for hydraulic support posture;

[0052] a. Randomly generate the initial population, perform population initialization and chromosome encoding and decoding operations;

[0053] b. Calculate the fitness value of each individual in the initial population and determine whether it meets the optimization criteria;

[0054] c. Perform selection, crossover and mutation operations on chromosomes;

[0055] d. Decode chromosomes and calculate the fitness of individuals in the population;

[0056] e. If the genetic termination condition is met, the optimal parameters calculated by the genetic algorithm are used as the final parameters of the LSTM-GRU two-layer neural network model; otherwise, return to step c;

[0057] 3) Training the hydraulic support posture prediction model;

[0058] a. Change m i Input hydraulic support posture prediction model, output of hydraulic support posture network model n i The position of the hydraulic support at the next moment;

[0059] b. The difference between the target output and the predicted output f(u) of the hydraulic support posture prediction model is expressed as error e, that is, e = n i -f(u);

[0060] c. Minimize the mean square error of the hydraulic support posture model to obtain the fitting error, and finally obtain the optimal fitting function between the input sample and the output sample;

[0061] 4) Predict the trajectory of the hydraulic support, send the experimental data to the hydraulic support posture neural network prediction model, and use the optimal fitting function to calculate the hydraulic support posture state at the next moment.

[0062] The beneficial effects of the present invention are:

[0063] This invention, based on digital twin technology and artificial intelligence deep learning algorithms, constructs a hydraulic support posture neural network prediction model to develop a hydraulic support posture adaptive method based on a digital twin model. This invention addresses the problems of low intelligence and weak adaptive adjustment capabilities in existing hydraulic support posture monitoring. By integrating real-time information between the hydraulic support and the surrounding rock environment, an improved hydraulic support posture prediction model integrating a LSTM-GRU two-layer neural network and a genetic algorithm is established to achieve adaptive adjustment of the hydraulic support posture, thereby improving the real-time performance, accuracy, and system robustness of the hydraulic support posture, as well as the level of intelligent mine support. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 Flowchart of the hydraulic support posture prediction model based on the improved LSTM-GRU two-layer neural network and genetic algorithm

[0065] Figure 2 Diagram of the hydraulic support posture adjustment method based on digital twin theory

[0066] Figure 3 The spatial coordinate diagram of the monitoring point of the hydraulic support

[0067] Figure 4 Coordinate diagram for monitoring points of hydraulic support DETAILED DESCRIPTION

[0068] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0069] Reference Figure 1 The following is a flowchart of the improved hydraulic support posture prediction model that integrates the LSTM-GRU two-layer neural network and the genetic algorithm. First, the monitoring data is processed, filtered, and missing values ​​are supplemented. The monitoring data is input, the dataset is divided, and the genetic algorithm parameters are initialized. The genetic algorithm is iterated repeatedly to select the parameters with the best fitness. The prediction model is trained using the LSTM-GRU two-layer neural network model and the MES mean squared error as the loss function.

[0070] The specific methods for parameter setting and optimization of the prediction model are as follows:

[0071] (1) LSTM layer

[0072] Assume the network input is x t , the hidden state is h t At time t, the calculations of each unit and gate are as follows:

[0073] f t =σ(W f ·(h t ,x t)+b f )

[0074] i t =σ(W f ·(h t-1 ,x t )+b c )

[0075]

[0076]

[0077] o t =σ(W o ·(h t-1 ,x t )+b o )

[0078] h t =o t tanh(C t )

[0079] Where: σ is the sigmoid activation function; b f is the bias; W f is the weight; C t-1 is the state of the unit that stores the previous memory information; f t is the degree of forgetting controlled by the forget gate; i t the extent to which input information should be updated; is the candidate vector; C t Represents the current unit state; o t is the output gate; the hidden layer output h t ;

[0080] There is no stable relationship between the number of hidden layer neurons and the training results, so continuous experimentation is required to obtain the optimal parameters. During model construction, the most commonly used sigmoid and tanh functions are used as activation functions. The mean square error (MSE) measures the error by averaging the sum of the squares of the differences between the predicted and target values. The formula is as follows:

[0081]

[0082] (2) GRU layer

[0083] GRU has improved the adverse effects of gradient disappearance and gradient explosion. t and the previous moment h t-1 Added linear correlation. The specific expression is as follows:

[0084] r t =σ(Wxr x t +W hr h t-1 +b f )

[0085] z t =σ(W xh x t +W hz h t-1 +b c )

[0086]

[0087]

[0088] Where: r t 、z t 、 Represent the reset gate, update gate and input vector respectively; ⊙ is the Hadamard product (matrix operation), which is the product of the corresponding elements in the operation matrix; W xr 、W hr 、W xh 、W hz 、W hh are the weight matrices of the corresponding functions respectively; the meanings of other variables are the same as those of the LSTM network.

