A hydraulic blasting safety risk early warning system based on deep learning

Abstract

The present invention discloses a hydraulic blasting safety risk early warning system based on deep learning, which belongs to the field of hydraulic blasting safety early warning and includes: a data acquisition module, which is used to collect safety parameter data of hydraulic blasting in real time; a data processing module, which is connected to the data acquisition module, which is used to first integrate the safety parameter data, and then process and analyze the data to obtain a target data set; a model construction module, which is connected to the data processing module, which is used to construct an initial multimodal network model, and train the initial multimodal network model based on the target data set to obtain a target multimodal network model; a monitoring and prediction module, which is connected to the model construction module, which is used to perform real-time safety monitoring of hydraulic blasting based on the target multimodal network model, and trigger corresponding control measures according to the monitoring results. The present invention significantly improves the safety risk early warning capability of hydraulic blasting, and provides a strong guarantee for the safety of hydraulic blasting projects.

Description

Technical Field

[0001] The present invention belongs to the field of hydraulic blasting safety warning, and in particular, relates to a hydraulic blasting safety risk warning system based on deep learning. Background Art

[0002] As an effective means of rock crushing, hydraulic blasting has been widely used in mining, tunneling and other projects. By controlling the water pressure, the intensity and range of the blasting can be effectively controlled, thereby reducing the impact on the surrounding environment. However, due to the various unstable factors involved in the blasting process, such as the uncertainty of rock properties and the fluctuation of blasting parameters, hydraulic blasting has certain safety risks. Improper operation may lead to serious consequences such as equipment damage and casualties. Traditional risk assessment methods mainly rely on empirical judgment and simple models, which are difficult to fully capture the complexity and variability of the blasting process, resulting in insufficient accuracy and real-time warning. At the same time, existing safety risk warning systems mostly rely on manual monitoring and simple sensor data. These systems often respond slowly, cannot achieve real-time monitoring and accurate warning, and have difficulty processing complex data and identifying potential risk patterns. Therefore, the existing technology has the following defects:

[0003] Incomplete data collection: Existing systems usually only collect limited physical parameters, such as pressure values, temperature, etc., and ignore other factors that may affect safety, such as environmental changes and equipment aging.

[0004] Limited data processing capabilities: Traditional early warning systems rely on manual experience and lack efficient data analysis capabilities. They are unable to extract useful information from large amounts of complex data, resulting in low accuracy of early warnings.

[0005] Imperfect early warning mechanism: Due to delays in data processing and analysis, the existing system is unable to achieve rapid response. The release of early warning information often lags behind the time when the risk occurs. There is a lack of effective early warning mechanism, and it is impossible to predict and respond to possible security risks in a timely manner.

[0006] Lack of self-learning ability: The existing system is unable to self-optimize based on historical data and experience, and the early warning model is updated and iterated slowly, making it difficult to adapt to changing working conditions. Summary of the invention

[0007] In order to solve the above technical problems, the present invention provides a water pressure blasting safety risk early warning system based on deep learning, comprising:

[0008] A data acquisition module is used to collect safety parameter data of hydraulic blasting in real time through a sensor network arranged in the blasting area;

[0009] A data processing module, connected to the data acquisition module, is used to first integrate the security parameter data, and then process and analyze the data to obtain a target data set;

[0010] A model building module, connected to the data processing module, for building an initial multimodal network model, and training the initial multimodal network model based on the target data set to obtain a target multimodal network model;

[0011] The monitoring and prediction module is connected to the model building module and is used to perform real-time safety monitoring of water pressure blasting based on the target multimodal network model and trigger corresponding control measures according to the monitoring results.

[0012] Preferably, the data acquisition module includes a first acquisition unit, a second acquisition unit, and a third acquisition unit;

[0013] The first acquisition unit is used to acquire hydraulic geological data of hydraulic blasting based on distributed sensors;

[0014] The second acquisition unit is used to obtain the surrounding environment data of the water pressure blasting;

[0015] The third collection unit is used to collect equipment operation data through regular and continuous monitoring.

