Opencast coal mine motor equipment state monitoring and early warning method based on digital twinning
By adopting a digital twin-based edge cloud collaborative architecture and deep learning algorithm methods in mining motors, accurate prediction and early warning of motor state is achieved, and the problem of insufficient processing and prediction of complex environmental factors in the existing technology is solved, and the accuracy and real-timeness of motor fault detection are improved.
Patent Information
- Application Number
- CN202510208043.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-25
- Publication Date
- 2025-06-10
AI Technical Summary
The prior art is difficult to effectively deal with complex external environmental factors in the fault monitoring of mining motors, and lacks systematicity, macroscopicity and predictability, making it difficult to achieve accurate motor status prediction and early warning.
The status monitoring and early warning method of open-pit coal mine motor equipment based on digital twins is adopted, combined with edge-cloud collaborative architecture, digital twin model and deep learning algorithm to achieve prediction and real-time monitoring of motor diversified states. Specific steps include obtaining IoT sensor data, performing preprocessing and feature extraction, using the AM-CNN-GRU model for state prediction, and setting an early warning threshold through the Gaussian kernel function.
It realizes accurate quantitative prediction of motor status, improves prediction accuracy and stability of results, can detect motor abnormal status in advance, provides comprehensive decision-making guidance, and makes up for the shortcomings of traditional methods in real-time, spatial and data value.
Smart Images
Figure CN120124467A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of fault warning of mine motors, and more specifically, to a method for monitoring and warning the state of open-pit coal mine motor equipment based on digital twin. Background Art
[0002] The stable operation of mine motors has a significant impact on the efficiency of the entire production line. Accurate prediction of the motor state is crucial for early warning of potential faults and active maintenance plan management.
[0003] Currently, the research based on digital twin mainly focuses on big data-driven monitoring, with more research in aspects such as device networking, device virtualization, and service-oriented modeling, but lacks systematicness, macroscopicity, and predictability. How to monitor and predict the operating state of production equipment connected to the cloud from a macroscopic perspective based on a large amount of data accumulated in the cloud from the perspective of edge-cloud collaboration and give full play to the advantages of edge-cloud computing is one of the issues worthy of in-depth discussion.
[0004] In addition, due to external constraint conditions such as the operating environment and load of mine motors, the problems of the cloudification mode, principle, method, etc. of motor equipment in the edge-cloud environment are quite different from the traditional device networking, virtualization, and service-oriented problems. There is still a need for in-depth research on how to unify factors such as device networking, virtualization, and services, integrate new information technologies to establish an edge-cloud collaboration theoretical framework and technical system, and how to predict the multiple states of mine motors and provide systematic theories and methods for state-based maintenance services in the edge-cloud environment.
[0005] Therefore, in the fault monitoring of mine motors, one of the major challenges is how to handle complex external environmental factors (such as the operating load and environmental conditions of the motor), and at the same time, how to combine big data, Internet of Things technology, and virtualization and service-oriented modeling methods to establish a more accurate and effective prediction model. These traditional methods sometimes fail to meet the requirements of real-time, spatial, and data value, especially in the remote monitoring and management of motor equipment. Summary of the Invention
[0006] In view of this, the present invention provides a method for monitoring and warning the state of motor equipment in open-pit coal mines based on digital twins, realizing the multi-state prediction of mine motors that integrates digital twin models and deep learning algorithms under the edge-cloud collaborative architecture. From a macroscopic perspective, it monitors and predicts the operating state of motor equipment connected to the cloud, studies the big data-driven state prediction and maintenance method based on edge-cloud collaboration, and solves the problems of the massive volume, diversity, real-time nature, spatial nature, and value of data in big data-driven Cyber-Physical Systems (CPS). The proposed architecture avoids the weaknesses of traditional modeling that consider model construction, update, spatio-temporal characteristics, and prediction accuracy in isolation, can fully express time constraints and spatial environmental conditions, and integrally realizes concepts such as complex calculations, networked real-time communication, and decision-making in a big data environment. Studying the real-time monitoring and state prediction of mine motors in the edge-cloud environment has important theoretical significance and application value for further understanding the internal laws of equipment cloud monitoring. Solving this problem will provide a theoretical basis for the monitoring and warning of mine motors connected to the industrial big data cloud platform.
