Engineering logistics digital safety production inspection method and system

Abstract

The invention relates to the technical field of safety production inspection, and particularly discloses an engineering logistics digital safety production inspection method and system, and the method comprises the steps: obtaining equipment state data of engineering logistics equipment, carrying out the processing of the equipment state data, obtaining a multi-vector set, carrying out the normalization processing of the multi-vector set, obtaining a high-dimensional multi-feature set, and carrying out the verification of the high-dimensional multi-feature set; calculating and obtaining an abnormal probability density value through a Gaussian mixture model based on the comprehensive representation, and comparing the abnormal probability density value with an abnormal threshold value to obtain an abnormal signal; calculating a comprehensive representation prediction value of the equipment state data, and calculating to obtain a prediction error value; according to the method, the prediction error value and the safety risk threshold value are compared, an early warning signal is generated, multi-dimensional data is used for calculation, the accuracy of equipment operation state inspection is improved, equipment operation evaluation is more comprehensive, and equipment can be predicted based on historical data before obvious faults and safety risks occur in the equipment, so that a maintenance decision is made in advance, and the maintenance efficiency is improved. And greater safety risks caused by abnormities are avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of production safety inspection, and in particular to a digital production safety inspection method and system for engineering logistics. Background Art

[0002] With the rise of e-commerce, the modern logistics industry has also developed rapidly. At present, the volume of logistics transportation has increased dramatically. As a key link in the transportation process, the safety of transportation vehicles has become increasingly prominent, especially in the field of engineering logistics. Higher requirements are placed on the safety of engineering transportation vehicles at all stages of transportation tasks.

[0003] In the digital production process, the stability and safety of equipment operation are extremely important, but the safety inspection in the production process is basically carried out by manual inspection, and the comprehensive digitalization is relatively low. Some digital inspections are carried out by manual troubleshooting after abnormalities occur, which is inefficient and cannot predict the working status of the equipment after the abnormality occurs. Usually, the equipment is shut down for inspection directly, which seriously affects production efficiency. In addition, some digital inspection methods are all carried out from a single angle, and the inspection results are seriously one-sided and limited, and the accuracy of the inspection results is low. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for digital production safety inspection of engineering logistics to solve the above-mentioned problems.

[0005] The purpose of the present invention can be achieved through the following technical solutions: A method for digital production safety inspection of engineering logistics, comprising the following steps: Step 1: Mark the equipment in the digital safety production process of engineering logistics, and mark the equipment as i, where i is 1, 2, 3...; Taking unit time t as the time interval, obtaining device state data at time T, including representation state data, appearance feature data and sound feature data, and processing the device state data to obtain a multi-vector set, the multi-vector set including a representation state vector set, an appearance feature vector set and a sound feature vector set; Step 2: Normalize the multi-vector set to obtain a high-dimensional multi-feature set, including state features, appearance features, and sound features, and then perform dimensionality reduction calculation on the high-dimensional multi-feature set to obtain a comprehensive representation; Step 3: Train to obtain a Gaussian mixture model based on comprehensive representation , the abnormal probability density value is obtained by calculating the Gaussian mixture model , and compare it with the abnormal threshold PY to obtain the abnormal signal; Step 4: Calculate the comprehensive representation prediction value of the equipment status data based on the abnormal signal , and obtain the comprehensive representation of time T+1 , calculate the prediction error value H T ; and compare the predicted error value with the safety risk threshold HY, and generate an early warning signal.

[0006] As a further solution of the present invention: the method of obtaining the characterizing state data is: obtaining the temperature value W of the device through the temperature sensor i (T) Vibration value V of the device obtained by the vibration sensor i (T) and the displacement value D of the device obtained by the displacement sensor i (T), the representation state data is processed and stored as a representation state vector set ,Right now , where T=n*t, n is 0, 1, 2, 3…

[0007] As a further solution of the present invention: the appearance feature data is obtained by using a high-speed camera to obtain the device appearance image at time T with a unit time t as the time interval, which is used to identify abnormal phenomena on the device surface, and then the device appearance image is extracted by a convolutional neural network to obtain a high-dimensional feature representation, and then processed by a ResNet feature extractor to obtain an appearance feature vector set , .

