Occupational hazard risk assessment system based on artificial intelligence

Abstract

The invention relates to the field of machine learning, provides an artificial intelligence-based occupational hazard risk assessment system, and aims to solve the problems of serious missing of multi-dimensional risk data, insufficient feature expression and low assessment precision. The system is composed of a data acquisition module, a data enhancement module and a risk assessment module. Unified tensorization processing is carried out on the data through a segmented weighted non-negative matrix factorization model, an error function of L1 and L2 norm combination is introduced, and the robustness and accuracy of data completion are improved; on this basis, constructing a Bargmann-SVM model, mapping standardized data to a Bargmann space, obtaining complex high-dimensional features by extracting a high-order derivative and a projection coefficient of a full-pure function, training a support vector machine in combination with a polynomial radial basis kernel, and performing accurate classification and grade evaluation; the system can be widely applied to high-risk environments such as factories and workshops, and the intelligent recognition and early warning capability of occupational hazard risks is improved.

Description

Technical Field

[0001] The present invention relates to the field of machine learning, and in particular to an occupational hazard risk assessment system based on artificial intelligence. Background Art

[0002] With the continuous acceleration of industrialization and the increasing complexity of production environments, occupational hazards pose an increasingly serious threat to the health and safety of workshop and factory workers. Traditional occupational hazard risk assessment methods are unable to meet the intelligent and refined needs of modern enterprises in occupational health management. Although some enterprises have introduced intelligent systems for occupational hazard risk assessment, they still have many shortcomings. First, when faced with missing multidimensional risk data caused by sensor failures, network delays, etc., existing systems typically use static methods such as mean filling and linear interpolation. These methods lack structural perception and have difficulty restoring the inherent correlations of the data, affecting the accuracy of subsequent risk assessments. Second, most systems use shallow artificial features or low-dimensional mapping based on raw data, which cannot fully explore the high-order coupling characteristics between environmental, behavioral, and health data, and are particularly difficult to identify complex risk patterns under nonlinear and multidimensional interactions. Finally, existing intelligent risk assessment models have poor adaptability to risk differences between different positions, environments, or personnel, and are prone to overfitting, misjudgment, or category confusion. They are particularly robust when used with new samples or unseen working conditions. Summary of the Invention

[0003] The present invention provides an occupational hazard risk assessment system based on artificial intelligence, aiming to solve the problems of weak multidimensional data missing processing capability, insufficient feature expression, and insufficient classification accuracy in existing intelligent assessment methods. The core of the system includes two key technical innovations: first, a piecewise weighted non-negative matrix decomposition data completion model is used to process multidimensional risk data, and the workshop environment, employee health and behavior characteristics are constructed into a unified high-dimensional risk tensor. The piecewise error function combining L1 and L2 norms is introduced to respectively process small errors and large errors, thereby improving the system's robustness to outliers and overall prediction accuracy. At the same time, a non-negative multiplication update strategy is used to optimize the three types of latent factor matrices of environment, health and behavior. To achieve accurate reconstruction of missing risk data; secondly, an SVM risk assessment model based on Bargmann space mapping (Bargmann-SVM) is proposed, which maps the reconstructed standardized risk data to the holomorphic function form in the Bargmann space, extracts its high-order derivatives and projection coefficients as high-order features, and then combines the polynomial radial basis function to construct a support vector machine classifier, which effectively improves the system's ability to express and classify nonlinear risk structures; through the synergistic effect of the above two modules, the system can eventually output multidimensional occupational hazard assessment results including health risks, environmental risks and behavioral risks, and provide factories with targeted risk warnings and prevention and control measures.

[0004] The present invention provides an occupational hazard risk assessment system based on artificial intelligence, which includes a data acquisition module, a data enhancement module and a risk assessment module;

[0005] The data acquisition module collects workshop environment data through industrial sensors, employee health data through employee smart bracelets, and dangerous behavior data through cameras and OpenCV, integrating them to obtain multi-dimensional risk data;

[0006] The data enhancement module constructs a piecewise weighted non-negative matrix factorization data completion model, processes multidimensional risk data through the piecewise weighted non-negative matrix factorization data completion model, and generates reconstructed multidimensional risk data;

[0007] The risk assessment module constructs a Bargmann-SVM model, processes and reconstructs multidimensional risk data through the Bargmann-SVM model, and generates occupational hazard risk assessment results.