[0007] To achieve the above object, the present invention adopts the following technical solutions:
[0008] A method for monitoring and warning the state of motor equipment in open-pit coal mines based on digital twins, comprising the following steps:
[0009] S1. Obtain monitoring data from Internet of Things sensors, divide the data into a training set and a validation set, and perform preprocessing; the data includes: current, temperature, and vibration parameters of the motor;
[0010] S2. Train a state prediction model in the cloud, and construct an AM-CNN module for feature extraction;
[0011] S3. Take the extracted features as the input of the GRU prediction model to obtain the feature prediction values of the motor in the future period;
[0012] S4. Compare the feature prediction values with the corresponding actual values, and use the root mean square error of the feature residuals as the state index of the motor;
[0013] S5. Based on the prediction residuals, set the warning threshold using a Gaussian kernel function; the prediction residuals are the differences between the feature prediction values and the corresponding actual values;
[0014] S6. Preprocess the real-time monitoring data in the edge layer, and input it into the trained AM-CNN-GRU model to obtain the motor operating state index to be detected;
[0015] S7. Calculate the root mean square error of the characteristic residuals of the motor operation data to be detected, and compare it with the warning threshold. If it does not exceed, it is judged to be in a normal state and continuous monitoring is carried out. If it exceeds the threshold, it is judged to be in an abnormal state and a warning is triggered.
[0016] As can be seen from the above technical solutions, compared with the prior art, the present invention has the following technical advantages:
[0017] 1. The present invention establishes a motor state prediction framework based on Digital Twin (DT), and uses edge-cloud collaboration technology to provide data support for the Digital Twin model, and conducts motor state prediction from a systematic and macroscopic perspective.
[0018] 2. The present invention realizes quantitative prediction of the motor state. The adopted AM-CNN-GRU model captures local features through the combination of a convolutional neural network (CNN), captures long-term dependencies through a gated recurrent unit (GRU), and focuses on key information through an attention mechanism (AM), realizing accurate and comprehensive data understanding. This fusion advantage not only improves the prediction accuracy, reduces errors in multi-tasks, but also enhances the stability of the results. In addition, in terms of computational efficiency, by reducing the number of parameters and accelerating the convergence process, the AM-CNN-GRU model efficiently utilizes computational resources and comprehensively demonstrates excellent performance.
[0019] 3. The present invention constructs an AM-CNN-GRU model based on twin data to realize quantitative prediction of the motor state, and realizes multi-state prediction of mine motors from the perspectives of single-parameter prediction and multi-parameter fusion respectively; the analysis results provide all-round and multi-angle decision-making guidance for motor state warning, giving more comprehensive and intuitive results than single prediction methods, making up for the shortcoming that sensors cannot directly measure the internal state variables of the motor, and realizing early fault prediction of the motor. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained according to the provided drawings without creative efforts.
[0021] Figure 1 It is a flowchart of the method for monitoring and warning the state of open-pit coal mine motor equipment based on Digital Twin provided by the present invention;
[0022] Figure 2 It is a schematic diagram of the virtual-real mapping scheme of mine motor equipment provided by the present invention;
[0023] Figure 3Schematic diagram of the digital twin-driven motor status prediction data stream provided by the present invention;
[0024] Figure 4 Architecture diagram of the AM-CNN-GRU model provided by the present invention;
[0025] Figure 5 Schematic diagram of the prediction residuals of the training set provided by the present invention;
[0026] Figure 6 Schematic diagram of the probability distribution of the prediction residuals provided by the present invention;
[0027] Figure 7 Schematic diagram of the state prediction results of the MCNN-SE-GRU model provided by the present invention;
[0028] Figure 8 Schematic diagram of the state prediction results when the motor is operating normally provided by the present invention. Specific embodiments
[0029] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.