[0016] Preferably, the distributed sensors include at least a water pressure sensor, a temperature sensor, a humidity sensor, a vibration sensor, and an acoustic wave sensor, which correspondingly collect water pressure parameters, temperature parameters, humidity parameters, vibration parameters, and acoustic wave parameters of hydraulic blasting;

[0017] The second acquisition unit includes at least a meteorological sensor, which is used to monitor the temperature, humidity, rainfall, wind speed, wind direction, and air pressure conditions in the blasting area;

[0018] The equipment operation data includes, but is not limited to, pressure vessel corrosion degree data, pump wear data, and sealing performance data.

[0019] Preferably, the data processing module includes a preprocessing unit and a feature extraction unit;

[0020] The preprocessing unit is used to perform multi-source data fusion on the security parameter data through a data fusion algorithm to obtain fused data; and then perform noise filtering and data cleaning on the fused data;

[0021] The feature extraction unit is used to extract features from the data processed by the preprocessing unit.

[0022] Preferably, the pre-processing unit includes a data synchronization processing sub-unit and a data fusion sub-unit;

[0023] The data synchronization processing subunit is used to perform synchronization and alignment processing on sensor data with different sampling rates and viewing angles in time and space to obtain time series data;

[0024] The data fusion subunit is used to synthesize the associated probability distribution of each sensor through a multi-Bayesian estimation method, provide a final fusion value, and perform multi-source data fusion according to the final fusion value to obtain fused data.

[0025] Preferably, the preprocessing unit further includes a data cleaning subunit, a data standardization subunit, and a data filtering subunit;

[0026] The data cleaning subunit is used to remove sensor noise through mean filtering, replace missing values ​​in sensor data with specific values ​​through the nan_to_num function by interpolation, and then identify and remove outliers through the Z-score method, or identify and correct outliers;

[0027] The data standardization subunit is used to perform data normalization by using StandardScaler of the Scikit-learn library to convert the data into a form with a mean of 0 and a standard deviation of 1; and then perform data standardization to convert the data into a form with zero mean and unit variance;

[0028] The data filtering subunit is used to perform filtering based on a low-pass filter combined with an exponential smoothing method to remove high-frequency noise.

[0029] Preferably, the data fusion subunit includes a priori probability allocation subunit, a likelihood function determination subunit, a posterior probability calculation subunit, a fused posterior probability distribution subunit, and a fusion value calculation subunit;

[0030] The prior probability allocation subunit is used to allocate a prior probability distribution to each sensor, and the prior probability distribution is used to reflect the cognition of the target variable in the absence of observation data;

[0031] The likelihood function determination subunit is used to determine the likelihood function of each sensor, and the likelihood function is used to describe the probability distribution of the target variable under the condition of given observation data;

[0032] The posterior probability calculation subunit is used to calculate the posterior probability distribution by combining the prior probability distribution and the likelihood function using the Bayesian formula; the posterior probability distribution is used to reflect the cognition of each sensor on the target variable when the observation data is available;

[0033] The fusion posterior probability distribution subunit is used to synthesize the posterior probability distributions of all sensors into a joint posterior probability distribution, and the joint posterior probability distribution is the final data fusion result;

[0034] The fusion value calculation subunit is used to calculate the final fusion value according to the fused joint posterior probability distribution.

[0035] Preferably, the model construction module includes a model construction unit, a model training unit, and a model fusion unit;

[0036] The model building unit is used to build a convolutional neural network model and a recurrent neural network model respectively;

[0037] The model training unit is used to train the convolutional neural network model based on the target data set, and then use the output of the trained convolutional neural network model as the input of the recurrent neural network model to train the recurrent neural network model to obtain the target multimodal network model.

[0038] Preferably, the model building unit includes a first building unit and a second building unit;

[0039] The first construction unit is used to construct two identical convolutional neural network models, and add a BatchNorm2d layer, a nonlinear ReLU layer and an attention mechanism module after the two-dimensional convolution layer in the convolutional neural network model to generate a twin adversarial network model;

[0040] The second construction unit is used to generate a convolutional neural network group based on the twin adversarial network model.

[0041] Preferably, the model training unit further includes a layer selection unit and a loss function introduction unit;

[0042] The layer selection unit is used to delete the input layer and the output layer of the multimodal network model;

[0043] The loss function introduction unit is used to introduce the cross entropy loss function and the attention mechanism module into the multimodal network model to reconstruct the model and obtain the target multimodal network model.

