Hydraulic support load prediction method, device and system and storage medium

Abstract

The invention discloses a hydraulic support load prediction method, device and system and a storage medium, and the method comprises the steps: collecting the load time sequence data of a coal face hydraulic support, and carrying out the normalization processing of the load time sequence data; decomposing the preprocessed load time sequence data to obtain a plurality of intrinsic mode components; meanwhile, redundant components are recognized and eliminated, and an effective feature input sequence is constructed; extracting local detail features, short-term dependency features and long-term trend features of the load data in parallel through convolution kernels of different scales; focusing long-term stable key features through a static attention mechanism, and adaptively adjusting weight distribution according to a feature input sequence through a dynamic attention mechanism to obtain output of residual connection; and according to the output of the residual connection, establishing a deep dependency relationship between the feature input sequence and the load time sequence data by using a long-short-term memory neural network, and realizing accurate prediction of the support load in the future time period.

Description

Technical Field

[0001] This invention belongs to the field of coal mine safety mining and intelligent monitoring technology, specifically relating to a method, device, system, and storage medium for predicting the load of hydraulic supports in coal mining faces. Background Technology

[0002] As mining depth increases, geological conditions become more complex, making roof management at the working face more challenging. Hydraulic support load, as a crucial indicator of changes in roof stress, reflects the stability of the mine's internal geological structure. Preventing hydraulic supports from operating under prolonged high loads, which could lead to production stoppages due to support failure, is essential. Therefore, accurately predicting hydraulic support load is vital for ensuring mine safety and improving production efficiency. Summary of the Invention

[0003] To address the problems existing in the prior art, the present invention provides a method, device, system, and storage medium for predicting the load of a hydraulic support.

[0004] To achieve the above objectives, the present invention provides the following solution: A method for predicting the load of a hydraulic support, comprising: Step S1: Collect load time series data of hydraulic support at coal mining face, and normalize the load time series data. Step S2: Decompose the preprocessed load time series data to obtain multiple intrinsic mode components; at the same time, identify and remove redundant components to construct an effective feature input sequence. Step S3: Extract local detail features, short-term dependency features, and long-term trend features of the payload data in parallel using convolutional kernels of different scales; Step S4: Focus on long-term stable key features through a static attention mechanism, and adaptively adjust the weight distribution according to the feature input sequence through a dynamic attention mechanism to obtain the output of the residual connection; Step S5: Based on the output of the residual connection, a deep dependency relationship between the feature input sequence and the load time series data is established using a long short-term memory neural network to achieve accurate prediction of the stent load in future periods.

[0005] Preferably, in step S2, the preprocessed load data is decomposed using an adaptive noise complete set empirical mode decomposition algorithm to obtain multiple intrinsic mode components.

[0006] Preferably, in step S2, redundant components are identified and removed based on the KL divergence criterion to construct an effective feature input sequence.

[0007] The present invention also provides a hydraulic support load prediction device, comprising: The first processing module is used to collect the load time series data of the hydraulic support of the coal mining face, and at the same time, to normalize the load time series data. The second processing module is used to decompose the preprocessed load time series data to obtain multiple intrinsic mode components; at the same time, it identifies and removes redundant components to construct an effective feature input sequence. The third processing module is used to extract local detail features, short-term dependency features and long-term trend features of the payload data in parallel using convolution kernels of different scales; The fourth processing module is used to focus on long-term stable key features through a static attention mechanism, and at the same time, to adaptively adjust the weight distribution according to the feature input sequence through a dynamic attention mechanism to obtain the output of the residual connection. The fifth processing module is used to establish a deep dependency relationship between the feature input sequence and the load time series data based on the output of the residual connection using a long short-term memory neural network, so as to achieve accurate prediction of the stent load in future periods.

[0008] Preferably, the second processing module uses an adaptive noise complete set empirical mode decomposition algorithm to decompose the preprocessed load data and obtain multiple intrinsic mode components.

[0009] Preferably, the second processing module identifies and removes redundant components based on the KL divergence criterion to construct an effective feature input sequence.

