A risk assessment method and system for extra-long tunnel construction site
Through data acquisition and analysis, combined with clustering algorithms and BP neural network model, the risk assessment at the construction site of special-length tunnels is automatically solved, and the subjectivity and inefficiency of traditional methods are achieved, and efficient and accurate risk management and customized control are achieved. Measures to ensure construction safety.
Patent Information
- Application Number
- CN202410439279.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-04-12
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-04-12
AI Technical Summary
The traditional special-time tunnel construction site risk assessment method relies on expert experience and historical data, and is subjective and inefficient, making it difficult to achieve real-time monitoring and customized risk control, and cannot meet the needs of efficient, accurate and real-time risk management.
Data acquisition and analysis methods are used to build a risk factor indicator system, and clustering algorithms and BP neural network models are used to automatically identify risk factors and conduct quantitative evaluations, forming a risk matrix and control suggestions, and achieving automated and intelligent risk assessment.
It improves the efficiency of risk management, provides more comprehensive and accurate risk identification, can formulate targeted risk prevention measures, reduce risk occurrence, ensure construction safety, and realize dynamic monitoring and evaluation of risk conditions on the construction site.
Smart Images

Figure CN118246744B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of tunnel assessment, and in particular to a method and system for risk assessment at a construction site of an extra-long tunnel. Background Art
[0002] With the rapid development of infrastructure construction, extra-long tunnels have become important projects connecting key areas such as mountainous areas and crossing water bodies. Extra-long tunnel construction has the characteristics of long construction period, high technical requirements and many risk factors. In order to ensure construction safety and project quality, accurate risk assessment and management have become particularly important. Traditional risk assessment methods mainly rely on expert experience and historical data analysis, and use qualitative or semi-quantitative methods to identify and assess risks.
[0003] Traditional risk assessment methods rely heavily on the experience and judgment of experts, which may lead to a certain degree of subjectivity in the assessment results, making it difficult to achieve a comprehensive and objective risk assessment. Due to reliance on manual data collection and analysis, traditional methods are inefficient in processing large amounts of real-time data, making it difficult to achieve real-time monitoring and assessment of construction site risks. As construction environments and conditions change, traditional risk assessment methods are difficult to quickly adapt to new risk factors, and it is also difficult to update assessment models and strategies. When proposing risk control measures, traditional methods often rely on general solutions and lack customized control strategies for specific risk factors.
[0004] Due to these defects, the application of traditional risk assessment methods in large and complex engineering projects such as extra-long tunnels is limited, and it is difficult to meet the needs of efficient, accurate and real-time risk management. Therefore, it is necessary to provide a new assessment method for extra-long tunnel construction sites. Summary of the invention
[0005] In view of this, the present invention provides a risk assessment method and system for a super-long tunnel construction site to overcome the above technical defects and improve the risk assessment and management level of the super-long tunnel construction site.
[0006] The technical purpose of the present invention is achieved in this way:
[0007] In one aspect, the present invention provides a method for risk assessment of a super-long tunnel construction site, comprising the following steps:
[0008] S1 collects the original data of the extra-long tunnel construction site and obtains the historical accident records related to the extra-long tunnel construction site;
[0009] S2 processes and analyzes historical accident records to identify a set of risk factors;
[0010] S3 establishes a risk assessment model for the extra-long tunnel construction site and performs pre-training in combination with a risk factor set, inputs the original data into the trained risk assessment model, and obtains a first probability matrix and a second probability matrix, wherein the first probability matrix describes the probability of occurrence under each risk factor, and the second probability matrix describes the probability of risk level distribution under each risk factor;
[0011] S4 combines the first probability matrix and the second probability matrix to form a risk matrix, and uses an evaluation method to quantitatively evaluate the risk matrix to obtain a risk evaluation result;
[0012] S5 formulates risk control recommendations based on risk assessment results.