[0008] As a further solution of the present invention: the sound feature data is obtained by: obtaining the noise detection area through a sound collection device, obtaining the acoustic signal at time T, and obtaining the Mel frequency cepstral coefficient value through Fourier edge exchange and feature extraction processing, thereby obtaining the sound feature vector set ; Where m is the acoustic feature dimension, m is 1, 2, 3…, and e is the cepstral coefficient value.

[0009] As a further solution of the present invention: the method of obtaining the comprehensive characterization calculation is: A021: Based on the representation state vector set , get the minimum value of the characterization state vector set and represents the maximum value of the state vector set , and then normalize it; pass Calculate the state characteristics ; A022: Based on appearance feature vector set , and then get the maximum value of the appearance feature vector set , and then perform maximum value normalization processing; pass Calculate the appearance features ; A023: Based on sound feature vector set , using normalization method, calculate the sound characteristics ; A024: Then use dimensionality reduction calculation to obtain a comprehensive representation of the high-dimensional multi-feature set ; .

[0010] As a further solution of the present invention: the training method of the Gaussian mixture model includes: A031: Initialize Gaussian weights for each Gaussian component , and use random samples to calculate each Gaussian mean vector , and calculate the Gaussian distribution covariance matrix ; A032: Calculate the Gaussian distribution probability of each comprehensive feature; A033: Update Gaussian weights based on Gaussian distribution probability , Gaussian mean vector and the Gaussian distribution covariance matrix ; A034: Repeat A032 and A033 until the parameter change is less than the convergence threshold ξ, and obtain the Gaussian mixture model, which is: ;in, is the probability density of the equipment status data; Among them, K is the number of Gaussian distribution values, which can be understood as the number of mixed components, that is, the status data of the device; is the kth Gaussian distribution.

[0011] As a further solution of the present invention: a probability density value of the device is calculated based on the device status data, and then the probability density value is compared with the abnormal threshold: If the probability density value is less than the abnormal threshold, an abnormal signal is generated; If the probability density value is greater than the abnormal threshold, a normal signal is generated.

[0012] As a further solution of the present invention: the prediction error value H T The calculation method is: A041: Obtain the comprehensive representation from the initial detection time to time T-1 within the detection cycle, and then calculate the comprehensive representation prediction value of the device status data at time T+1 through the LSTM learning model ; Specific: ; A042: Maintain the equipment running status and obtain comprehensive representation at time T+1 ; A043: Then through Calculate the prediction error value H T .

[0013] As a further solution of the present invention: the prediction error value H T Compare with the security risk threshold HY: If the prediction error value H T If it is greater than the safety risk threshold HY, a warning signal is generated; If the prediction error value H T If it is greater than the safety risk threshold HY, a hold detection signal is generated.

[0014] As a further solution of the present invention: a digital production safety inspection system for engineering logistics, comprising: Data acquisition module: Mark the equipment in the digital safety production process of engineering logistics, and mark the equipment as i, where i is 1, 2, 3...; Taking unit time t as the time interval, obtaining device state data at time T, including representation state data, appearance feature data and sound feature data, and processing the device state data to obtain a multi-vector set, the multi-vector set including a representation state vector set, an appearance feature vector set and a sound feature vector set; Data processing module: normalize the multi-vector set to obtain a high-dimensional multi-feature set, including state features, appearance features, and sound features, and then perform dimensionality reduction calculation on the high-dimensional multi-feature set to obtain a comprehensive representation; Abnormal judgment module: Using EM algorithm, training to obtain Gaussian mixture model, based on comprehensive representation , the abnormal probability density value is obtained by calculating the Gaussian mixture model , and compare it with the abnormal threshold PY to obtain the abnormal signal; Risk prediction module: Based on abnormal signals, calculate the comprehensive representation prediction value of equipment status data , and obtain the comprehensive representation of time T+1 , calculate the prediction error value H T ; and compare the predicted error value with the safety risk threshold HY, and generate an early warning signal.