[0010] The present invention also provides a hydraulic support load prediction system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes a hydraulic support load prediction method when executed by the processor.

[0011] The present invention also provides a storage medium storing a computer program that executes a hydraulic support load prediction method when running.

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention decomposes support load data using the CEEMDAN algorithm and removes redundant components using KL divergence, improving the effectiveness of input data. It utilizes the MTAM module (multi-scale convolution + dynamic-static temporal attention) to accurately capture key features of load changes and combines LSTM to construct a deep dependency relationship between features and loads, enabling advanced prediction of support loads. Ultimately, this provides technical support for the assessment of hydraulic support operating status, mine roof management, and prevention of accidents such as rockbursts, ensuring safe and efficient mine production. Attached Figure Description

[0013] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a flowchart of the hydraulic support load prediction method according to an embodiment of the present invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0017] Example 1 like Figure 1 As shown, the present invention provides a method for predicting the load of a hydraulic support, comprising: Step 1: Data Acquisition and Preprocessing Load time-series data of hydraulic supports in coal mining faces were collected. The raw data was normalized to eliminate the influence of dimensions and improve the stability of model training. Data normalization process: In the formula: Indicates the input load data; and They are The maximum and minimum values.

[0018] Step 2: Signal decomposition and feature extraction: The preprocessed load data is decomposed using an adaptive noise complete set empirical mode decomposition algorithm to obtain multiple intrinsic mode components; redundant components are identified and removed based on the KL divergence criterion to construct an effective feature input sequence; 1) For the original signal X =( x 1, x 2, …, x n CEEMDAN decomposition was performed to obtain the IMF components of the signal. Here, X represents the preprocessed support load time-series signal. x i (i =1,2,…,n) represents the first… i The support load at each sampling time point, where n is the total number of sampling points for the load time series data, and the decomposed IMF components (IMF1, IMF2, ..., IMF) are obtained. k These correspond to the high-frequency fluctuations, mid-frequency fluctuations, and low-frequency fluctuations of the load signal, respectively.

[0019] 2) Calculate the complete support load time-series signal and individual IMF components. E =( e 1, e 2, …, e n ) Calculate the original signal and individual IMF components E =( e 1, e 2, …, e n probability distribution p ( x )and q ( x ),get E Compared to X The KL divergence is: In the formula, p ( x The probability distribution of the original support load signal; q ( x The KL divergence value represents the probability distribution of a single IMF component. When the KL divergence value is small, it indicates that the distribution of the IMF component is not significantly different from that of the original load signal, and it contains more key information about load changes. When the KL divergence value is large, it indicates that the IMF component is a redundant component (such as meaningless fluctuations caused by measurement noise). IMF components with a KL divergence value less than 0.1 are removed.

[0020] Calculate the KL divergence value D ( p , q ): The result can be obtained by solving the Gaussian kernel density function. p ( x Similarly, calculate... q ( x Substituting these values ​​into equations (1) and (2), we can obtain the KL divergence value: In the formula: k (·) is the kernel function. h This is the bandwidth parameter for kernel density estimation.

[0021] Step 3: Multi-scale Feature Fusion: Three parallel 1D convolutional layers are used, with kernel sizes of 1, 3, and 5, and each layer having 16 channels, to construct a multi-scale convolutional module. The input is the IMF feature sequence filtered in Step 2 (i.e., the retained IMF components that reflect different frequencies of load fluctuations). Local detail features, short-term dependency features, and long-term trend features of the load data are extracted in parallel using convolutional kernels of different scales, achieving multi-level perception of complex load changes. Specific correlations are as follows: A 1×16 convolutional kernel (kernel size 1, 16 channels) is used to extract local detail features of the load, corresponding to the instantaneous changes in the support load. Through a single time step convolution operation, it captures abrupt changes in load over a very short time. A 3×16 convolutional kernel (kernel size 3, 16 channels) is used to extract short-term dependency features of the load, corresponding to load fluctuations within the time required for periodic changes in the support load over a short period. Through three time steps convolution operation, it captures the short-term correlation between the load and the rhythm of coal mining operations. A 5×16 convolutional kernel (kernel size 5, 16 channels) is used to extract long-term trend features of the load, corresponding to the cumulative changes in the support load over a longer period. Through five time steps convolution operation, it captures the long-term trend of load changes related to the accumulation of roof pressure. The feature maps output from the three convolutional layers are fused through a concatenation operation to form a multi-scale load feature sequence, providing comprehensive feature input for subsequent attention weighting.