[0013] On the basis of the above technical solution, preferably, step S2 includes the following steps:
[0014] S21 constructs a risk factor index system for the construction site of a super-long tunnel, including primary risk factor indicators and secondary risk factor indicators, and uses the analytic hierarchy process to assign initial weights to the primary risk factor indicators and secondary risk factor indicators;
[0015] S22 determines the k value according to the number of the first-level risk factor indicators, and determines the m value according to the number of the second-level risk factor indicators;
[0016] S23 extracts keywords from historical accident records to obtain accident keywords, wherein the historical accident records include accident frequencies;
[0017] S24 clusters the accident keywords using a clustering algorithm to form k accident sets and k first cluster centers, and performs secondary clustering on the k accident sets respectively, and obtains m second cluster centers and m accident subsets for each accident set;
[0018] S25 selects a first cluster center and matches it with the first-level risk factor indicator. If the match is successful, directly proceed to step S26; otherwise, the first cluster center is added as a new first-level risk factor indicator and then proceeds to step S26;
[0019] S26 selects a second cluster center to which the first cluster center belongs to match the secondary risk factor indicator. If the match is successful, the accident frequency corresponding to the second cluster center is quantified as a weighting factor, and the initial weights of the primary risk factor indicator and the secondary risk factor indicator are updated according to the weighting factor; if the match fails, the second cluster center is added as a new secondary risk factor indicator, and the initial weight of its secondary risk factor indicator is formed according to the accident frequency corresponding to the second cluster center, and the initial weight of its primary risk factor indicator is formed according to the accident frequency of the first cluster center to which it belongs;
[0020] S27 repeats steps S25-S26 until all first cluster centers and second cluster centers are matched, and the final first-level risk factor indicators, second-level risk factor indicators, and corresponding indicator weights are obtained to form a risk factor set.
[0021] Based on the above technical solution, preferably, the clustering algorithm includes a clustering loss function, which is expressed as follows:
[0022]
[0023] Where, L k is the clustering loss function, n1 is the number of data, n2 is the number of clusters, n2 is k or m, s i is the i-th data, z j is the cluster center of the jth cluster, λ is the regularization parameter, and ‖·‖ represents the distance between data.
[0024] On the basis of the above technical solution, preferably, step S3 includes the following steps:
[0025] S31 constructs a risk assessment model, which is a BP neural network model. The BP neural network model is divided into an input layer, a hidden layer and an output layer, and contains multiple neurons;
[0026] S32 prepares a sample set, uses the sample set in combination with the risk factor set to train the risk assessment model, and uses an optimization algorithm to optimize the number of neurons during the training process to obtain a trained risk assessment model, the number of neurons of which is N;
[0027] S33 inputs the original data into the trained risk assessment model for prediction, and the model outputs the probability of occurrence and the probability of risk level distribution under each risk factor;
[0028] S34 generates a first probability matrix and a second probability matrix according to the results of model prediction. The first probability matrix describes the occurrence probability of each risk factor, and the second probability matrix describes the risk level distribution probability of each risk factor.
[0029] Based on the above technical solution, preferably, step S32 includes the following steps:
[0030] S321 collects historical data of the construction site of the extra-long tunnel to prepare a sample set, and pre-processes the sample set, the sample set including a reference value;
[0031] S322 determines the number of neurons in each layer and the network weight in the risk assessment model, wherein the number of neurons in the input layer is the same as the number of risk factors, and the number of neurons in the output layer is the dimension of the risk matrix;
[0032] S323 encodes the number of neurons in the hidden layer of the risk assessment model and uses it as an optimization target of the optimization algorithm;
[0033] S324 randomly generates a set of different numbers of hidden layer neurons as the initial population of the optimization algorithm;
[0034] S325 defines the network loss function of the risk assessment model as a fitness function;
[0035] S326 selects individuals to determine the network structure of the risk assessment model, inputs the sample set into the risk assessment model, passes through the input layer, the hidden layer and finally the output layer in sequence, the neurons of each layer calculate the output according to the input and the network weight, and pass it to the next layer, in the output layer, calculates the value of the network loss function according to the actual output value and the reference value, calculates the gradient of the network loss function to each network weight according to the value of the network loss function, and passes the gradient back to the network, and updates the network weight according to the update formula;
[0036] S327 evaluates the performance of the individual according to the value of the network loss function, and iteratively updates the population according to the evaluation result;
[0037] S328 repeats steps S326-S327 until the preset number of iterations is reached or the network loss function converges, and a trained risk assessment model is obtained, the number of neurons of which is N.
[0038] Based on the above technical solution, preferably, the network loss function is:
[0039]
[0040] Where L is the network loss function, the first term is the loss function of the occurrence probability under each risk factor, the second term is the loss function of the risk level distribution probability under each risk factor, α and β are the weight parameters of the first and second terms respectively, M is the number of samples, is the true label of the i2th sample, which is 0 or 1. is the probability that the model predicts that the i2th sample category is 1, C is the total number of risk levels, is a binary indicator whose value is 0 or 1, indicating whether sample i2 belongs to category c. is the probability that the model predicts whether sample i2 belongs to category c.