[0044] Compared with the prior art, the present invention has the following advantages and technical effects:

[0045] The hydraulic blasting safety risk early warning system of the present invention can significantly improve the safety risk early warning capability of hydraulic blasting, specifically including:

[0046] Real-time: The system can monitor and analyze key parameters in the blasting process in real time and respond to emergencies in a timely manner. The system's real-time data processing capability ensures immediate response to risks and greatly shortens the warning time.

[0047] Accuracy: The deep learning model can accurately predict the safety risks that may occur during the blasting process, reduce false positives and false negatives, and can more accurately identify and predict potential safety risks.

[0048] Adaptability: The system can adapt to different blasting environments and geological conditions, improving the versatility and reliability of the early warning system. At the same time, the adaptive learning mechanism enables the system to continuously evolve according to new data and situations, improving the stability and reliability of long-term operation.

[0049] Safety: Comprehensive data collection and analysis reduce the possibility of human negligence and misjudgment, providing a higher level of safety for hydraulic blasting operations.

[0050] Through deep learning technology, the present invention realizes real-time monitoring and analysis of various parameters in the hydraulic blasting process, significantly improves the accuracy and timeliness of the early warning system, and provides a strong guarantee for the safety of hydraulic blasting projects. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0052] Figure 1 Schematic diagram of the system structure of an embodiment of the present invention. DETAILED DESCRIPTION

[0053] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0054] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0055] like Figure 1 As shown, this embodiment provides a water pressure blasting safety risk early warning system based on deep learning, including:

[0056] A data acquisition module is used to collect safety parameter data of hydraulic blasting in real time through a sensor network arranged in the blasting area;

[0057] The data processing module is connected to the data acquisition module and is used to first integrate the safety parameter data, and then process and analyze the data to obtain the target data set;

[0058] A model building module, connected to the data processing module, is used to build an initial multimodal network model and train the initial multimodal network model based on a target data set to obtain a target multimodal network model;

[0059] The monitoring and prediction module is connected with the model building module and is used to perform real-time safety monitoring of hydraulic blasting based on the target multimodal network model and trigger corresponding control measures according to the monitoring results.

[0060] Further, the data acquisition module includes a first acquisition unit, a second acquisition unit, and a third acquisition unit;

[0061] The first acquisition unit is used to collect hydraulic geological data of hydraulic blasting based on distributed sensors;

[0062] The second acquisition unit is used to obtain the surrounding environment data of the water pressure blasting;

[0063] The third collection unit is used to collect equipment operation data through regular and continuous monitoring.

[0064] Furthermore, in order to comprehensively assess safety risks, the system should integrate data from different sensors. Distributed sensors should include at least water pressure sensors, temperature sensors, humidity sensors, vibration sensors, and acoustic wave sensors, which should collect water pressure parameters, temperature parameters, humidity parameters, vibration parameters, and acoustic wave parameters of hydraulic blasting to ensure that all factors that may affect safety are taken into account.

[0065] The second acquisition unit includes at least a meteorological sensor, which is used to monitor the temperature, humidity, rainfall, wind speed, wind direction, and air pressure conditions in the blasting area; by deploying special environmental monitoring sensors to track environmental changes, these data can help the system evaluate the potential impact of weather conditions on hydraulic blasting operations.

[0066] Equipment operation data includes but is not limited to pressure vessel corrosion data, pump wear data, and seal performance data. Through regular and continuous monitoring, equipment operation data is collected to assess the health of the equipment and predict potential failures. All sensors are calibrated and maintained regularly to ensure data accuracy and consistency. Eliminating sensor failures or deviations that may lead to unreliable data can further improve the effectiveness of the early warning system.

[0067] In addition, the present embodiment can also dynamically adjust the frequency of data collection according to the operating conditions and risk levels. For example, during high-risk operations or when environmental conditions are severe, the sampling frequency is increased to obtain more detailed monitoring. Through the above-mentioned improvement measures, the data collection process will be more comprehensive and accurate, and various factors affecting the safety of hydraulic blasting can be better captured and analyzed, thereby improving the effectiveness of the early warning system.

[0068] Further, the data processing module includes a preprocessing unit and a feature extraction unit;

[0069] The preprocessing unit is used to fuse the multi-source data of the security parameter data through the data fusion algorithm to obtain the fused data; then the fused data is subjected to noise filtering and data cleaning;

[0070] The feature extraction unit is used to extract features from the data processed by the preprocessing unit.