[0022] Step 4: Enhance the expressive power of load feature representation by jointly modeling static and dynamic attention mechanisms. Dynamic attention dynamically adjusts the weight distribution based on the time step of the input features. Integration is performed through weighted fusion to consider both global load information and key local load information. The specific process and its relationship with the support load are as follows: Input the spliced ​​and fused feature signal. , , , W The attention weights are the merged values.

[0023] DY=G(S(δ(P(Y T ))))SY=Conv(TCN(Y)) W=SY⊙DY In the formula: P(·), S(·) and G(·) are one-dimensional convolution with 1 channel, normalization operation and global average pooling (used to compress the dimension of multi-scale load features and retain key load information); DY and SY are the outputs of dynamic and static attention respectively (DY corresponds to the dynamic weight of load features at different time steps, such as the high weight at abnormal load time, and SY corresponds to the long-term stable load feature weight, such as the basic weight in the normal load range).

[0024] Attention weights fully consider the guidance of global information and can effectively highlight key load features (such as anomalous loads and periodic load peaks). However, as the network depth increases, the ability to express load features weakens. Therefore, residual connections are introduced to preserve the original information while enhancing the ability to identify critical load values. The output of the residual connection... Y’ Represented as: In the formula, The load feature sequence is the attention-weighted version; Y The original multi-scale load feature sequence is used, and residual connections are made to ensure that key information in the original load features is not lost, thereby improving the model's adaptability to complex load changes.

[0025] Step 5: Utilize a Long Short-Term Memory (LSTM) neural network to establish a deep dependency between the feature sequence and the stent load, capturing long-term temporal patterns in the load data and achieving accurate prediction of stent load in future periods. The LSTM input gate controls the proportion of load features at the current moment entering the LSTM memory unit, calculated as follows: In the formula: x t This indicates the output of the residual connection in step four. Y 'In the t The load characteristic values ​​at each time step reflect the multi-scale fusion information of the support load at the current moment; This includes the load prediction results from the previous moment and historical load trends. It is the sigmoid function; and For input gate weights; and Forget gate weights. The feature update process for the forget gate: In the formula: and To update the weights of the state; This refers to the core payload information from the previous moment. For the Hadamard product. The update process of the output gate: In the formula: Information to be retained; and Weight for the forget gate.

[0026] Step Six: Model Training and Validation: Construct samples from the preprocessed feature sequences using "time window sliding sampling" (time window size n=12, sliding step m=5) to obtain the dataset. Optimize the model parameters using the adaptive moment estimation optimizer, employing root mean square error, coefficient of determination, and mean absolute error as evaluation metrics, and validate the model's prediction accuracy and generalization ability.

[0027] This invention effectively integrates the local details, short-term fluctuations, and long-term trends of hydraulic support loads through a multi-scale convolution module, significantly improving the model's ability to perceive complex load changes. Through the static and dynamic attention mechanisms of the time attention module, the influence of key time segments is adaptively strengthened, increasing the model's attention to important features. By establishing long-term dependencies in the load sequence through deep time series modeling, accurate prediction of support loads in future periods is achieved.