[0030] Refer to Figure 1 As shown, an open-pit coal mine motor equipment status monitoring and early warning method based on digital twin is disclosed in an embodiment of the present invention, including the following steps: S1 to S7;
[0031] S1. Obtain the monitoring data from the Internet of Things sensors, divide the data into a training set and a validation set; and perform preprocessing; the data includes: the current, temperature, and vibration parameters of the motor;
[0032] This step first collects the real-time operation data of the motor, such as temperature, vibration, current, etc., from the Internet of Things sensors as input data. These data are the basis for subsequent analysis and modeling. The data is divided into a training set (for training the model) and a validation set (for verifying the effect of the model) to ensure that the model can generalize better. Preprocessing includes data cleaning (such as removing noise and filling missing values), data standardization (ensuring that different features have similar scales), and data transformation, etc. Preprocessing is to ensure data quality and improve the performance of the model.
[0033] S2. Train the state prediction model in the cloud and construct an AM-CNN module for feature extraction;
[0034] Model training is carried out in the cloud. Using the model AM-CNN that combines the attention mechanism and the convolutional neural network (CNN), by introducing the attention mechanism, it can better capture the key information in the motor operating state, thereby improving the classification accuracy. The AM-CNN module extracts useful features from the original data to provide information for subsequent state prediction.
[0035] Among them, the AM-CNN module includes:
[0036] The convolutional layer is used to perform convolutional operations on the one-dimensional matrix composed of multi-dimensional time series; the operation process is as shown in Equation (1);
[0037]
[0038] Among them, represents the j-th element of the l-th layer, M j represents the j-th convolutional region of the feature map of the (l - 1)-th layer, is the element among them; represents the weight matrix of the l-th layer; represents the bias, f represents the non-linear ReLU activation function, as shown in Equation (2);
[0039] f(x) = max(0, x) (2)
[0040] The pooling layer is used to perform pooling on the convolutional calculation results, that is, downsampling; as shown in Equation (3);
[0041]
[0042] Among them and respectively represent the weight and bias of the j-th feature map in the l-th layer; down(·) represents the downsampling function; here, max pooling with shift-invariance characteristics is used to reduce the dimension of the output feature map.
[0043] The channel attention mechanism is used to calculate the importance of each channel to the state prediction result.
[0044] Introduce the channel attention mechanism to calculate the importance of each channel to the state prediction result. Through global average pooling, two excitation operations are used to estimate the weights for each channel;
[0045] s = F ex (z, W) = σ(g(z, W)) = σ(W 2 δ(W 1 , z)) (4)
[0046] In the formula, F ex(·) is a non - linear function; g(·) is the internal operation function of Fex(·); z is the real number after channel feature compression, W is the parameter of the fully - connected layer, σ is the S - type activation function, s is the channel weight value; δ is the ReLU activation function, W 1 ,W 2 is the weight matrix; the output of the channel attention mechanism (SE block) is obtained by re - scaling U using the activation s:
[0047]
[0048] where and F scale (u c ,s C ) represents the channel multiplication between the scalar s C and the feature map u C ;
[0049] S3. Take the extracted features as the input of the GRU prediction model to obtain the predicted feature values of the motor in the future time period;
[0050] GRU is a variant of the Recurrent Neural Network (RNN), which is good at processing time - series data. The features extracted by AM - CNN are passed to the GRU (Gated Recurrent Unit) model for time - series prediction. That is: the future state of the motor is predicted through the GRU model, that is, the operating state features of the motor in the future period are predicted.