[0015] Beneficial effects of the present invention: The present invention is based on the equipment status data obtained from multiple dimensions, and through data feature fusion, a unified feature vector is constructed, and the data features are reduced in dimension with principal component analysis to obtain more accurate and stable equipment status characterization data, and then a Gaussian mixture model is trained on the equipment data, and then it is calculated whether the equipment under the current equipment status data is abnormal, which solves the problem of incomplete inspection of a single data dimension, can capture complex abnormal patterns for different equipment states, and adapt to the inspection of the operating status of multiple devices. At the same time, the use of multi-dimensional data calculation can effectively solve the problem of the nonlinear relationship between the equipment operating status and the influencing factors, improve the accuracy of the equipment operating status inspection, and make the equipment operation evaluation more comprehensive; By performing predictive analysis and calculation on the equipment operation status after an abnormality, it is possible to determine whether the equipment has an operation risk problem, solve the problem of low efficiency of relying on manual physical inspection of the equipment when the equipment status data is abnormal, and solve the problem of potential failures in manual physical inspection; Moreover, after an equipment anomaly occurs, predictions can be made about the equipment based on historical data before the equipment exhibits obvious failures and safety risks, so that maintenance decisions can be made in advance to avoid greater safety risks caused by the anomaly, which can lead to further production losses. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The present invention will be further described below in conjunction with the accompanying drawings.

[0017] Figure 1 It is a schematic flow chart of the method of the present invention; Figure 2 It is a system block diagram of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] Example 1: Please refer to Figure 1 As shown, the present invention is a method for digital production safety inspection of engineering logistics, comprising the following steps: Step 1: Mark the equipment in the digital safety production process of engineering logistics, and mark the equipment as i, where i is 1, 2, 3...; Taking unit time t as the time interval, obtaining device state data at time T, including representation state data, appearance feature data and sound feature data, and processing the device state data to obtain a multi-vector set, including a representation state vector set, an appearance feature vector set and a sound feature vector set; A011: Get the data representing the status, specifically, the temperature value W of the device obtained by the temperature sensor i (T) Vibration value V of the device obtained by the vibration sensor i (T) and the displacement value D of the device obtained by the displacement sensor i (T), the representation state data is processed and stored as a representation state vector set ,Right now , where T=n*t, n is 0, 1, 2, 3, etc.; the characterization status data is the operating status and ambient temperature of the equipment during operation, which can be obtained through detection by various sensors deployed around the equipment to realize the detection of the equipment and ensure the operating status of the equipment; A012: Acquire appearance feature data. Specifically, the appearance image of the device at time T is acquired by a high-speed camera with a unit time t as the time interval, which is used to identify abnormal phenomena on the surface of the device. The abnormal phenomena include but are not limited to cracks, corrosion, and wear. The abnormal phenomena are represented by n. For example, when the value of n is 1, it represents a crack phenomenon, and when the value of n is 2, it represents a corrosion phenomenon. The abnormal phenomenon is set by the inspector. Then, the device appearance image is extracted by a convolutional neural network to obtain a high-dimensional feature representation, and then processed by the ResNet feature extractor to obtain an appearance feature vector set. , ; Real-time images of the scene can be obtained through smart cameras, and then whether there is any abnormality based on the status of the equipment in the image; A013: Obtain sound feature data. Specifically, obtain the noise detection area through the sound collection device, record in real time, obtain the acoustic signal at time T, and obtain the Mel frequency cepstral coefficient value through Fourier edge exchange and feature extraction processing, and then obtain the sound feature vector set. ; Wherein, m is the acoustic feature dimension, including but not limited to sharp sound, low-frequency vibration, etc. Similarly, m is 1, 2, 3..., and numbers can be used to represent different acoustic feature dimensions respectively, and e is the cepstrum coefficient value; the sound feature data is used to express the dynamic effect of the equipment during operation; By acquiring equipment status data, the operating parameters are analyzed from multiple dimensions such as equipment load, vibration, temperature, etc., and abnormal reminders and risk warnings are provided to prevent abnormal problems such as damage during equipment operation and improve equipment operation efficiency; Step 2: Normalize the multi-vector set to obtain a high-dimensional multi-feature set, including state features, appearance features, and sound features, and then perform dimensionality reduction calculation on the high-dimensional multi-feature set to obtain a comprehensive representation; A021: Based on the representation state vector set , get the minimum value of the characterization state vector set and represents the maximum value of the state vector set , and then normalize it; pass Calculate the state characteristics ; A022: Based on appearance feature vector set , and then get the maximum value of the appearance feature vector set , and then perform maximum value normalization processing; pass Calculate the appearance features ; A023: Based on sound feature vector set , the sound features are calculated using the same normalization method as the state vector set ; A024: Then use dimensionality reduction calculation to obtain a comprehensive representation of the high-dimensional multi-feature set ; Specifically, ; Step 3: Train to obtain a Gaussian mixture model based on comprehensive representation , the abnormal probability density value is obtained by calculating the Gaussian mixture model , and compare it with the abnormal threshold PY to obtain the abnormal signal; Specifically: The training method of Gaussian mixture model includes: A031: Initialize Gaussian weights for each Gaussian component , and use random samples to calculate each Gaussian mean vector , and calculate the Gaussian distribution covariance matrix ; A032: Calculate the Gaussian distribution probability of each comprehensive feature; A033: Update Gaussian weights based on Gaussian distribution probability , Gaussian mean vector and the Gaussian distribution covariance matrix ; A034: Repeat A032 and A033 until the parameter change is less than the convergence threshold ξ, and obtain the Gaussian mixture model, which is: ;in, is the probability density of the equipment status data; Among them, K is the number of Gaussian distribution values, which can be understood as the number of mixed components, that is, the status data of the device; is the kth Gaussian distribution; The convergence threshold in the Gaussian mixture model training process is set by the staff based on the inspection experience and historical performance of the inspection accuracy. If the convergence threshold is too large, the model algorithm will quickly end the training during the Gaussian mixture model training process, and the accuracy of the results will be reduced. If the convergence threshold is too small, the model algorithm is prone to enter an infinite loop. Therefore, in determining the convergence threshold, it is necessary to adjust and determine it based on historical data to ensure the accuracy of the results of the Gaussian mixture model in the subsequent calculation process; The probability density value of the device is calculated based on the current device status data, and then the probability density value is compared with the abnormal threshold: If the probability density value is less than the abnormal threshold, an abnormal signal is generated. At this time, it means that the device has an abnormality during operation at the current moment and an abnormal reminder is required; If the probability density value is greater than the abnormal threshold, a normal signal is generated, indicating that the device is in a normal and stable state during operation, and no early warning is required, and monitoring can be continued; Based on the equipment status data obtained from multiple dimensions, a unified feature vector is constructed through data feature fusion, and the data features are reduced in dimension with the help of principal component analysis to obtain more accurate and stable equipment status representation data. The Gaussian mixture model is then trained on the equipment data, and then it is calculated whether the equipment under the current equipment status data is abnormal. This solves the problem of incomplete inspection of a single data dimension, can capture complex abnormal patterns for different equipment states, and adapt to the inspection of the operating status of multiple devices. At the same time, the use of multi-dimensional data calculation can effectively solve the problem of the nonlinear relationship between the equipment operating status and the influencing factors, improve the accuracy of the equipment operating status inspection, and make the equipment operation evaluation more comprehensive.