[0041] Based on the above technical solution, preferably, in step S326, the update formula is:
[0042]
[0043] In the formula, t is the current iteration number, t+1 represents the next iteration, Represents the new network weight value for the next iteration, used to connect neuron h2 and neuron h1. Represents the network weight value in the current iteration, and also connects neuron h2 and neuron h1. represents the error gradient of the h2th neuron for the i2th sample, It represents the output value of the i2th sample passing through the h2th neuron, and ε is the learning speed.
[0044] On the basis of the above technical solution, preferably, step S4 includes the following steps:
[0045] The S41 risk factor set includes p first-level risk factor indicators and q second-level risk factor indicators. The size of the first probability matrix P is p*k, and the size of the second probability matrix L is p*k. The risk matrix R is calculated based on the first probability matrix and the second probability matrix:
[0046]
[0047] in, represents element-wise multiplication, that is:
[0048]
[0049] Among them, g1 is the g1th first-level risk factor indicator, and g2 is the g2th second-level risk factor indicator;
[0050] S42 calculates the first comprehensive risk value of each first-level risk factor indicator:
[0051]
[0052] In the formula, It represents the first comprehensive risk value of the g1th first-level risk factor indicator;
[0053] S43 sorts the first-level risk factor indicators from high to low according to the first comprehensive risk value to obtain a first sequence, and determines the risk priority of the first sequence, and the first first-level risk factor indicator has the highest risk priority;
[0054] S44 calculates the second comprehensive risk value of each secondary risk factor indicator:
[0055]
[0056] In the formula, It represents the second comprehensive risk value of the g2-th secondary risk factor indicator;
[0057] S45 sorts the secondary risk factor indicators from high to low according to the second comprehensive risk value to obtain a second sequence, and determines the risk priority of the second sequence, with the first secondary risk factor indicator being the highest risk priority;
[0058] S46 According to The cross term of the first-level risk factor indicator and the second-level risk factor indicator is found in the first sequence and the second sequence, which is taken as a high-risk risk factor combination. The risk factor combination is analyzed to form a risk assessment result.
[0059] Based on the above technical solution, preferably, step S46 includes the following steps:
[0060] S461 sets the risk threshold and divides the risk into three levels: low risk, medium risk and high risk according to the risk threshold;
[0061] S462 Calculate the risk value of each risk factor combination Comparing the risk value with the risk threshold, and classifying the risk factor combination as low risk, medium risk or high risk according to the comparison result;
[0062] S463 screens out all risk factor combinations classified as high risk from the classification results;
[0063] S464 sorts the risk factor combinations with high risks from high to low according to the risk values to obtain a third sequence, and marks the third sequence with a final risk priority;
[0064] S465 analyzes the high-risk risk factors in the third sequence in turn, determines the risk causes, risk consequences, control measures and their effectiveness, and forms a risk assessment report as the risk assessment result.
[0065] On the other hand, the present invention further provides a risk assessment system for a super-long tunnel construction site, the system being used to execute any of the above methods, the system comprising:
[0066] A data acquisition module configured to collect raw data and historical data from a construction site of a super-long tunnel;
[0067] A data processing and analysis module is configured to process and analyze the collected raw data to identify a set of risk factors;
[0068] A risk assessment model training module is configured to construct a risk assessment model and train the risk assessment model in combination with an optimization algorithm to obtain a trained risk assessment model;
[0069] A risk assessment module is configured to evaluate the original data using the trained risk assessment model, predict the risk probability and the risk level distribution probability, output a first probability matrix and a second probability matrix, combine the first probability matrix and the second probability matrix to form a risk matrix, and use an assessment method to quantitatively evaluate the risk matrix to obtain a risk assessment result;
[0070] The control suggestion generation module is configured to select risk control measures based on the risk assessment results to form risk control suggestions.