[0051] Among them, the calculation steps of the forward propagation of the GRU network are as follows:
[0052] S31. Calculate the output value R of the reset gate t ; the previous - time hidden - state vector H t-1 and the current - time input value X t are passed to the reset gate together to obtain a reset - gate output value between 0 and 1. The larger the value, the more information is retained; the calculation process is shown in formula (6):
[0053] R t = σ ( X t w xr + H t-1 w hr + b r ) (6)
[0054] In the formula, σ represents the S - type activation function; w xr represents the weight value connected between the input layer and the reset gate; w hr represents the weight value connected between the hidden layer and the reset gate; b r represents the bias of the cell state in the reset gate;
[0055] S32. Calculate the output value Q of the update gate t and the candidate hidden state vector H t '; The current time input vector and the hidden state of the previous time enter the update gate together to calculate a value between 0 and 1 as the output. At the same time, a candidate hidden state is created through the activation function tanh. The calculation process is shown in formulas (7) and (8):
[0056] Q t = σ ( X t w xz + H t-1 w hz + b z ) (7)
[0057] H t '=tanh ( X t w xh + ( R t * H t-1 )w hh + b h ) (8)
[0058] In the formula, w xz represents the weight value connecting the input layer and the update gate; w hz represents the weight value connecting the hidden layer and the update gate; b z represents the bias of the cell state in the update gate; w xh represents the weight value connecting the input layer and the hidden layer; w hh represents the weight value connecting the hidden layer of the previous time and the hidden layer of the current time; b h represents the bias of the cell state in the hidden layer;
[0059] S33. Calculate the hidden state vector at the current time; Use the output value of the update gate as the weight vector, and the candidate hidden state vector and the hidden state vector of the previous time are weighted and averaged to obtain the hidden state vector H at the current time t , and the calculation process is shown in formula (9):
[0060] H t =Q t *H t-1 +(1 - Q t )*H′ t (9)
[0061] S34. Calculate the network output value Y t ; Add the weighted hidden state vector at the current time and the output layer bias, and then jointly output the final output value. The calculation process is shown in formula (10):
[0062] Y t = σ(H t w y + b y ) (10)
[0063] Wherein, w y represents the weight value of the unit state in the output layer; b y represents the bias of the unit state in the output layer.
[0064] S4. Compare the feature prediction value with the corresponding actual value, and use the root mean square error of the feature residual as the state index HI of the motor;
[0065] This step compares the predicted value with the actual value and calculates the root mean square error (RMSE), which is the square root mean of the prediction error. This error reflects the gap between the predicted value and the actual value and is used as an index (HI) of the motor state. The smaller the value of HI, the more accurate the prediction.
[0066] The calculation of the root mean square error of the feature residual is as follows:
[0067]
[0068] Wherein, HI represents the root mean square error of the feature residual; represents the feature value obtained by model prediction and is the predicted feature; y i represents the actual feature, and n represents the number of samples.
[0069] S5. Based on the prediction residual, set the warning threshold using the Gaussian kernel function; the prediction residual is the difference between the feature prediction value and the corresponding actual value;
[0070] Among them, the prediction residual refers to the difference between the predicted value and the actual value. The warning threshold is set through the Gaussian kernel function (usually used in methods such as support vector machines), that is, by learning the residual distribution of the data, a threshold is determined, and if the threshold is exceeded, it is judged that the motor has an abnormality.
[0071] The Gaussian kernel is used to smooth the data or map the data to a higher-dimensional space for more complex calculations. In this step, the Gaussian kernel function helps to set a reasonable threshold according to the distribution of the data.
[0072] Specifically, it includes:
[0073] Determine that the probability density function of the health index during normal operation of the motor is Then the probability density function based on kernel density estimation is as shown in Equation (12):
[0074]
[0075] where Z is the detection index of the unit state, and Z t is the monitoring index of the t-th sample data point in the sample data. The number of sample data N is T - β + 1, the width of the sliding window is β (i.e., the sliding window contains β time-point data), h is the bandwidth, T is the time length of the training sample, and K(.) is the kernel density function, as shown in Equation (13):
[0076]
[0077] where Z i represents the detection index of the i-th sample data point.
[0078] Set the confidence level to α, then the threshold of the generator's health index is t d , as shown in Equation (14):
[0079]
[0080] where P(t < t d ) represents the probability that the health index is less than the warning threshold.
[0081] S6. Preprocess the real-time monitoring data in the edge layer, input it into the trained AM-CNN-GRU model, and obtain the operating state index of the motor to be detected;
[0082] Among them, the real-time monitoring data in the edge layer refers to the real-time acquisition data of the motor. These data need to be preprocessed and then input into the trained AM-CNN-GRU model. The model outputs the current operating state of the motor based on these data. Edge computing can ensure real-time data processing and response, avoid sending all data to the cloud, and reduce latency.