[0008] Furthermore, the piecewise weighted non-negative matrix factorization data completion model generates and reconstructs the multi-dimensional risk data, which specifically includes the following steps:

[0009] Step S1: Convert the multidimensional risk data into a unified format, construct a high-dimensional tensor, mark the missing values ​​in the multidimensional risk data and record the missing locations, and perform standardization to ensure the scale consistency of the data, and generate a standardized high-dimensional data tensor;

[0010] Step S2: Initialize the latent factor matrix and use it to calculate the missing risk value of the standardized high-dimensional data tensor to generate the prediction error; the latent factor matrix is ​​used to model the potential risk structure of the three dimensions of environment, health, and behavior;

[0011] Step S3: constructing a matrix decomposition error function by introducing a joint norm and a regularization term, and expanding the matrix decomposition error function by introducing a piecewise weighted norm. Combining the prediction error to process small errors and large errors separately, a piecewise objective function is generated.

[0012] Step S4: Calculate the gradient of the segmented objective function with respect to the latent factor matrix, generate gradient information, and update the latent factor matrix based on the gradient information. In the process of updating the latent factor matrix, adopt a non-negative multiplication update strategy that depends on a single latent factor. By introducing non-negative constraints and weighted update rules, optimize the latent factor matrix to ensure that all elements in the matrix remain non-negative, and obtain the optimized factor matrix.

[0013] Step S5: Reconstruct the standardized high-dimensional data tensor by optimizing the factor matrix to generate reconstructed multi-dimensional risk data.

[0014] Furthermore, the Bargmann-SVM model,generates the occupational hazard risk assessment results, including the following steps:

[0015] Step B1: Define the Bargmann space and normalize the reconstructed multidimensional risk data to obtain standardized risk data, ensuring that the data in each dimension has a unified dimension. For each data point in the standardized risk data, map it to a holomorphic function representation in the Bargmann space to obtain the Bargmann space data representation.

[0016] Step B2: Based on the Bargmann space data representation, the numerical differentiation method in the Bargmann space is used to calculate the high-order derivatives of the holomorphic function representation, and the high-order features of the holomorphic function representation are extracted based on the high-order derivatives to obtain the high-order features of the Bargmann space. The high-order features of the Bargmann space include projection coefficients and derivative information.

[0017] Step B3: Define the risk level, build the SVM model, combine the Bargmann space high-order features and the risk level into a training set, and train the SVM model;

[0018] Step B4: Construct a polynomial radial basis function and use it to tune the parameters of the SVM model to obtain a trained SVM model. The polynomial radial basis function includes a Gaussian radial basis function and a polynomial kernel. Combining the two improves the flexibility and accuracy of the SVM model in processing data.

[0019] Step B5: Use the trained SVM model to perform risk assessment and generate occupational hazard risk assessment results. The occupational hazard risk assessment results include health risk assessment results, environmental risk assessment results, and behavioral risk assessment results. Preventive measures are formulated based on the assessment results.

[0020] By adopting the above scheme, the beneficial effects achieved by the present invention are as follows:

[0021] This invention effectively improves the system's ability to restore missing risk information in factory workshops. Through a piecewise weighted non-negative matrix factorization data completion model, environmental parameters, employee health indicators, and behavioral monitoring data collected on-site are integrated into a unified tensor form. A piecewise weighted strategy is introduced into the error function to effectively distinguish the impact of small and large errors, thereby improving the model's adaptability and robustness in dealing with workshop environments with strong noise interference and frequent data missing. This strategy not only improves the accuracy of risk data completion, but also provides a stable and complete data foundation for subsequent high-precision assessment models.

