Intelligent risk evaluation system for enterprise safety production

Abstract

The invention relates to the technical field of intelligent risk evaluation, and discloses an intelligent risk evaluation system for enterprise safety production, which comprises a data real-time acquisition module, a data acquisition module, a data preprocessing module, a data storage module, a risk identification module, a feature extraction module and a model training module, the problem of overfitting occurring in model training is solved in a learning rate adjustment mode, the learning rate adjustment comprises a learning rate initialization module, an attenuation strategy module, a current learning rate module, a training progress monitoring module and a parameter adjustment module, and the learning rate initialization module is used for determining an initial learning rate; the attenuation strategy module dynamically adjusts the learning rate according to the training progress. The loop learning rate can accelerate initial training, the adaptive learning rate attenuation can automatically adjust the learning rate in the later period, the training process is more stable, overfitting is prevented by reducing the learning rate, and the stability of model training is kept in the later period.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent risk assessment, and in particular to an intelligent risk assessment system for enterprise production safety. Background Art

[0002] With the expansion of production scale and the improvement of technological complexity of modern enterprises, the risks faced by enterprises in production safety are increasing. Traditional risk assessment methods rely on manual inspection and empirical judgment, which have problems such as strong subjectivity, low efficiency and poor real-time performance. In order to improve the scientificity, accuracy and timeliness of risk assessment, enterprises have gradually introduced intelligent technology and developed an intelligent risk assessment system for enterprise production safety. The system realizes dynamic risk monitoring and intelligent assessment of the entire process of enterprise production safety through functions such as data collection, analysis and decision support.

[0003] After searching, application number 202311774118.5 discloses a risk intelligent assessment system for enterprise safety production, which belongs to the technical field of risk intelligent assessment system, including real-time data acquisition module, data collection module, data preprocessing module, data storage module, risk identification module, feature extraction module, model training module, model evaluation module, risk assessment module, risk level module, risk reporting module, risk warning module, and warning notification module. The above scheme can realize automatic identification and extraction of potential risks in enterprise safety production through the above modules, train risk assessment models using machine learning algorithms, conduct risk assessment and level classification according to model evaluation results, issue warning signals in time and notify relevant personnel to take corresponding warning response measures, record warning information and analysis results, improve the accuracy and real-time nature of risk assessment of enterprise safety production, and reduce enterprise production safety risks.

[0004] In the above technical solutions: the model may overfit during the training process, resulting in poor performance on new data. At the same time, different algorithms have different performances in different scenarios. How to choose the algorithm that best suits the enterprise's production safety characteristics and how to integrate multiple algorithms to improve the accuracy and stability of the model is still an important issue.

[0005] To this end, we propose a risk intelligent assessment system for enterprise safety production. Summary of the invention

[0006] The present invention mainly solves the above-mentioned technical problems and provides a risk intelligent assessment system for enterprise safety production.

[0007] In order to achieve the above-mentioned purpose, the present invention adopts the following technical scheme: a risk intelligent assessment system for enterprise safety production, comprising a real-time data acquisition module, a data collection module, a data preprocessing module, a data storage module, a risk identification module, a feature extraction module, a model training module, a model evaluation module and a risk assessment module. Through the collaborative work of multiple modules, comprehensive monitoring and risk assessment of the enterprise production environment can be achieved, and the overfitting problem in model training can be solved by adjusting the learning rate;

[0008] The learning rate adjustment includes a learning rate initialization module, a decay strategy module, a current learning rate module, a training progress monitoring module and a parameter adjustment module. The learning rate initialization module is used to determine the initial learning rate. The learning rate initialization module serves as the starting point of the entire training process and determines the initial learning speed of the model. The decay strategy module dynamically adjusts the learning rate according to the training progress. The function of the current learning rate module is to track and record the currently used learning rate in real time to ensure that the optimizer can apply the correct learning rate for parameter update, update the current learning rate in each training cycle and under specific conditions, construct a training progress monitoring module, collect training status information in real time, and provide input for the decay strategy module.

[0009] Preferably, the attenuation strategy module dynamically adjusts the learning rate according to the training progress, and the specific steps of adjusting the learning rate include:

[0010] S1: Initialize the learning rate. Set the initial learning rate before starting training.