[0028] Example 2 The present invention also provides a hydraulic support load prediction device, comprising: The first processing module is used to collect the load time series data of the hydraulic support of the coal mining face, and at the same time, to normalize the load time series data. The second processing module is used to decompose the preprocessed load time series data to obtain multiple intrinsic mode components; at the same time, it identifies and removes redundant components to construct an effective feature input sequence. The third processing module is used to extract local detail features, short-term dependency features and long-term trend features of the payload data in parallel using convolution kernels of different scales; The fourth processing module is used to focus on long-term stable key features through a static attention mechanism, and at the same time, to adaptively adjust the weight distribution according to the feature input sequence through a dynamic attention mechanism to obtain the output of the residual connection. The fifth processing module is used to establish a deep dependency relationship between the feature input sequence and the load time series data based on the output of the residual connection using a long short-term memory neural network, so as to achieve accurate prediction of the stent load in future periods.

[0029] Preferably, the second processing module uses an adaptive noise complete set empirical mode decomposition algorithm to decompose the preprocessed load data and obtain multiple intrinsic mode components.

[0030] Preferably, the second processing module identifies and removes redundant components based on the KL divergence criterion to construct an effective feature input sequence.

[0031] Example 3 The present invention also provides a hydraulic support load prediction system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes a hydraulic support load prediction method when executed by the processor.

[0032] Example 4 The present invention also provides a storage medium storing a computer program that executes a hydraulic support load prediction method when running.

[0033] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A hydraulic support load prediction method characterized by, The method comprises the following steps: Step S1, collecting the load time series data of the hydraulic support of the coal mining face, and simultaneously performing normalization processing on the load time series data; Step S2, decomposing the preprocessed load time series data to obtain a plurality of intrinsic mode components; simultaneously, identifying and removing redundant components to construct an effective feature input sequence; Step S3, extracting local detail features, short-term dependence features and long-term trend features of the load data in parallel through different scale convolution kernels; Step S4, focusing on the long-term stable key features through a static attention mechanism, and simultaneously adjusting the weight distribution adaptively according to the feature input sequence through a dynamic attention mechanism to obtain the output of the residual connection; Step S5, establishing the deep dependence relationship between the feature input sequence and the load time series data by using a long short-term memory neural network according to the output of the residual connection, and realizing accurate prediction of the support load in the future period.

2. The hydraulic support load prediction method of claim 1, wherein, In step S2, the preprocessed load data is decomposed to obtain a plurality of intrinsic mode components by using an adaptive noise complete ensemble empirical mode decomposition algorithm.

3. The hydraulic support load prediction method of claim 2, wherein, In step S2, the redundant components are identified and removed based on the K-L divergence criterion to construct an effective feature input sequence.

4. A hydraulic support load prediction device characterized by comprising: The method comprises the following steps: A first processing module is configured to collect the load time series data of the hydraulic support of the coal mining face, and simultaneously perform normalization processing on the load time series data; A second processing module is configured to decompose the preprocessed load time series data to obtain a plurality of intrinsic mode components; simultaneously, identify and remove redundant components to construct an effective feature input sequence; A third processing module is configured to extract local detail features, short-term dependence features and long-term trend features of the load data in parallel through different scale convolution kernels; A fourth processing module is configured to focus on the long-term stable key features through a static attention mechanism, and simultaneously adjust the weight distribution adaptively according to the feature input sequence through a dynamic attention mechanism to obtain the output of the residual connection; A fifth processing module is configured to establish the deep dependence relationship between the feature input sequence and the load time series data by using a long short-term memory neural network according to the output of the residual connection, and realize accurate prediction of the support load in the future period.

5. The hydraulic support load prediction device as claimed in claim 4, wherein The second processing module decomposes the preprocessed load data to obtain a plurality of intrinsic mode components by using an adaptive noise complete ensemble empirical mode decomposition algorithm.

6. The hydraulic support load prediction device as claimed in claim 5, wherein The second processing module identifies and removes redundant components based on the K-L divergence criterion to construct an effective feature input sequence.

7. A hydraulic support load prediction system, characterised in that, The method comprises the following steps: A memory and a processor, wherein the memory stores a computer program which is run by the processor, and the computer program performs the hydraulic support load prediction method according to any one of claims 1-3 when being run by the processor.

8. A storage medium, characterized by The storage medium stores a computer program, and the computer program performs the hydraulic support load prediction method according to any one of claims 1-3 when being run.

Images