[0022] Furthermore, the present invention constructs a Bargmann-SVM model to address the problems of overlapping risk factors and complex patterns in workshop scenarios. By embedding the standardized risk data into the Bargmann space and converting it into a holomorphic function form, and extracting the projection coefficients and high-order derivatives in its Taylor expansion, it is possible to capture small changes in the data and characterize the global structure. Subsequently, the SVM model is trained using a polynomial radial basis function that fuses a Gaussian RBF kernel with a polynomial kernel, which greatly enhances the system's modeling capabilities for nonlinear risk distributions, achieves accurate classification and grading of employee health risks, environmental risks, and behavioral risks, and improves the intelligence and precision of risk assessment.

[0023] In summary, the present invention not only breaks through the technical bottlenecks of existing factory workshop intelligent risk assessment systems in terms of missing value completion, nonlinear modeling and multidimensional feature fusion, but also effectively enhances the structural cognition and generalized prediction capabilities of occupational hazard risks under complex working conditions by constructing a high-order feature recognition and classification mechanism; the system can provide systematic, explainable and refined risk assessment services for high-risk workshops, and provide powerful decision-making support for enterprise safety production management. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 A schematic diagram of a module of an occupational hazard risk assessment system based on artificial intelligence provided by the present invention;

[0025] Figure 2 This is the heart rate and PM2.5 trend graph of employee A proposed in Example 6;

[0026] Figure 3 This is the TVOC heat map of the workshop and the activity trajectory of employee A in Example 6.

[0027] Employee A's heart rate and PM2.5 trend chart: the horizontal axis is "time", the left vertical axis is "heart rate (beats / minute)", and the right vertical axis is "PM2.5 (micrograms / cubic meter)";

[0028] TVOC heat map of the workshop and the activity trajectory of employee A: the red area indicates a higher TVOC concentration, and the blue dotted line is the movement trajectory of employee A. DETAILED DESCRIPTION

[0029] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0030] Example 1, according to Figure 1,The present invention provides an occupational hazard risk assessment system based on artificial intelligence, which includes a data acquisition module, a data enhancement module and a risk assessment module;

[0031] The data acquisition module collects workshop environment data through industrial sensors, employee health data through employee smart bracelets, and dangerous behavior data through cameras and OpenCV, integrating them to obtain multi-dimensional risk data;

[0032] The data enhancement module constructs a piecewise weighted non-negative matrix factorization data completion model, processes multidimensional risk data through the piecewise weighted non-negative matrix factorization data completion model, and generates reconstructed multidimensional risk data;

[0033] The risk assessment module constructs a Bargmann-SVM model, processes and reconstructs multidimensional risk data through the Bargmann-SVM model, and generates occupational hazard risk assessment results.

[0034] Example 2: This example is based on Example 1. In this example, a piecewise weighted non-negative matrix factorization data completion model is used to generate and reconstruct multidimensional risk data. The process specifically includes the following steps:

[0035] Step S1: Convert the multidimensional risk data into a unified format, construct a high-dimensional tensor, mark the missing values ​​in the multidimensional risk data and record the missing locations, and perform standardization to ensure the scale consistency of the data, and generate a standardized high-dimensional data tensor;

[0036] Step S2: Initialize the latent factor matrix and use it to calculate the missing risk value of the standardized high-dimensional data tensor to generate the prediction error. The latent factor matrix is ​​used to model the potential risk structure of the three dimensions of environment, health, and behavior. The formula used is as follows:

[0037] Risk value prediction formula:

[0038] ;

[0039] in, Represents the dimension index, represents the predicted missing risk value, represents the latent factor index, represents the number of latent factors, Represents the environmental risk factor matrix The element in The sensor and The projection component on the latent factor dimension, Represents the health risk factor matrix The element in time periods and The projection component on the latent factor dimension, Represents the behavioral risk factor matrix The elements in represent the The behavioral characteristics of each employee and The projection components on the latent factor dimensions;

[0040] The missing data values ​​in the standardized high-dimensional data tensor are predicted by the latent factor matrix, which represents the potential risk factors of environment, health and behavior. They are associated with each data point in the standardized high-dimensional data tensor, and the missing risk values ​​can be estimated by this formula;