[0011] S2: Select the decay strategy and determine the learning rate decay strategy to be used;

[0012] S3: Update the learning rate during training. At the end of each epoch, update the current learning rate according to the set decay strategy.

[0013] S4: Update the learning rate of the optimizer. After each update of the learning rate, ensure that the optimizer uses the new learning rate.

[0014] S5: Monitor and record the learning rate. Record the changes in the learning rate in the training log to facilitate subsequent analysis and adjustment.

[0015] Preferably, the attenuation strategy module combines the cyclic learning rate with the adaptive learning rate during the training process. The cyclic learning rate switches between high and low values ​​by adjusting the learning rate in different epochs, and the adaptive learning rate dynamically adjusts the learning rate according to actual gradient changes.

[0016] Preferably, the attenuation strategy module adds a patience value mechanism during the learning rate adjustment process, and automatically starts the learning rate attenuation strategy when the validation set performance has not been significantly improved within multiple epochs.

[0017] Preferably, the attenuation strategy module establishes a dynamic switching mechanism to switch the use strategy of cyclic learning rate and adaptive learning rate according to the training progress. In the early stage of training, the cyclic learning rate is used to quickly explore the solution space. In the middle and late stages of training, the adaptive learning rate strategy is gradually introduced to ensure stable convergence of the model.

[0018] Preferably, in the attenuation strategy module, the method of superimposing the cyclic learning rate and the adaptive learning rate includes:

[0019] S1: Initialize the adaptive learning rate optimizer;

[0020] S2: Set the cyclic learning rate strategy;

[0021] S3: Combined with the patience value mechanism;

[0022] S4: Dynamic switching.

[0023] Preferably, the attenuation strategy module adjusts the strategy through a learning rate, and the risk assessment module accurately identifies risk levels at different levels.

[0024] Preferably, the training progress monitoring module provides feedback for the learning rate adjustment, and the parameter adjustment module is responsible for applying the calculated new learning rate to the optimizer.

[0025] Preferably, there are close connections and interactions between the various modules in adjusting the learning rate, and together they constitute a dynamic learning rate adjustment system.

[0026] Beneficial Effects

[0027] The present invention provides a risk intelligent assessment system for enterprise safety production. It has the following beneficial effects:

[0028] (1) This risk intelligent assessment system for enterprise safety production dynamically adjusts the learning rate according to the training progress. The learning rate can be changed in a large range by cyclic learning rate, thereby helping the model to explore the global optimal solution and avoid falling into the local optimum. The adaptive learning rate dynamically adjusts the step size according to the gradient, which is particularly suitable for high-dimensional space or gradient sparse problems. The model parameters can be finely adjusted locally. The combination of the two can optimize the model in a large range while ensuring the fine adjustment of the local optimum. The cyclic learning rate can accelerate the initial training and avoid the problem of slow convergence caused by too small a learning rate. The adaptive learning rate decay automatically adjusts the learning rate in the later stage to make the training process smoother. By reducing the learning rate, overfitting can be prevented. The combination of the two can improve the model convergence speed and maintain the stability of the training in the later stage.

[0029] (2) This is a risk intelligent assessment system for enterprise safety production. In the risk intelligent assessment system for enterprise safety production, the ability of the model training module to stably converge to the global optimum is crucial to improving the overall performance and accuracy of the system. Stable convergence to the global optimum means that the system can make full use of historical data and real-time production safety related data to conduct accurate risk assessment, ensuring that the model not only performs well in local areas, but also makes the best decision globally. This helps to identify potential production safety hazards, reduce omissions or false alarms, and enable the system to detect potential risks more effectively.

[0030] (3) This intelligent risk assessment system for enterprise safety production has stable convergence to ensure the generalization ability of the model. The system can maintain high assessment accuracy and robustness when facing new data, ensuring that the enterprise can still carry out effective risk prediction and prevention and control in a changing production environment. Through stable model training and optimization, the risk assessment module can accurately identify different levels of risk according to different production conditions, environmental changes, equipment status and other factors, thereby providing strong support for enterprise decision-making.