[0071] Further, the pre-processing unit includes a data synchronization processing sub-unit and a data fusion sub-unit;

[0072] The data synchronization processing subunit is used to synchronize and align sensor data with different sampling rates and viewing angles in time and space to obtain time series data;

[0073] The data fusion subunit is used to synthesize the associated probability distribution of each sensor through the multi-Bayesian estimation method, provide the final fusion value, and perform multi-source data fusion according to the final fusion value to obtain fused data.

[0074] Furthermore, the preprocessing unit also includes a data cleaning subunit, a data standardization subunit, and a data filtering subunit;

[0075] The data cleaning subunit is used to remove sensor noise through mean filtering. For missing values ​​in sensor data, the missing values ​​are replaced with specific values ​​through the nan_to_num function using interpolation, and then outliers are identified and removed or identified and corrected through the Z-score method.

[0076] The data standardization subunit is used to normalize the data by using the StandardScaler of the Scikit-learn library to convert the data into a form with a mean of 0 and a standard deviation of 1; then data standardization is performed to convert the data into a form with zero mean and unit variance;

[0077] The data filtering subunit is used to perform filtering based on a low-pass filter combined with an exponential smoothing method to remove high-frequency noise.

[0078] Further, the data fusion subunit includes a priori probability allocation subunit, a likelihood function determination subunit, a posterior probability calculation subunit, a fused posterior probability distribution subunit, and a fused value calculation subunit;

[0079] The prior probability assignment subunit is used to assign a prior probability distribution to each sensor, and the prior probability distribution is used to reflect the cognition of the target variable in the absence of observed data;

[0080] The likelihood function determination subunit is used to determine the likelihood function of each sensor, and the likelihood function is used to describe the probability distribution of the target variable under the condition of given observation data;

[0081] The posterior probability calculation subunit is used to calculate the posterior probability distribution by combining the prior probability distribution and the likelihood function using the Bayesian formula; the posterior probability distribution is used to reflect each sensor's cognition of the target variable when the observation data is available;

[0082] The fusion posterior probability distribution subunit is used to synthesize the posterior probability distributions of all sensors into a joint posterior probability distribution, which is the final data fusion result;

[0083] The fusion value calculation subunit is used to calculate the final fusion value according to the fused joint posterior probability distribution.

[0084] Furthermore, the model construction module includes a model construction unit, a model training unit, and a model fusion unit;

[0085] The model building unit is used to build a convolutional neural network model and a recurrent neural network model respectively;

[0086] The model training unit is used to train the convolutional neural network model based on the target data set, and then use the output of the trained convolutional neural network model as the input of the recurrent neural network model to train the recurrent neural network model to obtain the target multimodal network model.

[0087] Further, the model building unit includes a first building unit and a second building unit;

[0088] The first construction unit is used to construct two identical convolutional neural network models, adding a BatchNorm2d layer, a nonlinear ReLU layer and an attention mechanism module after the two-dimensional convolution layer in the convolutional neural network model to generate a twin adversarial network model;

[0089] The second building unit is used to generate a convolutional neural network group based on the twin adversarial network model.

[0090] Furthermore, the model training unit also includes a layer selection unit and a loss function introduction unit;

[0091] The layer selection unit is used to delete the input layer and output layer of the multimodal network model;

[0092] The loss function introduction unit is used to introduce the cross entropy loss function and the attention mechanism module into the multimodal network model to reconstruct the model and obtain the target multimodal network model.

[0093] Specifically, the convolutional neural network model (CNN) of this embodiment specifically includes:

[0094] Based on the convolution layer, multiple convolution kernels are used to slide calculations on the input data to extract local features. Each convolution kernel represents a specific feature identifier;

[0095] The activation function uses the ReLU function to increase nonlinearity, which helps to learn more complex features.

[0096] The pooling layer reduces the feature dimension through maximum pooling or average pooling while retaining important information and enhancing the spatial invariance of features.

[0097] The fully connected layer expands and fully connects the outputs of the convolutional layer and the pooling layer for the final classification or regression task.

[0098] Specifically, this embodiment uses a recurrent neural network (RNN) model to extract time-dependent features, including:

[0099] The recurrent hidden layer captures the temporal dependencies in the sequence data through recurrent connections, memorizes the information of the previous time step, and combines it with the input of the current time step.