[0083] S7. Calculate the root mean square error of the characteristic residuals of the operating data of the motor to be detected, compare it with the warning threshold. If it does not exceed, it is judged to be in a normal state and continuous monitoring is carried out. If it exceeds the threshold, it is judged to be in an abnormal state and a warning is triggered.
[0084] As Figure 2 shown, first, simulate the power loss of the motor based on the twin model, qualitatively predict whether the motor state is abnormal, and visualize the real-time operating state of the motor. Then construct a convolutional-gated neural network model based on channel attention, quantitatively calculate the health index of the motor, and predict the state of the motor.
[0085] As Figure 3As shown in the figure, the workflow of a device status prediction system based on Internet of Things sensor data is presented. The system is divided into two parts: the edge layer and the cloud layer. In the edge layer, there is historical data (which can be stored locally or in the cloud) and real-time monitoring data from Internet of Things sensors, and all this data needs to be preprocessed. In the cloud layer, the preprocessed data is divided into a training set and a test set. The training set is used to train the status prediction model (using the AM-CNN-GRU model), and the test set is used to evaluate the model. The model makes predictions through feature extraction to obtain prediction residuals, and compares the prediction residuals with the warning threshold. If it exceeds the threshold, it is determined as an abnormal state and the status prediction result is output; if it does not exceed, it is a normal state. The entire system realizes the effective monitoring and warning of device status.
[0086] As Figure 4 shown. First, based on AM-CNN, the spatial features of the motor status are extracted and used as the input of the GRU network to predict the change trend of the status parameters; the status is predicted according to the evaluation result of the residual between the predicted value and the actual measured value. The kernel density estimation method is used to calculate the distribution of the prediction residuals during normal operation of the motor, and then the threshold is set. Based on the real-time monitoring data in the edge layer, the motor status is predicted and compared with the threshold, and if it exceeds the threshold, a warning is triggered.
[0087] This method combines Internet of Things data, deep learning models (AM-CNN and GRU), prediction residual analysis, and edge-cloud collaborative computing. By establishing an accurate motor status prediction model and triggering a warning in the abnormal state, it effectively realizes the fault warning and status monitoring of motor devices.
[0088] In this embodiment, an induction motor in an open-pit coal mine in Xinjiang is taken as the research object to verify the proposed digital twin-driven status prediction method and achieve early warning. The AM-CNN-GRU model is used to predict the status indicators of the motor. Querying the device fault records in this open-pit coal mine, it can be known that the main motor of the belt conveyor system on the No. 1 production line issued a warning of overheating of the front bearing at 17:50 on October 21, 2021.
[0089] In this embodiment, the AM-CNN-GRU model is used to predict the status indicators of the motor. The stator current, winding temperature, and rear bearing temperature of the motor are used as relevant status variables, and the ambient temperature and the instantaneous quantity of the belt scale are used as time-varying working condition parameters to predict the front bearing temperature of the motor. The monitoring parameters of the motor are shown in Table 1:
[0090] Table 1 Measured variables
[0091]
[0092] The present invention uses the historical motor normal operation data (from July to September 2021) stored in the cloud and preprocessed as the training set, and the real-time data received by the edge layer as the test data to train and evaluate the model prediction effect. The data sampling frequency is 1 time per minute.
[0093] The model parameters of AM-CNN-GRU constructed based on the training set are shown in Table 2:
[0094] Table 2 Parameter table of AM-CNN-GRU model
[0095]
[0096] The preprocessed training set is divided into new sample sequences using a sliding time window with a size of 300. The new samples are input into the state prediction model for training. The number of training epochs is 100 and the batch size is 256. Calculate the prediction residuals and use kernel density estimation to calculate the health threshold of the motor. The calculation results of the prediction residuals of the training set are as Figure 5 shown, and the probability distribution is as Figure 6 shown, where Figure 5 shows the prediction residuals of the training set. The horizontal axis is the time step, and the vertical axis is the root mean square error (RMSE). In the figure, the prediction residuals changing with the time step are plotted with blue lines, and there is a red horizontal line as a reference line. Figure 6 shows the probability distribution of the prediction residuals. The horizontal axis is the root mean square error (RMSE), and the vertical axis is the probability density. In the figure, the probability density corresponding to different RMSE values is represented by green histograms. Based on formulas (12)-(14), the bandwidth of the kernel density estimation is calculated to be 0.1, the confidence level is 0.95, and the normalized state threshold is 0.457.