[0071] The method of the present invention has the following beneficial effects compared with the prior art:
[0072] (1) The present invention greatly improves the efficiency of risk management through an automated and intelligent risk assessment method, providing support for rapid decision-making at the construction site;
[0073] (2) The present invention uses data analysis and clustering algorithms to automatically identify risk factors, avoiding risk points that may be overlooked in traditional methods and making risk identification more comprehensive;
[0074] (3) The present invention establishes a risk assessment model based on BP neural network and integrates the optimization algorithm into the training process to effectively improve the performance of the BP neural network model, find the optimal number of hidden layer neurons, reduce the risk of overfitting, and enhance the generalization ability of the model, thereby providing more reliable and accurate results for risk assessment;
[0075] (4) Based on the accurate risk assessment results, the present invention can formulate more targeted risk prevention measures and emergency plans, effectively reduce or avoid the occurrence of risks, and ensure construction safety;
[0076] (5) The present invention can realize dynamic monitoring and evaluation of the risk status of the construction site through the continuously iteratively updated risk assessment model, thereby making the risk management process more systematic and scientific;
[0077] (6) The extra-long tunnel construction site risk assessment system provided by the present invention can provide a comprehensive, accurate and practical risk management tool for the construction team, helping them to effectively identify and control various risks in the construction process, ensuring construction safety and smooth progress of the project. BRIEF DESCRIPTION OF THE DRAWINGS
[0078] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0079] Figure 1 is a flow chart of a method according to an embodiment of the present invention;
[0080] Figure 2 A flow chart of a clustering algorithm according to an embodiment of the present invention;
[0081] Figure 3 A schematic diagram of forming a risk factor set according to an embodiment of the present invention;
[0082] Figure 4 A diagram showing the training process of the risk assessment model according to an embodiment of the present invention;
[0083] Figure 5 4 is a system structure diagram of an embodiment of the present invention. DETAILED DESCRIPTION
[0084] The following will be combined with 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.
[0085] like Figure 1 As shown, the present invention provides a method for risk assessment of a super-long tunnel construction site, comprising the following steps:
[0086] S1 collects the original data of the extra-long tunnel construction site and obtains the historical accident records related to the extra-long tunnel construction site;
[0087] S2 processes and analyzes historical accident records to identify a set of risk factors;
[0088] S3 establishes a risk assessment model for the extra-long tunnel construction site and performs pre-training in combination with a risk factor set, inputs the original data into the trained risk assessment model, and obtains a first probability matrix and a second probability matrix, wherein the first probability matrix describes the probability of occurrence under each risk factor, and the second probability matrix describes the probability of risk level distribution under each risk factor;
[0089] S4 combines the first probability matrix and the second probability matrix to form a risk matrix, and uses an evaluation method to quantitatively evaluate the risk matrix to obtain a risk evaluation result;
[0090] S5 formulates risk control recommendations based on risk assessment results.
[0091] Specifically, in one embodiment of the present invention, step S1 includes the following steps:
[0092] Collecting raw data from the construction site of the extra-long tunnel: Collecting various data from the construction site, including but not limited to the number of construction personnel, construction equipment, construction progress, material usage, environmental conditions, etc. Real-time monitoring of construction site data, such as temperature, humidity, air pressure, vibration and other parameters, through sensors and monitoring equipment.
[0093] Obtain historical accident records related to the extra-long tunnel construction site: Collect accident records that occurred in the past at the extra-long tunnel construction site, including information such as accident type, accident cause, accident consequences, and accident handling measures. Sort and classify historical accident records and establish an accident database, including detailed information such as the time, location, and casualties of the accident.
[0094] Specifically, see Figure 2 and Figure 3 In one embodiment of the present invention, step S2 includes the following steps:
[0095] S21 constructs a risk factor index system for the construction site of a super-long tunnel, including primary risk factor indicators and secondary risk factor indicators, and uses the analytic hierarchy process to assign initial weights to the primary risk factor indicators and secondary risk factor indicators;
[0096] S22 determines the k value according to the number of the first-level risk factor indicators, and determines the m value according to the number of the second-level risk factor indicators;
[0097] S23 extracts keywords from historical accident records to obtain accident keywords, wherein the historical accident records include accident frequencies;
[0098] S24 clusters the accident keywords using a clustering algorithm to form k accident sets and k first cluster centers, and performs secondary clustering on the k accident sets respectively, and obtains m second cluster centers and m accident subsets for each accident set;
[0099] S25 selects a first cluster center and matches it with the first-level risk factor indicator. If the match is successful, directly proceed to step S26; otherwise, the first cluster center is added as a new first-level risk factor indicator and then proceeds to step S26;
[0100] S26 selects a second cluster center to which the first cluster center belongs to match the secondary risk factor indicator. If the match is successful, the accident frequency corresponding to the second cluster center is quantified as a weighting factor, and the initial weights of the primary risk factor indicator and the secondary risk factor indicator are updated according to the weighting factor; if the match fails, the second cluster center is added as a new secondary risk factor indicator, and the initial weight of its secondary risk factor indicator is formed according to the accident frequency corresponding to the second cluster center, and the initial weight of its primary risk factor indicator is formed according to the accident frequency of the first cluster center to which it belongs;
[0101] S27 repeats steps S25-S26 until all first cluster centers and second cluster centers are matched, and the final first-level risk factor indicators, second-level risk factor indicators, and corresponding indicator weights are obtained to form a risk factor set.