[0041] Step S3: The matrix decomposition error function is constructed by introducing the joint norm and regularization term, and the segmented weighted norm is introduced to expand the matrix decomposition error function. The small error and large error are processed separately in combination with the prediction error to generate a segmented objective function. The formula used is as follows:

[0042] The matrix decomposition error function is:

[0043] ;

[0044] in, represents the objective function, represents the actual observed value, represents a collection of normalized high-dimensional data tensors, represents the prediction error, Indicates that the square of the prediction error is measured using the SL1 norm. Indicates that the square of the prediction error is measured using the L2 norm. represents the regularization parameter, represents the sum of squares of each element in the latent factor matrix; represents the joint norm;

[0045] The SL1 norm part improves the robustness of the model and suppresses noise through sparse regularization;

[0046] The L2 norm part optimizes the smoothness of the model and reduces the risk of overfitting;

[0047] Piecewise objective function:

[0048] ;

[0049] When the error is small, the L2 norm (i.e., square error) is used to fit the error more carefully;

[0050] When the error is large, the L1 norm (i.e., absolute value) is introduced to increase the robustness to outliers;

[0051] Additional regularization term to prevent overfitting;

[0052] Step S4: Calculate the gradient of the segmented objective function with respect to the latent factor matrix, generate gradient information, and update the latent factor matrix based on the gradient information. In the process of updating the latent factor matrix, adopt a non-negative multiplication update strategy that depends on a single latent factor. By introducing non-negative constraints and weighted update rules, optimize the latent factor matrix to ensure that all elements in the matrix remain non-negative, and obtain the optimized factor matrix.

[0053] Step S5: Reconstruct the standardized high-dimensional data tensor by optimizing the factor matrix to generate reconstructed multi-dimensional risk data.

[0054] Example 3: This example is based on Example 1 and provides another preferred solution. In this example, the process of generating and reconstructing multi-dimensional risk data specifically includes the following steps:

[0055] Step E1: Convert the multidimensional risk data into a unified format, construct a high-dimensional tensor, mark the missing values ​​in the multidimensional risk data and record the missing locations, and perform standardization to ensure the scale consistency of the data, and generate a standardized high-dimensional data tensor;

[0056] Step E2: Initialize the latent factor matrix and use it to calculate the missing risk value of the standardized high-dimensional data tensor to generate the prediction error; the latent factor matrix represents environmental risk factors, health risk factors, and behavioral risk factors respectively;

[0057] Step E3: Constructing the matrix decomposition error function by introducing the piecewise weighted norm and regularization term;

[0058] Step E4: Calculate the gradient of the matrix decomposition error function with respect to the latent factor matrix, generate gradient information, and update the latent factor matrix based on the gradient information to obtain the optimized factor matrix;

[0059] Step E5: Reconstruct the standardized high-dimensional data tensor by optimizing the factor matrix to generate reconstructed multidimensional risk data.

[0060] Example 4: This example is based on Example 2. In this example, the Bargmann-SVM model generates an occupational hazard risk assessment result, specifically comprising the following steps:

[0061] Step B1: Define the Bargmann space and standardize the reconstructed multidimensional risk data to obtain standardized risk data, ensuring that the data in each dimension has a unified dimension. For each data point of the standardized risk data, map it to a holomorphic function representation in the Bargmann space to obtain the Bargmann space data representation. The Bargmann space data representation is the holomorphic function representation of each standardized risk data point in the Bargmann space, that is, the data point is represented in the form of a Taylor series expansion of the holomorphic function. It contains projection coefficients, which characterize the variation characteristics of the data point at different orders in the Bargmann space, helping to reveal the complex patterns and high-order characteristics of the data. The Bargmann space is a Hilbert space containing holomorphic functions, which has the characteristics of Gaussian weighting and square integrability. The formula used is as follows:

[0062] Taylor series expansion of a holomorphic function:

[0063] ;

[0064] in, represents the power term, represents the projection coefficient, Indicates the numerical representation of the standardized risk data, express of Power, represents a holomorphic function, that is, a holomorphic function representation of each data point in the Bargmann space;

[0065] Step B2: Based on the Bargmann space data representation, the numerical differentiation method in the Bargmann space is used to calculate the high-order derivatives of the holomorphic function representation, and the high-order features of the holomorphic function representation are extracted based on the high-order derivatives to obtain the high-order features of the Bargmann space. The high-order features of the Bargmann space include projection coefficients and derivative information.