[0100] The output layer generates the output of each time step based on the cyclic update of the hidden state, and performs sequence generation or labeling. The LSTM or GRU unit is used to alleviate the gradient disappearance problem, so that the model can capture long-term dependencies.

[0101] This embodiment uses the Adam optimizer to adjust the model weights of the target multimodal network model, calculates the gradient of the loss function with respect to the parameters through the back propagation algorithm, and updates the model parameters to minimize the loss. Then, the model performance and generalization ability are optimized by adjusting hyperparameters such as the learning rate, batch size, and number of iterations.

[0102] The fused data is input into the trained target multimodal network model, and the stability of the rock mass is analyzed and judged in real time to predict the safety risks in the blasting process, detect anomalies and potential safety hazards in time, and provide corresponding early warning information and decision support.

[0103] Early warning includes triggering corresponding early warning mechanisms based on risk prediction results, such as sound and light alarms, automatic cessation of blasting operations, etc., to prevent and control safety risks.

[0104] More specifically, the early warning mechanism triggering process includes:

[0105] Threshold setting: According to the safety requirements and historical data of blasting operations, the early warning threshold of the risk level is set. When the risk level predicted by the model exceeds this threshold, the early warning mechanism will be triggered.

[0106] Real-time monitoring: The system needs to monitor the risk level of the model output in real time. Once the risk level exceeds the warning threshold, the warning mechanism will be activated immediately. Real-time monitoring can be achieved through scheduled tasks or event triggering to ensure the timeliness and accuracy of the warning mechanism.

[0107] Multi-level warning: Design a multi-level warning mechanism to trigger different levels of warnings according to different risk levels. For example, when the risk level is medium risk, a warning prompt can be triggered to remind the operator to pay attention; when the risk level is high risk, an emergency stop operation can be triggered to ensure the safety of personnel and equipment.

[0108] Warning information transmission: Warning information needs to be delivered to relevant personnel in a timely and accurate manner. This embodiment ensures that information can be received and understood in a timely manner through various means such as sound and light alarms, SMS notifications, and system pop-up windows. Warning information should include risk levels, possible impacts, and recommended response measures to help operators make correct decisions.

[0109] Emergency response: After the early warning mechanism is triggered, corresponding emergency response measures are required, including stopping blasting operations, evacuating personnel, checking equipment status, etc., to ensure the safety of personnel and equipment. Emergency response measures should be formulated with detailed operation procedures and division of responsibilities based on risk assessment results and early warning information.

[0110] Feedback mechanism: After the early warning mechanism is triggered, feedback information is collected to evaluate the accuracy of the early warning and the effectiveness of the emergency response. By using this feedback information to optimize the model and early warning mechanism, the early warning effect in the future can be improved. The feedback mechanism can be implemented through log records, user surveys or expert reviews to ensure its comprehensiveness and objectivity. At the same time, in order to improve the effectiveness of the risk assessment and early warning module, the system needs to be continuously optimized. This includes model updates, parameter adjustments, user interactions and other aspects. The model update of this embodiment can be achieved through retraining or incremental learning to adapt to new data distributions and changes in the blasting environment. Parameter adjustment optimizes the accuracy and timeliness of risk assessment based on the early warning effect and actual needs. User interaction allows operators to adjust early warning parameters according to actual conditions, or provide feedback to help the system better adapt to different operating conditions by providing a user interface.