[0020] Embodiment 2: Based on the above embodiment, in this embodiment, if an abnormality is detected in the equipment during the inspection process, in order to determine whether there is a risk if the abnormal equipment continues to operate, this embodiment uses a deep learning model to predict the operating status of the equipment. Specifically: Step 4: Calculate the comprehensive representation prediction value of the equipment status data based on the abnormal signal , and obtain the comprehensive representation of time T+1 , calculate the prediction error value H T ; and compare the predicted error value with the safety risk threshold HY, and generate an early warning signal; Specifically, regarding the comprehensive representation prediction value The calculation method is: A041: Obtain the comprehensive representation from the initial detection time to time T-1 within the detection cycle, and then calculate the comprehensive representation prediction value of the device status data at time T+1 through the LSTM learning model ; Specific: ; A042: Maintain the equipment running status and obtain comprehensive representation at time T+1 ; A043: Then through Calculate the prediction error value H T ; The prediction error value H T Compare with the security risk threshold HY: If the prediction error value H T If it is greater than the safety risk threshold HY, a warning signal is generated; If the prediction error value H T If it is greater than the safety risk threshold HY, a hold detection signal is generated; By performing predictive analysis and calculation on the equipment operation status after an abnormality, it is possible to determine whether the equipment has an operation risk problem, solve the problem of low efficiency of relying on manual physical inspection of the equipment when the equipment status data is abnormal, and solve the problem of potential failures in manual physical inspection; Moreover, after an equipment anomaly occurs, predictive analysis can be used to make predictions about the equipment based on historical data before the equipment has obvious failures and safety risks, so that maintenance decisions can be made in advance to avoid greater safety risks caused by the anomaly, which can lead to further production losses.