[0066] Step B3: Define the risk level, build the SVM model, combine the Bargmann space high-order features and the risk level into a training set, and train the SVM model;

[0067] Step B4: Construct a polynomial radial basis function and use it to tune the parameters of the SVM model to obtain a trained SVM model. The polynomial radial basis function includes a Gaussian radial basis function and a polynomial kernel. Combining the two improves the flexibility and accuracy of the SVM model in processing data. The formula used is as follows:

[0068] ;

[0069] in, represents a data point in the Bargmann space data representation, represents the polynomial radial basis function, Represents the weighting factor, which is used to balance the influence of Gaussian RBF components and polynomial components in the kernel function; represents the bandwidth parameter, which controls the expansion range of the Gaussian kernel. represents the square of the Euclidean distance, represents the Gaussian RBF kernel; express and The dot product of represents the offset constant, represents the polynomial power, represents the polynomial kernel;

[0070] Step B5: Use the trained SVM model to perform risk assessment and generate occupational hazard risk assessment results. The occupational hazard risk assessment results include health risk assessment results, environmental risk assessment results, and behavioral risk assessment results. Preventive measures are formulated based on the assessment results.

[0071] Example 5: This example is based on Example 2 and provides another preferred method. In this example, the process of generating occupational hazard risk assessment results specifically includes the following steps:

[0072] Step R1: Define risk levels, build an SVM model, combine the reconstructed multidimensional risk data and risk levels into a training set, and train the SVM model;

[0073] Step R2: Construct a polynomial radial basis function, and use the polynomial radial basis function to tune the parameters of the SVM model to obtain the trained SVM model;

[0074] Step R3: Use the trained SVM model to perform risk assessment and generate occupational hazard risk assessment results.

[0075] Example 6, according to Figure 2 、 Figure 3 ,This embodiment is based on the fourth embodiment.,In this embodiment, the risk assessment module,constructs a Bargmann-SVM model, processes and reconstructs multidimensional risk data through the Bargmann-SVM model,,and generates an occupational hazard risk assessment result;

[0076] In this embodiment:

[0077] Reconstructing multidimensional risk data:

[0078] ;

[0079] Generates a heart rate and PM2.5 trend chart for employee A, showing the changing trends of heart rate (bpm) and PM2.5 concentration (µg / m³) from 08:00 to 17:00, helping to identify peak stress periods.

[0080] Generate a TVOC heat map of the workshop and the activity trajectory of employee A. The TVOC concentration heat distribution in the workshop plan is overlaid with employee A's movement trajectory, with high-risk areas highlighted in red.

[0081] After training, the SVM model classifies the input of employee A:

[0082] Health risk level: moderately high (Level 3);

[0083] Environmental risk level: High risk (Level 4);

[0084] Behavioral risk level: Medium (Level 2);

[0085] Occupational hazard risk assessment results:

[0086] Health risk: Employee A has a high heart rate and stress score, indicating a stressful work state;

[0087] Environmental risks: TVOC and temperature are both high, PM2.5 exceeds the standard, and there are risks of occupational poisoning and heat stress;

[0088] Behavioral risks: Bending over frequently and approaching dangerous equipment; operating habits need to be improved;

[0089] Preventive measures recommended:

[0090] Health risk prevention and control:

[0091] Increase the frequency of monthly psychological stress screening for employee A;

[0092] Equipped with a smart bracelet to monitor heart rate and SpO2 in real time, automatically alerting when abnormalities occur;

[0093] Arrange a weekly meditation or counseling session for employees;

[0094] Environmental risk control:

[0095] Optimize the workshop exhaust system and add local exhaust fans at employee stations;

[0096] Add TVOC alarm, which will automatically activate exhaust ventilation when the set threshold is exceeded;

[0097] Behavioral risk correction:

[0098] Install a danger warning line near the reactor, and automatically record any illegal approach;

[0099] Add operating area signs and strengthen daily safety training.