[0031] (4) The intelligent risk assessment system for enterprise safety production can stably converge to the global optimum, which also means that the system's response in generating risk reports and issuing early warning notifications is more accurate and timely. Enterprises can rely on this system to formulate safety production strategies more efficiently, reduce the probability of accidents, and achieve the effect of improving the overall safety level of the enterprise. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the implementation of the present invention or the technical solution in the prior art, the following is a brief introduction to the drawings required for the implementation or the prior art description. Obviously, the drawings described below are only exemplary, and for ordinary technicians in this field, other implementation drawings can be derived from the provided drawings without creative work.

[0033] The structures, proportions, sizes, etc. disclosed in this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with the technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantial technical significance. Any structural modification, change in proportion or adjustment of size should still fall within the scope of the technical contents disclosed in the present invention without affecting the effects and purposes that can be achieved by the present invention.

[0034] Figure 1 This is a diagram of the model training module architecture of the present invention;

[0035] Figure 2A learning rate adjustment strategy diagram of the present invention;

[0036] Figure 3 Schematic diagram of the system performance evaluation process of the present invention. DETAILED DESCRIPTION

[0037] 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.

[0038] Embodiment: A risk intelligent assessment system for enterprise safety production, referring to Figure 1-Figure 3 , including real-time data acquisition module, data collection module, data preprocessing module, data storage module, risk identification module, feature extraction module, model training module, model evaluation module, and risk assessment module. The model training module may overfit during the training process, and the problem in model training can be solved by adjusting the learning rate;

[0039] Adjust the learning rate. Use the learning rate decay strategy during training to adjust the learning rate according to the training progress to avoid overfitting caused by excessive learning rate in the later stage of training. Adjust the learning rate, including the learning rate initialization module, decay strategy module, current learning rate module, training progress monitoring module and parameter adjustment module. There is a close connection and interaction between the various modules of learning rate adjustment, which together constitute a dynamic learning rate adjustment system. The function of the learning rate initialization module is to set the initial learning rate. As the starting point of the entire training process, the learning rate initialization module determines the initial learning speed of the model. The decay strategy module defines how the learning rate is adjusted as the training progresses, including the specific decay algorithm. The decay strategy provides a basis for the current learning rate module, enabling it to update the current learning rate in each training cycle or under specific conditions. The function of the current learning rate module is to track and record the currently used learning rate in real time to ensure that the optimizer can apply the correct learning rate for parameter update;

[0040] The training progress monitoring module works closely with other modules to collect training status information in real time, such as the current epoch, training loss, and validation set performance. This information provides the necessary input for the decay strategy module to determine whether the learning rate needs to be adjusted. The training progress monitoring module provides feedback for the learning rate adjustment, so that the model can respond dynamically according to the training situation. The parameter adjustment module is responsible for applying the calculated new learning rate to the optimizer to ensure that the latest learning rate is used for each update, thereby optimizing the learning effect of the model. The interrelationship and collaboration of these modules make the learning rate adjustment process flexible and efficient, which helps to improve the training performance and final effect of the model. By realizing intelligent adjustment of the learning rate during the training process, overfitting can be effectively avoided and the generalization ability of the model can be improved.

[0041] The specific steps to adjust the learning rate according to the training progress are as follows:

[0042] S1: Initialize the learning rate. Set the initial learning rate before starting training.

[0043] S2: Select the decay strategy and determine the learning rate decay strategy to be used;

[0044] S3: Update the learning rate during training. At the end of each epoch, or according to the set decay strategy, update the current learning rate.

[0045] S4: Update the learning rate of the optimizer. After each update of the learning rate, ensure that the optimizer uses the new learning rate.

[0046] S5: Monitor and record the learning rate. Record the learning rate changes in the training log to facilitate subsequent analysis and adjustment.

[0047] The validation set is used to decide how to select the decay strategy and plays an important role in selecting the learning rate decay strategy. The validation set is a part of the data separated from the training data. According to the performance monitoring of the validation set, if the model performance does not improve within a period of time, the decay strategy is selected and the learning rate is reduced to effectively prevent the training instability caused by excessive learning rate. Usually, a patience value is set. When the validation index does not improve in multiple consecutive epochs, the learning rate is reduced. In each training cycle, the learning rate changes cyclically between two boundaries, so that the model hovers near the global optimal solution and avoids overfitting. In one cycle, the learning rate gradually increases from the minimum value to the maximum value, and then decreases back to the minimum value.