[0021] Embodiment 3: Figure 2 As shown, a digital safety production inspection system for engineering logistics includes: Data acquisition module: Mark the equipment in the digital safety production process of engineering logistics, and mark the equipment as i, where i is 1, 2, 3...; Taking unit time t as the time interval, obtaining device state data at time T, including representation state data, appearance feature data and sound feature data, and processing the device state data to obtain a multi-vector set, the multi-vector set including a representation state vector set, an appearance feature vector set and a sound feature vector set; Data processing module: normalize the multi-vector set to obtain a high-dimensional multi-feature set, including state features, appearance features, and sound features, and then perform dimensionality reduction calculation on the high-dimensional multi-feature set to obtain a comprehensive representation; Abnormal judgment module: Using EM algorithm, training to obtain Gaussian mixture model, based on comprehensive representation , the abnormal probability density value is obtained by calculating the Gaussian mixture model , and compare it with the abnormal threshold PY to obtain the abnormal signal; Risk prediction module: Based on abnormal signals, calculate the comprehensive representation prediction value of equipment status data , and obtain the comprehensive representation of time T+1 , calculate the prediction error value H T ; and compare the predicted error value with the safety risk threshold HY, and generate an early warning signal; By analyzing multi-dimensional data such as equipment temperature, pressure, vibration, and sound, the assessment is more comprehensive, reducing blind spots caused by inspections using a single data source, and effectively solving the problem of inaccurate inspection results caused by the uncertainty of a single data source.

[0022] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A digital safety production inspection method for engineering logistics, characterized in that: The following steps are involved: Step 1: Mark the equipment in the digital safety production process of engineering logistics, and mark the equipment as i, where i is 1, 2, 3...; Taking unit time t as the time interval, obtaining device state data at time T, including representation state data, appearance feature data and sound feature data, and processing the device state data to obtain a multi-vector set, the multi-vector set including a representation state vector set, an appearance feature vector set and a sound feature vector set; Step 2: Normalize the multi-vector set to obtain a high-dimensional multi-feature set, including state features, appearance features, and sound features, and then perform dimensionality reduction calculation on the high-dimensional multi-feature set to obtain a comprehensive representation; Step 3: Train to obtain a Gaussian mixture model based on comprehensive representation , the abnormal probability density value is obtained by calculating the Gaussian mixture model , and compare it with the abnormal threshold PY to obtain the abnormal signal; Step 4: Calculate the comprehensive representation prediction value of the equipment status data based on the abnormal signal , and obtain the comprehensive representation of time T+1 , calculate the prediction error value H T ; and compare the predicted error value with the safety risk threshold HY, and generate an early warning signal.

2. A method for digital production safety inspection of engineering logistics according to claim 1, characterized in that: The method of obtaining the characterization status data is: obtaining the temperature value W of the device through the temperature sensor i (T) Vibration value V of the device obtained by the vibration sensor i (T) and the displacement value D of the device obtained by the displacement sensor i (T), the representation state data is processed and stored as a representation state vector set ,Right now , where T=n*t, n is 0, 1, 2, 3… 3. A method for digital production safety inspection of engineering logistics according to claim 1, characterized in that: The acquisition method of the appearance feature data is as follows: a high-speed camera is used to obtain the device appearance image at time T with a unit time t as the time interval, which is used to identify abnormal phenomena on the device surface. The device appearance image is then extracted by a convolutional neural network to obtain a high-dimensional feature representation, and then processed by a ResNet feature extractor to obtain an appearance feature vector set. , .