[0089] (3) Genetic Algorithm

[0090] a. Chromosome encoding

[0091] The number of hidden neurons and learning rate in the LSTM network are used as the initialization objects of the genetic algorithm, and chromosome encoding is performed in the form of real number encoding. The number of genes per chromosome is 2; respectively, they represent the number of hidden neurons Zt and the learning rate Ts of the LSTM model, so each chromosome is represented as X = (Zt, Ts).

[0092] b. Initial parameter settings

[0093] The solution is a set of possible (Zt, Ts) values. Such a set of possible solutions is called a population. A random method is selected to generate the initial population. When randomly generating, it is important to set the random generation range for the two genes of the chromosome, otherwise the generated solutions will exceed the practical significance.

[0094] c. Fitness function construction

[0095] Based on the overall fit between the estimated value and the true value of the hydraulic support position, a fitness function is constructed; in order to make the network parameters obtained by the genetic algorithm more suitable for the LSTM model and improve the generalization ability of the model, the data is divided into training samples and test samples; the LSTM network is trained using the training samples to obtain the training sample output value and test sample output value of the LSTM network. The individual fitness function is defined as:

[0096]

[0097] in, are the predicted values ​​of the training sample and the test sample, respectively, y j t 、y k v is the actual value; the error of the test sample directly reflects the prediction effect of the model; the error of the training sample and the error of the test sample are given the same weight, that is, 0.5, and the product of the two and the weight is used as the fitness function of the model.

[0098] d. Selection operator, crossover operator and mutation operator

[0099] 1) The selection operator selects individuals with better adaptability in the current population as parents and passes genetic information to offspring. Here, the league selection algorithm is used as the GA selection strategy. The change law of the number of hidden layer neurons and the learning rate is shown in the formula:

[0100] c.Ln=abs(c.Ln+random.randint(-3, 3))

[0101] c.ln=abs(c.ln+random.uniform(-0.001,0.001))

[0102] Where c.Ln is the number of hidden neurons in the population, and c.ln is the learning rate;

[0103] 2) The crossover operation is based on a pair of randomly selected individuals, and the chromosome genes are exchanged and recombined to produce a new chromosome. The simulated binary crossover operator is selected for the crossover operation. The calculation formula for the new individual after the crossover is:

[0104]

[0105]

[0106] in:

[0107]

[0108] η is a custom distribution factor, generally set to 1; x is the number of cells in the parent generation and

[0109] 3) Mutation Operation

[0110] Set the population to perform mutation operation with a mutation probability of 0.03; select the polynomial mutation operator for mutation operation; the polynomial mutation operator formula is:

[0111] v k =v k +δ·( u kl k )

[0112] in,

[0113]

[0114] δ1=(v k -l k ) / (u k -l k )

[0115] δ2=(u k -v k ) / (u k -l k )

[0116] Random refers to the probability of mutation, v k is a parent individual, is the distribution index, which is generally set to 1; after the above genetic algorithm operation process, a new population is generated, and after calculating the fitness value, after repeated iterations of the population, the parameters that construct the optimal fitness are selected as the optimal parameters found by the genetic algorithm.

[0117] Reference Figure 2 , which is a diagram of the hydraulic support posture adjustment method based on digital twin theory. Using the coal seam roof and floor data cut by a real coal mining machine in physical space and the real hydraulic support prototype, the corresponding coal seam environment roof and floor are created in virtual space, and the hydraulic support model is imported. Then, the hydraulic support model is repaired and physical components are added to enable it to adaptively fit the coal seam roof and floor surfaces. Finally, through the kinematic analysis of the hydraulic support model, a dynamic coupling model of the hydraulic support and the coal seam environment is established, thereby obtaining the optimal support posture of the hydraulic support for the coal seam environment. The support related information of the virtual space hydraulic support can provide a reference for the support posture of the physical space hydraulic support.

[0118] Reference Figure 3 , is the spatial coordinate solution of the monitoring points of the hydraulic support. Knowing the coordinates of the three monitoring points in the local coordinate system of the reference support, the position of the movable support and the attitude surface △ABC can be determined. The spatial coordinates of any point in space can be determined based on the direction angle and vector length. Figure 3As shown, a rectangular coordinate system O-XYZ is established in space, with the origin of the coordinate system being O. There is a point P in space. The angle between the projection of the vector OP on the YOZ plane and the positive direction of the z axis is defined as θ, and the angle between the projection of the vector OP on the XOY plane and the positive direction of the z axis is defined as Will Figure 3 Substituting the geometric relationship shown in the figure into the parametric equation of the circle formed by the unit vector OP on the x-axis and y-axis, the solution formula for the coordinate point P is finally obtained:

[0119]

[0120] where x p ,y p , z p are the coordinates of point P in the rectangular coordinate system O-XYZ, and l is the length of the vector OP.