[0048] Used to reduce the learning rate when the validation set performance has not improved for a period of time. The formula is:

[0049] new_Ir=initial,Ir×factor

[0050] initial,Ir: initial learning rate;

[0051] factor: the multiple by which the learning rate is reduced;

[0052] Condition: When the validation set loss or accuracy does not improve within the patience value, update the learning rate;

[0053] The learning rate continuously cycles between two boundaries (minimum and maximum) during training, and the formula is as follows:

[0054]

[0055] min-Ir is the lower bound of the learning rate;

[0056] max_Ir is the upper limit of the learning rate;

[0057] currentstep is the current step number (number of iterations in training);

[0058] cycle___steps is the total number of steps in a cycle. The learning rate changes cyclically between the maximum and minimum values. It is usually used to prevent the model from falling into the local optimum and improve the robustness of training.

[0059] Combine cyclic learning rate or adaptive learning rate decay. Cyclic learning rate can prevent the model from overfitting in local areas by adjusting the learning rate in different epochs and switching between high and low values. Adaptive learning rate will adjust the learning rate according to the actual gradient change and further reduce the learning rate to avoid overfitting in complex models. The combination of the two can effectively improve the generalization ability of the model. Adaptive learning rate performs well when processing data sets with high data sparsity, while cyclic learning rate can perform global search for training at different stages. The combination of the two can better handle the characteristics of different data sets, whether it is data with large gradient changes or problems that are difficult to converge quickly in the early stages of training.

[0060] The method of superimposing the cyclic learning rate and the adaptive learning rate is as follows:

[0061] S1: Initialize the adaptive learning rate optimizer. During model training, you can select the adaptive learning rate optimizer to dynamically adjust the learning rate according to the gradient changes. The adaptive learning rate algorithm calculates an independent learning rate for each parameter and automatically attenuates the impact of large gradients.

[0062] S2: Set a cyclic learning rate strategy: Based on the adaptive learning rate, a cyclic learning rate mechanism is introduced to set the minimum and maximum learning rate boundaries. The learning rate is gradually increased to the maximum value within a cycle, and then gradually decayed back to the minimum value. This method can work together with the adaptive learning rate, so that the model can not only adaptively adjust the learning rate, but also periodically explore the global optimal solution;

[0063] S3: Combined with the patience value mechanism, when the performance of the validation set does not improve significantly within multiple epochs, the learning rate decay strategy is automatically started. This ensures that if the model performance does not improve in a specific training phase, the system will adaptively reduce the learning rate further to help the model better fit the data.

[0064] S4: Dynamic switching, switching between the two strategies according to the training progress. For example, in the early stage of training, the cyclic learning rate is used to quickly explore the solution space, while in the middle and late stages of training, the adaptive learning rate strategy and patience value are gradually introduced to ensure that the model can stably converge to the global optimum.

[0065] The system with two learning rate adjustment strategies can take into account both rapid convergence in the early stage and stable optimization in the later stage, while avoiding overfitting and improving the overall performance of the model. The learning rate is dynamically adjusted according to the training progress. The learning rate can be changed in a larger range through cyclic learning rate, thereby helping the model to explore the global optimal solution and avoid falling into the local optimal solution. The adaptive learning rate dynamically adjusts the step size according to the gradient, which is especially suitable for high-dimensional space or gradient sparse problems. The model parameters can be finely adjusted locally. The combination of the two can optimize the model in a large range while ensuring fine adjustment of the local optimal solution. The cyclic learning rate can accelerate the initial training and avoid the problem of slow convergence caused by too small a learning rate. The adaptive learning rate decay automatically adjusts the learning rate in the later stage to make the training process smoother. By reducing the learning rate, overfitting can be prevented. The combination of the two can improve the model convergence speed and maintain the stability of training in the later stage.

[0066] In the intelligent risk assessment system for enterprise safety production, it is crucial for the model training module to be able to stably converge to the global optimum to improve the overall performance and accuracy of the system. Stable convergence to the global optimum means that the system can make full use of historical data and real-time safety production related data to conduct accurate risk assessment, ensuring that the model not only performs well in local areas, but also makes the best decision globally, which helps to identify potential production safety hazards, reduce missed reports or false reports, and enable the system to detect potential risks more effectively.