4. A method for digital production safety inspection of engineering logistics according to claim 1, characterized in that: The sound feature data is obtained by: obtaining the noise detection area through a sound collection device, obtaining the acoustic signal at time T, and obtaining the Mel frequency cepstral coefficient value through Fourier edge exchange and feature extraction processing, thereby obtaining the sound feature vector set ; Where m is the acoustic feature dimension, m is 1, 2, 3…, and e is the cepstral coefficient value.

5. A method for digital production safety inspection of engineering logistics according to claim 1, characterized in that: The method of obtaining the comprehensive characterization calculation is as follows: A021: Based on the representation state vector set , get the minimum value of the characterization state vector set and represents the maximum value of the state vector set , and then normalize it; pass Calculate the state characteristics ; A022: Based on appearance feature vector set , and then get the maximum value of the appearance feature vector set , and then perform maximum value normalization processing; pass Calculate the appearance features ; A023: Based on sound feature vector set , using normalization method, calculate the sound characteristics ; A024: Then use dimensionality reduction calculation to obtain a comprehensive representation of the high-dimensional multi-feature set ; 。 6. A method for digital production safety inspection of engineering logistics according to claim 1, characterized in that: The training method of the Gaussian mixture model includes: A031: Initialize Gaussian weights for each Gaussian component , and use random samples to calculate each Gaussian mean vector , and calculate the Gaussian distribution covariance matrix ; A032: Calculate the Gaussian distribution probability of each comprehensive feature; A033: Update Gaussian weights based on Gaussian distribution probability , Gaussian mean vector and the Gaussian distribution covariance matrix ; A034: Repeat A032 and A033 until the parameter change is less than the convergence threshold ξ, and obtain the Gaussian mixture model, which is: ;in, is the probability density of the equipment status data; Among them, K is the number of Gaussian distribution values, which can be understood as the number of mixed components, that is, the status data of the device; is the kth Gaussian distribution.

7. A method for digital production safety inspection of engineering logistics according to claim 6, characterized in that: The probability density value of the device is calculated based on the device status data, and then the probability density value is compared with the abnormal threshold: If the probability density value is less than the abnormal threshold, an abnormal signal is generated; If the probability density value is greater than the abnormal threshold, a normal signal is generated.

8. A method for digital production safety inspection of engineering logistics according to claim 1, characterized in that: The prediction error value H T The calculation method is: A041: Obtain the comprehensive representation from the initial detection time to time T-1 within the detection cycle, and then calculate the comprehensive representation prediction value of the device status data at time T+1 through the LSTM learning model ; Specific: ; A042: Maintain the equipment running status and obtain comprehensive representation at time T+1 ; A043: Then through Calculate the prediction error value H T .

9. A method for digital production safety inspection of engineering logistics according to claim 8, characterized in that: The prediction error value H T Compare with the security risk threshold HY: If the prediction error value H T If it is greater than the safety risk threshold HY, a warning signal is generated; If the prediction error value H T If it is greater than the safety risk threshold HY, a hold detection signal is generated.

10. A digital safety production inspection system for engineering logistics, characterized in that: The system is used to execute the engineering logistics digital production safety inspection method described in any one of claims 1 to 9, including: Data acquisition module: Mark the equipment in the digital safety production process of engineering logistics, and mark the equipment as i, where i is 1, 2, 3...; Taking unit time t as the time interval, obtaining device state data at time T, including representation state data, appearance feature data and sound feature data, and processing the device state data to obtain a multi-vector set, the multi-vector set including a representation state vector set, an appearance feature vector set and a sound feature vector set; Data processing module: normalize the multi-vector set to obtain a high-dimensional multi-feature set, including state features, appearance features, and sound features, and then perform dimensionality reduction calculation on the high-dimensional multi-feature set to obtain a comprehensive representation; Abnormal judgment module: Using EM algorithm, training to obtain Gaussian mixture model, based on comprehensive representation , the abnormal probability density value is obtained by calculating the Gaussian mixture model , and compare it with the abnormal threshold PY to obtain the abnormal signal; Risk prediction module: Based on abnormal signals, calculate the comprehensive representation prediction value of equipment status data , and obtain the comprehensive representation of time T+1 , calculate the prediction error value H T ; and compare the predicted error value with the safety risk threshold HY, and generate an early warning signal.

Images