[0111] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A hydraulic blasting safety risk early warning system based on deep learning, characterized in that: include: A data acquisition module is used to collect safety parameter data of hydraulic blasting in real time through a sensor network arranged in the blasting area; A data processing module, connected to the data acquisition module, is used to first integrate the security parameter data, and then process and analyze the data to obtain a target data set; A model building module, connected to the data processing module, for building an initial multimodal network model, and training the initial multimodal network model based on the target data set to obtain a target multimodal network model; A monitoring and prediction module, connected to the model building module, is used to perform real-time safety monitoring of hydraulic blasting based on the target multimodal network model and trigger corresponding control measures according to the monitoring results; The data processing module includes a preprocessing unit and a feature extraction unit; The preprocessing unit is used to perform multi-source data fusion on the security parameter data through a data fusion algorithm to obtain fused data; Then, the fused data is subjected to noise filtering and data cleaning; The feature extraction unit is used to extract features from the data processed by the preprocessing unit; The pre-processing unit includes a data synchronization processing sub-unit and a data fusion sub-unit; The data synchronization processing subunit is used to perform synchronization and alignment processing on sensor data with different sampling rates and viewing angles in time and space to obtain time series data; The data fusion subunit is used to synthesize the associated probability distribution of each sensor through a multi-Bayesian estimation method, provide a final fusion value, and perform multi-source data fusion according to the final fusion value to obtain fused data; The preprocessing unit also includes a data cleaning subunit, a data standardization subunit, and a data filtering subunit; The data cleaning subunit is used to remove sensor noise through mean filtering, replace missing values ​​in sensor data with specific values ​​through the nan_to_num function by interpolation, and then identify and remove outliers through the Z-score method, or identify and correct outliers; The data standardization subunit is used to perform data normalization by using StandardScaler of the Scikit-learn library to convert the data into a form with a mean of 0 and a standard deviation of 1; and then perform data standardization to convert the data into a form with zero mean and unit variance; The data filtering subunit is used to perform filtering based on a low-pass filter combined with an exponential smoothing method to remove high-frequency noise; The data fusion subunit includes a priori probability allocation subunit, a likelihood function determination subunit, a posterior probability calculation subunit, a fused posterior probability distribution subunit, and a fusion value calculation subunit; The prior probability allocation subunit is used to allocate a prior probability distribution to each sensor, and the prior probability distribution is used to reflect the cognition of the target variable in the absence of observation data; The likelihood function determination subunit is used to determine the likelihood function of each sensor, and the likelihood function is used to describe the probability distribution of the target variable under the condition of given observation data; The posterior probability calculation subunit is used to calculate the posterior probability distribution by combining the prior probability distribution and the likelihood function using the Bayesian formula; the posterior probability distribution is used to reflect the cognition of each sensor on the target variable when the observation data is available; The fusion posterior probability distribution subunit is used to synthesize the posterior probability distributions of all sensors into a joint posterior probability distribution, and the joint posterior probability distribution is the final data fusion result; The fusion value calculation subunit is used to calculate the final fusion value according to the fused joint posterior probability distribution; The model construction module includes a model construction unit, a model training unit, and a model fusion unit; The model building unit is used to build a convolutional neural network model and a recurrent neural network model respectively; The model training unit is used to train the convolutional neural network model based on the target data set, and then use the output of the trained convolutional neural network model as the input of the recurrent neural network model to train the recurrent neural network model to obtain the target multimodal network model.

2. The hydraulic blasting safety risk early warning system based on deep learning according to claim 1 is characterized in that: The data acquisition module includes a first acquisition unit, a second acquisition unit, and a third acquisition unit; The first acquisition unit is used to acquire hydraulic geological data of hydraulic blasting based on distributed sensors; The second acquisition unit is used to obtain the surrounding environment data of the water pressure blasting; The third collection unit is used to collect equipment operation data through regular and continuous monitoring.

3. The water pressure blasting safety risk early warning system based on deep learning according to claim 2 is characterized in that: The distributed sensors include at least a water pressure sensor, a temperature sensor, a humidity sensor, a vibration sensor, and an acoustic wave sensor, which correspondingly collect water pressure parameters, temperature parameters, humidity parameters, vibration parameters, and acoustic wave parameters of hydraulic blasting; The second acquisition unit includes at least a meteorological sensor, which is used to monitor the temperature, humidity, rainfall, wind speed, wind direction, and air pressure conditions in the blasting area; The equipment operation data includes, but is not limited to, pressure vessel corrosion degree data, pump wear data, and sealing performance data.

4. The water pressure blasting safety risk early warning system based on deep learning according to claim 1 is characterized in that: The model building unit includes a first building unit and a second building unit; The first construction unit is used to construct two identical convolutional neural network models, and add a BatchNorm2d layer, a nonlinear ReLU layer and an attention mechanism module after the two-dimensional convolution layer in the convolutional neural network model to generate a twin adversarial network model; The second construction unit is used to generate a convolutional neural network group based on the twin adversarial network model.

5. The water pressure blasting safety risk early warning system based on deep learning according to claim 1 is characterized in that: The model training unit also includes a layer selection unit and a loss function introduction unit; The layer selection unit is used to delete the input layer and the output layer of the multimodal network model; The loss function introduction unit is used to introduce the cross entropy loss function and the attention mechanism module into the multimodal network model to reconstruct the model and obtain the target multimodal network model.

Images