[0102] In this embodiment, the k of the clustering algorithm is 4, and m is 6. When constructing the risk factor index system for the extra-long tunnel construction site, there are 4 first-level risk factor indicators and 6 second-level risk factor indicators, that is, one first-level risk factor indicator corresponds to 6 subordinate second-level risk factor indicators.
[0103] In a specific example, the risk factor indicator system initially constructed is as follows:
[0104] Level 1 risk factor indicators: personnel safety, equipment safety, environmental safety, and construction process safety. The level 2 risk factor indicators corresponding to each level 1 risk factor indicator are:
[0105] Personnel safety: number of workers, worker training, workers' safety awareness, workers' health status, workers' work standardization, and workers' personal protective equipment.
[0106] Equipment safety: equipment integrity, equipment operation standardization, equipment maintenance status, completeness of equipment safety protection devices, technical level of equipment operators, and equipment emergency response capabilities.
[0107] Environmental safety: cleanliness of the construction site environment, ventilation conditions at the construction site, gas detection conditions at the construction site, noise control conditions at the construction site, emissions of harmful substances at the construction site, and emergency response capabilities at the construction site.
[0108] Construction process safety: rationality of construction process, standardization of construction process, risk assessment of construction process, coordination between construction process steps, inspection and acceptance procedures of construction process, and formulation of emergency plans for construction process.
[0109] The analytic hierarchy process is a multi-criteria decision-making method used to solve complex decision-making problems. It decomposes the decision-making problem hierarchically, determines the importance weights of factors at each level, and finally obtains the optimal decision solution. The following is the specific process of the analytic hierarchy process:
[0110] Building a hierarchy:
[0111] First, determine the decision-making goal, decompose the goal into several first-level criteria (risk factor indicators), and then subdivide each first-level criterion into second-level criteria (corresponding to the six subordinate second-level risk factor indicators) to form a hierarchical structure.
[0112] Construct a judgment matrix:
[0113] For pairwise comparisons between primary criteria, expert judgment or questionnaire surveys are used to construct a consistency judgment matrix to determine the relative importance of each pair of criteria.
[0114] Calculate the weight vector:
[0115] The consistency judgment matrix is calculated using the eigenvector method or the consistency index method to obtain the weight vector of each criterion, reflecting its importance in the overall goal.
[0116] Consistency check:
[0117] The calculated weight vector is checked for consistency to determine the consistency of the judgment matrix and to ensure the reliability of expert judgment or questionnaire survey.
[0118] Comprehensive weight:
[0119] The weight vectors of the criteria at all levels are combined to obtain the final weight vector, which reflects the relative importance of each criterion in the overall goal.
[0120] Consistency check and adjustment:
[0121] If the consistency ratio exceeds a certain threshold, the judgment matrix needs to be adjusted until the consistency requirement is met.
[0122] The analytic hierarchy process is used to assign initial weights to the risk factor indicators of the extra-long tunnel construction site to accurately reflect the importance of each indicator in risk management.
[0123] In this embodiment, keyword extraction from historical accident records is to extract key information from a large amount of historical accident data to help identify and understand the risk factors existing at the construction site of a super-long tunnel. The process includes:
[0124] Collect historical accident records of long tunnel construction sites. These records include accident reports, accident investigation reports, safety inspection records, safety meeting records and other forms of documents.
[0125] Clean and organize the collected historical accident records to remove duplicate data, erroneous data and irrelevant information.
[0126] Keywords are extracted using the TF-IDF method. Based on the extracted keywords, accident keywords related to the construction site of the extra-long tunnel are identified. These keywords involve construction technology, equipment operation, personnel behavior, safety management, etc. In addition to extracting keywords, it is also necessary to count the accident frequency in historical accident records.
[0127] In this embodiment, the clustering algorithm used is the k-means++ algorithm, which includes a clustering loss function, which is expressed as follows:
[0128]
[0129] Where, L k is the clustering loss function, n1 is the number of data, n2 is the number of clusters, n2 is k or m, s i is the i-th data, z j is the cluster center of the jth cluster, λ is the regularization parameter, and ‖·‖ represents the distance between data.