[0097] The main motor of the belt conveyor system on production line 1 issued a warning of overheating of the front bearing at 17:50 on October 21, 2021 (the 867th time step). It can be seen from the Figure 7 gray line that the temperature of the front bearing had been continuously rising for 40 minutes at this time, causing certain damage to the motor bearing. The digital twin-driven AM-CNN-GRU model proposed by the present invention triggered a warning immediately at the 826th time step, that is, when the temperature of the front bearing showed a continuous rising trend. At this time, the health index of the motor was 0.457. The state index corresponding to the 867th time step was 0.89, approaching the complete failure state. The prediction result of this model detected the motor anomaly 41 minutes earlier than the actual fault.
[0098] To verify the prediction effect of the proposed AM-CNN-GRU model of the present invention, it is respectively compared with the CNN-GRU, CNN-SE-GRU, and MCNN-GRU models. The prediction accuracy of the models in the training set is as Figure 8 shown. As can be seen from the figure, the AM-CNN-GRU model shows good tracking effect on the real temperature signal, especially with high prediction sensitivity at the peak and valley of the temperature signal change.
[0099] The prediction accuracy of each model is shown in Table 3.
[0100] Table 3 Prediction accuracy of the models
[0101] Model CNN-GRU CNN-SE-GRU MCNN-GRU AM-CNN-GRU Root Mean Square Error 0.206 0.024 0.146 0.016
[0102] The present invention proposes a method for monitoring and warning the state of motor equipment in open-pit coal mines based on digital twin technology. This method integrates a deep learning model and Internet of Things sensor data, extracts features through the AM-CNN module, and uses the GRU prediction model to predict the state features of the motor in the future period. In addition, this method also introduces a Gaussian kernel function to set the warning threshold to achieve real-time monitoring and warning of the abnormal state of the motor. The experimental results show that this method has high prediction accuracy and can effectively detect the abnormal state of the motor in advance, providing technical support for the maintenance management of motor equipment in open-pit coal mines.
[0103] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The same or similar parts among the various embodiments can be referred to each other. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method part.
[0104] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for monitoring and early warning the condition of open-pit coal mine motor equipment based on digital twins, characterized in that: The following steps are involved: S1, obtain monitoring data from IoT sensors and divide the data into training set and validation set; And preprocessing is performed; the data includes: current, temperature and vibration parameters of the motor; S2, train the state prediction model in the cloud and build the AM-CNN module for feature extraction; S3, using the extracted features as the input of the GRU prediction model to obtain the feature prediction value of the motor in the future period; S4, comparing the characteristic prediction value with the corresponding actual value, and taking the root mean square error of the characteristic residual as the state indicator of the motor; S5. Based on the prediction residual, a Gaussian kernel function is used to set a warning threshold; the prediction residual is the difference between the feature prediction value and the corresponding actual value; S6, preprocessing the real-time monitoring data in the edge layer, inputting the trained AM-CNN-GRU model, and obtaining the motor operation status index to be detected; S7. Calculate the root mean square error of the characteristic residual of the motor operation data to be detected, and compare it with the warning threshold. If it does not exceed the threshold, it is judged as a normal state and continuous monitoring is performed. If it exceeds the threshold, it is judged as an abnormal state and a warning is triggered.