[0121] Reference Figure 4 , is the coordinate solution diagram of the hydraulic support monitoring point. According to the single point coordinate solution formula, the coordinate solution formula of the three monitoring points of the movable support in the local coordinate system of the reference support is as follows:

[0122]

[0123]

[0124]

[0125] It can be seen from the above formula that by solving the coordinates of the monitoring points, the posture plane of the hydraulic support is determined, and the nine parameters of the monitoring points, θ1, θ2, θ3, l1, l2, and l3 are in the coordinate system of adjacent hydraulic supports.

[0126] The above content is a further detailed description of the present invention in combination with a specific preferred embodiment. It cannot be determined that the specific embodiments of the present invention are limited to this. For ordinary technicians in the technical field to which the present invention belongs, simple replacements and changes made without departing from the design concept of the present invention should be deemed to fall within the scope of protection of the claims submitted by the present invention.

Claims

1. A method for self-adaptive adjustment of hydraulic support posture based on digital twin, characterized in that: The digital twin technology is used to establish a three-dimensional model of the hydraulic support. The hydraulic support posture prediction model that integrates the improved LSTM-GRU two-layer neural network and the genetic algorithm is adopted to achieve adaptive adjustment of the hydraulic support posture. The steps include: Step 1: Create a 3D model of the hydraulic support: Use SolidWorks to create a 3D model of the hydraulic support, import it into Unity3D software in .fbx format, and add a parent-child relationship for the virtual hydraulic support in the 3D model; use C# scripts to write relative pose inversion and accuracy solution methods, and embed them into the bottom layer of the virtual 3D model system; Step 2: Obtain measurement information of the relative posture of the hydraulic support: define the relative posture matrix and posture plane of the hydraulic support to describe the posture state of the support, arrange the monitoring points according to the rays of the posture matrix, and install measurement sensors on the hydraulic support to obtain measurement information of the relative posture of the monitoring points, and pre-process the measurement information; Step 3, network model training: using an improved LSTM-GRU two-layer neural network and genetic algorithm fusion optimization method to establish a hydraulic support posture prediction model, the hydraulic support measurement parameters are input into the hydraulic support posture prediction model, an attention mechanism is added to the LSTM and GRU layers of the hydraulic support posture prediction model, and the genetic algorithm is used to tune the model parameters to obtain the optimal solution, which is used as the number of layers and the number of neurons in each layer of the hydraulic support posture prediction model to predict the posture state of the hydraulic support; Step 4: Use the UGUI component in Unity3D to build an information panel, where the information panel includes test information and posture information, wherein the test information is used to display the measured sensor data, and the posture information is used to display the obtained posture information; Step 5: Build a decision-making control platform based on Unity3D development software; the decision-making control platform consists of a three-dimensional state monitoring module, an aircraft parameter display module, and a remote intervention display module. A classification model is constructed using a decision tree algorithm and a training set, and the decision-making control effect of the classification model is verified using a test set. It is also characterized in that the relative posture matrix measurement method of the hydraulic support in step 2 is as follows: Determine the position and attitude plane of the moving hydraulic support based on the coordinates of the three known monitoring points in the hydraulic support reference coordinate system O-XYZ; For any point P in space, its coordinates can be solved by the following formula: where x p ,y p , z p is the coordinate of point P in the rectangular coordinate system O-XYZ, l is the length of vector OP, θ is the angle between the projection of vector OP on the YOZ plane and the positive direction of the z axis, is the angle between the projection of vector OP on the XOY plane and the positive direction of the Z axis; The coordinates of the three monitoring points in the reference hydraulic support coordinate system can be solved by the following formula: Among them, θ1, θ2, and θ3 are the angles between the projections of the measurement points A, B, and C on the YOZ plane and the positive direction of the Z axis, respectively. are the angles between the projections of the measuring points A, B, and C on the XOZ plane and the positive direction of the Z axis, and l1, l2, and l3 are the distances from the measuring points A, B, and C to point O respectively; By solving the coordinates of monitoring points A, B, and C, the posture plane of the hydraulic support is determined; The measurement information preprocessing process of the method is as follows: Step 2-1: Remove invalid data: remove the data containing a large number of null values ​​and data that exceeds the range of the hydraulic support; Step 2-2: Missing value processing: For data with individual missing data, use the average value for interpolation; Step 2-3: Normalize the measurement information using the deviation method: Among them, X min is the minimum value in the experimental data, X max is the maximum value, X is the original data value, X ′ is the normalized data value; The method uses a LSTM-GRU two-layer neural network to enhance key information in the input sequence and uses an attention mechanism to extract internal features to enhance the model's learning ability and output performance. The method uses the attention mechanism to optimize the hydraulic support posture prediction model. The specific algorithm process is as follows: 1) Get the output of the network layer of the LSTM-GRU two-layer neural network as the input x of the attention mechanism layer, that is, {x1,x2,…,x n }; 2) Use Linear linear transformation function and ReLU() activation function to build a fully connected layer and get new samples {e1,e2,…,e n } and then pass the dimension compression function squeeze({e1,e2,…,e n })convert; 3) Input the samples into the sofamax() function according to the time series parameter column to obtain the weight vector W. The number of column vector elements in W is equal to the input time step, and each element represents the weight of a time step. Then W is transformed into a different dimension by the dimension expansion function unsqueeze(W); 4) Multiply the initial sample x and the weight vector W to obtain the sample {x1W, x2W, ..., x n W}; 5) Sum the samples according to the time series parameter column to obtain the final output of the attention mechanism layer; The method uses a genetic algorithm to optimize the parameters of the LSTM layer, GRU layer, fully connected layer and the number of neurons in each layer of the hydraulic support posture prediction model, and uses the root mean square error of the test data as the loss function to solve the optimal solution; The training and optimization process of the hydraulic support posture prediction model of the method is as follows: Step 3-1: Data acquisition, processing the hydraulic support posture information and filtering the data; Step 3-2: Input the monitoring data of the hydraulic support, using the monitoring data of the hydraulic support as the input variable; Step 3-3: Initialize the genetic algorithm parameter settings, set the DNA length, chromosome length, population size, crossover rate, mutation rate and number of iterations; Step 3-4: Randomly generate N populations, perform crossover and mutation, calculate fitness, and select the solution with the best fitness value in each generation; Step 3-5: Generate a new population and perform the next iteration until the number of iterations reaches the maximum value. The optimal solution obtained is used as the number of layers and the number of neurons in each layer of the hydraulic support posture prediction model; Step 3-6: Train and obtain the hydraulic support posture prediction model; The hydraulic support posture prediction model adopts an LSTM-GRU two-layer neural network model and an attention mechanism, and uses the mean square error as the loss function to train the prediction model. Then, a genetic algorithm is used to iterate repeatedly to select the optimal parameters with the best fitness. The hydraulic support posture prediction model uses a genetic algorithm to optimize the hyperparameters of the hydraulic support posture neural network to obtain the optimal parameter combination of learning rate and number of hidden layer neurons; The hydraulic support posture prediction model is constructed by using the optimal parameter combination as a nonlinear transformation function between the input sample and the output sample at the current moment; The hydraulic support posture prediction model applies a nonlinear transformation function to obtain the posture information of the hydraulic support at the next moment.