[0130] In this embodiment, after obtaining the clustering result, it is matched with the constructed risk factor index system to update and improve the risk factor index system to obtain the final risk factor set, which includes the final first-level risk factor index and its index weight, and the final second-level risk factor index and its index weight. Since this embodiment uses text clustering when implementing the above clustering algorithm, the distance between the data is obtained by calculating the similarity between the word vectors of the keywords, and the cluster center obtained afterwards is also in the form of word vectors, so when matching, the first-level risk factor index or the second-level risk factor index is also processed as text and converted into vector form, so as to perform matching, and the method for calculating vector similarity can use cosine similarity.
[0131] This embodiment first matches each first cluster center with an existing first-level risk factor indicator. If the match is successful, directly enter step S26, indicating that the first cluster center already corresponds to an existing first-level risk factor indicator. If the match fails, the first cluster center is added as a new first-level risk factor indicator, and then enters step S26. Then match the second cluster center to which it belongs with the existing second-level risk factor indicator. If the match is successful, the accident frequency corresponding to the second cluster center is quantified as a weighting factor, and the initial weights of the first-level and second-level risk factor indicators are updated according to the weighting factor. If the match fails, the second cluster center is added as a new second-level risk factor indicator, and the initial weight of its second-level risk factor indicator is formed according to the accident frequency corresponding to the second cluster center, and the initial weight of its first-level risk factor indicator is formed according to the accident frequency of the first cluster center to which it belongs.
[0132] In this embodiment, when quantizing the weighting factor and updating the initial weight, the steps are as follows:
[0133] Assume that the accident frequency corresponding to the second cluster center is f, the weighting factor is w, the initial weight of the first-level risk factor indicator is w1, and the initial weight of the second-level risk factor indicator is w2.
[0134] Quantify the accident frequency as a weighting factor: Among them, f irepresents the accident frequency corresponding to each second cluster center, and k×m represents the total number of second cluster centers.
[0135] Update the initial weights of the first-level risk factor indicators: w'1 = w1 + γ × w, where γ is the learning rate.
[0136] Update the initial weights of the secondary risk factor indicators: w'2 = w2 + ξ × w, where ξ is the learning rate.
[0137] This embodiment can more finely analyze and evaluate the risk factors at the extra-long tunnel construction site by matching and updating the weights of the risk factor indicators. The weights of the risk factor indicators are updated in real time based on the frequency information in the historical accident records, which helps to adjust the risk management strategy in a timely manner. Through the process of clustering and matching, different types of risk factors can be more accurately identified and classified, which helps to formulate targeted risk management measures.
[0138] Specifically, in one embodiment of the present invention, step S3 includes the following steps:
[0139] S31 constructs a risk assessment model, which is a BP neural network model. The BP neural network model is divided into an input layer, a hidden layer and an output layer, and contains multiple neurons.
[0140] This embodiment selects to use the BP neural network model as the risk assessment model. The BP neural network is a neural network model suitable for solving nonlinear problems. In risk assessment, the relationship between risk factors is often a complex nonlinear relationship, and the BP neural network can better capture this nonlinear relationship. The BP neural network has a strong fitting ability and can learn complex data patterns and laws through training. This enables the BP neural network to model and predict risk factors more accurately.
[0141] S32 prepares a sample set, uses the sample set in combination with the risk factor set to train the risk assessment model, and uses an optimization algorithm to optimize the number of neurons during the training process to obtain a trained risk assessment model with N neurons.
[0142] The hidden layer in the BP neural network can help the model learn more complex features and relationships, and improve the expressive power of the model. By adjusting the number and structure of neurons in the hidden layer, the performance of the model can be further optimized. Therefore, this embodiment integrates the optimization algorithm into the training process, optimizes while training, and continuously adjusts the number and structure of neurons in the hidden layer, so that the model better fits the training data and improves the performance and accuracy of the model. The optimization algorithm can help find the optimal number of neurons in the hidden layer, so that the model can show good generalization ability on both training data and test data. With the assistance of the optimization algorithm, the model can avoid overfitting during the training process, that is, the model overfits the training data and loses the generalization ability. The optimization algorithm can help the model find the appropriate complexity to avoid the situation where the model is too complex on the training data and performs poorly on the test data.
[0143] S33 inputs the original data into the trained risk assessment model for prediction, and the model outputs the probability of occurrence and the probability of risk level distribution under each risk factor;
[0144] S34 generates a first probability matrix and a second probability matrix according to the results of model prediction. The first probability matrix describes the occurrence probability of each risk factor, and the second probability matrix describes the risk level distribution probability of each risk factor.