2. The method for monitoring and early warning the state of open-pit coal mine motor equipment based on digital twin according to claim 1 is characterized in that: The step S2 comprises: S21, performing a convolution operation on the one-dimensional matrix composed of the multi-dimensional time series, the operation process is as shown in formula (1); in, represents the jth element of the lth layer, M j represents the jth convolution region of the feature map of layer l-1, It is an element of it; represents the weight matrix of the lth layer; represents the deviation, and f represents the nonlinear ReLU activation function, as shown in formula (2); f(x)=max(0,x) (2) S22, pooling the convolution calculation results, i.e. downsampling, as shown in formula (3); in and denote the weight and bias of the jth feature map in the lth layer respectively; down(·) denotes the downsampling function, which is used to reduce the dimension of the output feature map; S23, introduce the channel attention mechanism to calculate the importance of each channel to the state prediction result, and estimate the weight of each channel through global average pooling and two excitation operations; s=F ex (z,W)=σ(g(z,W))=σ(W2δ(W1,z)) (4) In the formula, F ex (·) is a nonlinear function; g(·) is the internal operation function of Fex(·); z is the real number after channel feature compression, W is the fully connected layer parameter, σ is the S-type activation function, s is the channel weight value, δ is the ReLU activation function, W1, W2 are weight matrices; the output of the channel attention mechanism is obtained by rescaling u using the activation s: in and F scale (u c ,s C ) represents the scalar s C and feature map u C The channel multiplication between .
3. The method for monitoring and early warning the state of open-pit coal mine motor equipment based on digital twin according to claim 1 is characterized in that: The step S3 comprises: The calculation steps of the forward propagation of the GRU network are as follows: S31, calculate the output value R of the reset gate t ; The hidden state vector H at the previous time t-1 and the input value X at the current time t The information is then passed to the reset gate to obtain a reset gate output value between 0 and 1. The larger the value, the more information is retained. The calculation process is shown in formula (6): R t =σ(X t w xr +H t-1 w hr +b r ) (6) Where σ represents the S-type activation function; w xr represents the weight value of the connection between the input layer and the reset gate; w hr represents the weight value of the connection between the hidden layer and the reset gate; b r Represents the bias amount for resetting the state of the unit in the gate; S32, calculate the output value Q of the update gate t and the candidate hidden state vector H t '; The current time input vector and the hidden state of the previous time enter the update gate together, calculate the output value from 0 to 1, and create a candidate hidden state through the activation function tanh. The calculation process is shown in formulas (7) and (8): Q t =σ(X t w xz +H t-1 w hz +b z ) (7) H t '=tanh(X t w xh +(R t *H t-1 )w hh +b h ) (8) In the formula, w xz Represents the weight value of the connection between the input layer and the update gate; w hz represents the weight value of the connection between the hidden layer and the update gate; b z represents the bias of the unit state in the update gate; w xh Represents the weight value of the connection between the input layer and the hidden layer; w hh represents the weight value of the connection between the previous hidden layer and the current hidden layer; b h The bias that represents the unit state in the hidden layer; S33, calculate the hidden state vector at the current time; use the update gate output value as the weight vector, and obtain the hidden state vector H at the current time by weighted average of the candidate hidden state vector and the hidden state vector at the previous time t , the calculation process is shown in formula (9): H t =Q t *H t-1 +(1-Q t )*H′ t (9) S34, calculate network output value Y t ; Add the weighted hidden state vector at the current time to the output layer bias, and then output the final output value together. The calculation process is shown in formula (10): Y t =σ(H t w y +b y (10) In the formula, w y Represents the weight value of the unit state in the output layer; b y Represents the bias of the unit state in the output layer.
4. The method for monitoring and early warning the state of open-pit coal mine motor equipment based on digital twin according to claim 1 is characterized in that: In step S4, the root mean square error of the feature residual is calculated as follows: Among them, HI represents the root mean square error of feature residuals; Represents the characteristic value obtained by model prediction, which is the prediction feature; y i represents the actual feature, and n represents the number of samples.
5. The method for monitoring and early warning the state of open-pit coal mine motor equipment based on digital twin according to claim 1 is characterized in that: The step S5 comprises: The probability density function of the health index when the motor is operating normally is determined as follows: The probability density function based on kernel density estimation is shown in formula (12): In the formula, Z is the unit status detection index, Z t is the monitoring indicator of the t-th sample data point in the sample data, the number of sample data N is T-β+1, the width of the sliding window is β, that is, the sliding window contains β time point data; h is the bandwidth, T is the time length of the training sample, and K(.) is the kernel density function, as shown in formula (13): In the formula, Z i is the detection index of the i-th sample data point; Assuming the confidence level is α, the motor health indicator threshold is t d , as shown in formula (14): In the formula, P(t <t d ) represents the probability that the health indicator is less than the warning threshold.