[0100] The present invention and its embodiments are described above. Such description is not restrictive. What is shown in the accompanying drawings is only one of the embodiments of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in this field are inspired by it and do not depart from the purpose of the invention, they can creatively design structural methods and embodiments similar to the technical solution, which should all fall within the scope of protection of the present invention.

Claims

1. An artificial intelligence-based occupational hazard risk assessment system, comprising a data acquisition module, wherein the data acquisition module collects multi-dimensional risk data; characterized in that: The system also includes a data enhancement module and a risk assessment module; The data enhancement module constructs a piecewise weighted non-negative matrix decomposition data completion model, processes the multidimensional risk data through the piecewise weighted non-negative matrix decomposition data completion model, and generates reconstructed multidimensional risk data; The risk assessment module constructs a Bargmann-SVM model, processes and reconstructs multidimensional risk data through the Bargmann-SVM model, and generates occupational hazard risk assessment results.

2. The artificial intelligence-based occupational hazard risk assessment system according to claim 1, characterized in that: The process of generating and reconstructing multidimensional risk data using the piecewise weighted non-negative matrix factorization data completion model specifically includes the following steps: Step S1: Mark missing values ​​in multidimensional risk data and record missing locations to generate a standardized high-dimensional data tensor; Step S2: Initialize the latent factor matrix, use the latent factor matrix to calculate the missing risk value of the standardized high-dimensional data tensor, and generate the prediction error; Step S3: constructing a matrix decomposition error function by introducing a joint norm and a regularization term, and expanding the matrix decomposition error function by introducing a piecewise weighted norm. Combining the prediction error to process small errors and large errors separately, a piecewise objective function is generated. Step S4: Calculate the gradient of the segmented objective function with respect to the latent factor matrix, generate gradient information, and update the latent factor matrix based on the gradient information to obtain the optimized factor matrix; Step S5: Reconstruct the standardized high-dimensional data tensor by optimizing the factor matrix to generate reconstructed multi-dimensional risk data.

3. The artificial intelligence-based occupational hazard risk assessment system according to claim 2, characterized in that: In step S4, during the latent factor matrix update process, a single latent factor-dependent non-negative multiplication update strategy is adopted to optimize the latent factor matrix by introducing non-negative constraints and weighted update rules.

4. The artificial intelligence-based occupational hazard risk assessment system according to claim 2, characterized in that: The latent factor matrix was used to model the potential risk structure in the three dimensions of environment, health, and behavior.

5. The artificial intelligence-based occupational hazard risk assessment system according to claim 1, characterized in that: The Bargmann-SVM model generates occupational hazard risk assessment results, which specifically includes the following steps: Step B1: Define the Bargmann space and process the reconstructed multidimensional risk data to obtain standardized risk data. For each data point of the standardized risk data, map it to a holomorphic function representation in the Bargmann space to obtain the Bargmann space data representation. Step B2: Based on the Bargmann space data representation, the numerical differentiation method in the Bargmann space is used to calculate the high-order derivatives of the holomorphic function representation, and the high-order features of the holomorphic function representation are extracted based on the high-order derivatives to obtain the high-order features of the Bargmann space; Step B3: Define the risk level, build the SVM model, combine the Bargmann space high-order features and the risk level into a training set, and train the SVM model; Step B4: construct a polynomial radial basis function, and optimize the parameters of the SVM model using the polynomial radial basis function to obtain a trained SVM model; Step B5: Use the trained SVM model to perform risk assessment and generate occupational hazard risk assessment results.

6. The artificial intelligence-based occupational hazard risk assessment system according to claim 5, characterized in that: The high-order features of Bargmann space include projection coefficients and derivative information.

7. The artificial intelligence-based occupational hazard risk assessment system according to claim 5, characterized in that: Polynomial radial basis functions include Gaussian radial basis functions and polynomial kernels.

Images