[0145] Specifically, see Figure 4 , step S32 comprises the following steps:
[0146] S321 collects historical data from the construction site of the extra-long tunnel to produce a sample set, and pre-processes the sample set, which includes reference values.
[0147] Collect historical data from the construction site of the extra-long tunnel to create a sample set. The historical data is collected according to the risk factor set, which includes data on various risk factors and corresponding risk levels or probability of occurrence data. The sample set is preprocessed to ensure the accuracy and completeness of the data. The sample set also includes reference values for model training and evaluation.
[0148] S322 determines the number of neurons in each layer and the network weight in the risk assessment model, wherein the number of neurons in the input layer is the same as the number of risk factors, and the number of neurons in the output layer is the dimension of the risk matrix;
[0149] S323 encodes the number of neurons in the hidden layer of the risk assessment model and uses it as an optimization target of the optimization algorithm;
[0150] S324 randomly generates a set of different numbers of hidden layer neurons as the initial population of the optimization algorithm;
[0151] S325 defines the network loss function of the risk assessment model as a fitness function;
[0152] S326 selects individuals to determine the network structure of the risk assessment model, inputs the sample set into the risk assessment model, passes through the input layer, the hidden layer and finally the output layer in sequence, the neurons of each layer calculate the output according to the input and the network weight, and pass it to the next layer, in the output layer, calculates the value of the network loss function according to the actual output value and the reference value, calculates the gradient of the network loss function to each network weight according to the value of the network loss function, and passes the gradient back to the network, and updates the network weight according to the update formula;
[0153] S327 evaluates the performance of the individual according to the value of the network loss function, and iteratively updates the population according to the evaluation result;
[0154] S328 repeats steps S326-S327 until the preset number of iterations is reached or the network loss function converges, and a trained risk assessment model is obtained, the number of neurons of which is N.
[0155] The training process of step S326 in this embodiment is as follows: Forward propagation: input data passes through the input layer, hidden layer and output layer, and the neurons in each layer calculate the output according to the input and weight, and pass it to the next layer. Calculate loss: In the output layer, the value of the network loss function is calculated according to the actual output and the expected output to evaluate the performance of the model. Backward propagation: According to the value of the network loss function, the gradient of each weight is calculated, and these gradients are passed back to the network to update the weight according to the update formula. Among them, the network loss function is:
[0156]
[0157] Where L is the network loss function, the first term is the loss function of the occurrence probability under each risk factor, the second term is the loss function of the risk level distribution probability under each risk factor, α and β are the weight parameters of the first and second terms respectively, M is the number of samples, is the true label of the i2th sample, which is 0 or 1. is the probability that the model predicts that the i2th sample category is 1, C is the total number of risk levels, is a binary indicator whose value is 0 or 1, indicating whether sample i2 belongs to category c. is the probability that the model predicts whether sample i2 belongs to category c.
[0158] The update formula is:
[0159]
[0160] In the formula, t is the current iteration number, t+1 represents the next iteration, Represents the new network weight value for the next iteration, which is used to connect neuron h2 and neuron h1. The weight update is based on the back propagation algorithm, which aims to reduce the prediction error of the network. Represents the network weight value in the current iteration, which also connects neuron h2 and neuron h1. This value is the weight value before the current iteration and is used as the basis for calculating the next weight update. It represents the error gradient of the h2th neuron for the i2th sample. In the output layer, this value is based on the actual output and expected output The difference between is calculated, which reflects the sensitivity of the output error to the network weight. In the hidden layer, It is calculated based on the function of the error gradient of the next layer and the output of the current layer, and is used to update the weights. Represents the output value of the i2th sample through the h2th neuron. In the input layer, is equal to the input value; in the hidden layer and the output layer, It is calculated by multiplying the input value with the corresponding weight, summing it, and then passing it through an activation function (such as the sigmoid function). ε is the learning speed. By iteratively adjusting the weights The network aims to minimize the predicted output The actual expected output , thereby improving the accuracy of the model.
[0161] The encoding method used in this embodiment is binary encoding, and a binary string is used to represent the number of neurons in each hidden layer. For example, if it is agreed that the number of neurons in each hidden layer is at most 256, an 8-bit binary number can be used to represent the number of neurons in a hidden layer.
[0162] Assume that the BP neural network contains at most 3 hidden layers, and the number of neurons in each hidden layer is at most 256. Each hidden layer can be assigned an 8-bit binary number, so the entire network structure can be represented by a 24-bit binary string.