[0067] Stable convergence ensures the generalization ability of the model. The system can maintain high assessment accuracy and robustness when facing new data, ensuring that enterprises can still carry out effective risk prediction and prevention and control in a changing production environment. Through stable model training and optimization, the risk assessment module can accurately identify different levels of risk based on different production conditions, environmental changes, equipment status and other factors, thereby providing strong support for enterprise decision-making;

[0068] Stable convergence to the global optimum means that the system's responses in generating risk reports, issuing early warning notifications, etc. are more accurate and timely. Enterprises can rely on this system to formulate safe production strategies more efficiently, reduce the probability of accidents, and ultimately improve the overall safety level of the enterprise.

[0069] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A risk intelligent assessment system for enterprise safety production, including real-time data acquisition module, data collection module, data preprocessing module, data storage module, risk identification module, feature extraction module, model training module, model evaluation module and risk assessment module. Through the collaborative work of multiple modules, comprehensive monitoring and risk assessment of the enterprise production environment can be achieved, and the overfitting problem in model training can be solved by adjusting the learning rate; Features: The learning rate adjustment includes a learning rate initialization module, a decay strategy module, a current learning rate module, a training progress monitoring module and a parameter adjustment module. The learning rate initialization module is used to determine the initial learning rate. The learning rate initialization module serves as the starting point of the entire training process and determines the initial learning speed of the model. The decay strategy module dynamically adjusts the learning rate according to the training progress. The function of the current learning rate module is to track and record the currently used learning rate in real time to ensure that the optimizer can apply the correct learning rate for parameter update, update the current learning rate in each training cycle, construct a training progress monitoring module, collect training status information in real time, and provide input for the decay strategy module.

2. The intelligent risk assessment system for enterprise production safety according to claim 1 is characterized by: The attenuation strategy module dynamically adjusts the learning rate according to the training progress. The specific steps of adjusting the learning rate include: S1: Initialize the learning rate. Set the initial learning rate before starting training. S2: Select the decay strategy and determine the learning rate decay strategy to be used; S3: Update the learning rate during training. At the end of each epoch, update the current learning rate according to the set decay strategy. S4: Update the learning rate of the optimizer. After each update of the learning rate, ensure that the optimizer uses the new learning rate. S5: Monitor and record the learning rate. Record the changes in the learning rate in the training log to facilitate subsequent analysis and adjustment.

3. The intelligent risk assessment system for enterprise safety production according to claim 2 is characterized by: The attenuation strategy module combines the cyclic learning rate with the adaptive learning rate during the training process. The cyclic learning rate switches between high and low values ​​by adjusting the learning rate in different epochs, and the adaptive learning rate dynamically adjusts the learning rate according to the actual gradient changes.

4. The intelligent risk assessment system for enterprise safety production according to claim 1 is characterized by: The attenuation strategy module adds a patience value mechanism during the learning rate adjustment process, and automatically starts the learning rate attenuation strategy when the validation set performance does not improve significantly within multiple epochs.

5. The intelligent risk assessment system for enterprise safety production according to claim 1 is characterized by: The attenuation strategy module establishes a dynamic switching mechanism to switch the use strategy of cyclic learning rate and adaptive learning rate according to the training progress. In the early stage of training, the cyclic learning rate is used to quickly explore the solution space. In the middle and late stages of training, the adaptive learning rate strategy is gradually introduced to ensure stable convergence of the model.

6. The intelligent risk assessment system for enterprise safety production according to claim 1 is characterized by: In the attenuation strategy module, the method of superimposing the cyclic learning rate and the adaptive learning rate includes: S1: Initialize the adaptive learning rate optimizer; S2: Set the cyclic learning rate strategy; S3: Combined with the patience value mechanism; S4: Dynamic switching.

7. The intelligent risk assessment system for enterprise safety production according to claim 1 is characterized by: The attenuation strategy module adjusts the strategy through the learning rate, and the risk assessment module accurately identifies the risk levels at different levels.

8. The intelligent risk assessment system for enterprise safety production according to claim 1 is characterized by: The training progress monitoring module provides feedback for the learning rate adjustment, and the parameter adjustment module is responsible for applying the calculated new learning rate to the optimizer.

9. The intelligent risk assessment system for enterprise production safety according to claim 1 is characterized by: There is a close connection and interaction between the various modules in the learning rate adjustment, which together constitute a dynamic learning rate adjustment system.

Images