2. The method according to claim 1, wherein The method for constructing a hydraulic support posture prediction model further comprises the following steps: Step 2-1: Select the training data set. The input data of the hydraulic support posture prediction model consists of the posture information of the three monitoring points A, B, and C of the hydraulic support at different times. The input vector m of the monitoring point i and the target output vector n i Expressed as: m i =[A i-2 ,A i-1 ,A i ,B i-2 ,B i-1 ,B i ,C i-2 ,C i-1 ,C i ,P i-2 ,P i-1 ,P i ] Where A i-2 , A i-1 , A i , B i-2 , B i-1 , B i , C i-2 , C i-1 , C i , P i-2 , P i-1 , P i are the coordinates of points A, B, C, and P at time i-2, i-1, and i respectively. i+1 , B i+1 , C i+1 are the coordinates of points A, B, and C at time i+1; Step 2-2: Optimize the parameters of the hydraulic support posture neural network using genetic algorithm; a. Randomly generate the initial population, perform population initialization and chromosome encoding and decoding operations; b. Calculate the fitness value of each individual in the initial population and determine whether it meets the optimization criteria; c. Perform selection, crossover and mutation operations on chromosomes; d. Decode chromosomes and calculate the fitness of individuals in the population; e. If the genetic termination condition is met, the optimal parameters calculated by the genetic algorithm are used as the final parameters of the LSTM-GRU two-layer neural network model; otherwise, return to step c; Step 2-3: Training the hydraulic support posture prediction model; a. Change m i Input hydraulic support posture prediction model, output of hydraulic support posture network model n i The position of the hydraulic support at the next moment; b. The difference between the target output and the predicted output f(u) of the hydraulic support posture prediction model is expressed as error e, that is, e = n i -f(u); c. Minimize the mean square error of the hydraulic support posture model to obtain the fitting error, and finally obtain the optimal fitting function between the input sample and the output sample; Step 2-4: Predict the hydraulic support posture, send the experimental data to the hydraulic support posture neural network prediction model, and use the optimal fitting function to calculate the hydraulic support posture state at the next moment.

3. The method according to claim 1, wherein The measuring sensor device includes a laser distance measuring sensor, a laser transmitter, an inclination sensor and a travel sensor.

Images