[0163] Let's take an example to illustrate:
[0164] A specific network structure is 2 hidden layers, the first hidden layer has 64 neurons, and the second hidden layer has 128 neurons. It can be encoded as:
[0165] First hidden layer: 01000000 (binary representation of 64)
[0166] Second hidden layer: 10000000 (binary representation of 128)
[0167] Third hidden layer: Since there are only 2 hidden layers, the number of neurons in the third layer can be encoded as 0: 00000000
[0168] Full code: 010000001000000000000000
[0169] In this embodiment, during optimization, the network loss function is used as the fitness function. During the optimization process, the operations during iterative updating include selection, crossover and mutation. Selection: select individuals with better performance as parents according to the fitness function, and there is no need to modify the code. Crossover: single-point crossover, multi-point crossover or uniform crossover and other methods can be used. For example, in a single-point crossover, a point is randomly selected, and the coding parts of the two parents from this point to the end are exchanged to generate a new offspring. Mutation: Randomly select one or more bits in the code and flip them (0 to 1, 1 to 0) to introduce new mutations. In each generation of the optimization algorithm, the code needs to be converted back to the structure of the neural network for training and evaluation. The decoding process is to convert the binary string back to the corresponding number of neurons and construct a BP neural network with these parameters.
[0170] In this embodiment, if the iteration stop condition of training is not reached after the individual optimization iteration, for example, the loss function does not converge, the population is updated and the optimization and training process is restarted.
[0171] After the above-mentioned iterative optimization training, the obtained model will have the optimal number of neurons N.
[0172] Specifically, in one embodiment of the present invention, step S4 includes the following steps:
[0173] The S41 risk factor set includes p first-level risk factor indicators and q second-level risk factor indicators. The size of the first probability matrix P is p*k, and the size of the second probability matrix L is p*k. The risk matrix R is calculated based on the first probability matrix and the second probability matrix:
[0174]
[0175] in, represents element-wise multiplication, that is:
[0176]
[0177] Among them, g1 is the g1th first-level risk factor indicator, and g2 is the g2th second-level risk factor indicator;
[0178] S42 calculates the first comprehensive risk value of each first-level risk factor indicator:
[0179]
[0180] In the formula, It represents the first comprehensive risk value of the g1th first-level risk factor indicator;
[0181] S43 sorts the first-level risk factor indicators from high to low according to the first comprehensive risk value to obtain a first sequence, and determines the risk priority of the first sequence, and the first first-level risk factor indicator has the highest risk priority;
[0182] S44 calculates the second comprehensive risk value of each secondary risk factor indicator:
[0183]
[0184] In the formula, It represents the second comprehensive risk value of the g2-th secondary risk factor indicator;
[0185] S45 sorts the secondary risk factor indicators from high to low according to the second comprehensive risk value to obtain a second sequence, and determines the risk priority of the second sequence, with the first secondary risk factor indicator being the highest risk priority;
[0186] S46 According to The cross term of the first-level risk factor indicator and the second-level risk factor indicator is found in the first sequence and the second sequence, which is taken as a high-risk risk factor combination. The risk factor combination is analyzed to form a risk assessment result.
[0187] By constructing a risk matrix and adopting quantitative assessment methods, we can clarify the priority of risk factors at all levels, as well as the combination of risk factors that need to be focused on and prevented, which provides a basis for subsequent risk control.
[0188] Specifically, step S46 includes the following steps:
[0189] S461 sets the risk threshold and divides the risk into three levels: low risk, medium risk and high risk according to the risk threshold.
[0190] First, one or more risk thresholds need to be set based on historical data, expert experience or industry standards to distinguish different levels of risks.
[0191] S462 Calculate the risk value of each risk factor combination Comparing the risk value with the risk threshold, and classifying the risk factor combination as low risk, medium risk or high risk according to the comparison result;
[0192] S463 Filter out all risk factor combinations classified as high risk from the classification results; Filter out all risk factor combinations classified as high risk from the classification results. These are the intersections of primary and secondary risk factors that need to be focused on and managed.
[0193] S464 sorts the risk factors of high risk from high to low according to the risk value to obtain the third sequence, and marks the third sequence as the final risk priority; in order to effectively allocate resources and attention, it can be further based on The size of the value sorts all high-risk combinations to determine which risk combinations have the highest priority.
[0194] S465 analyzes the high-risk risk factors in the third sequence in turn, determines the risk causes, risk consequences, control measures and their effectiveness, forms a risk assessment report, and uses it as the risk assessment result. Detailed analysis is conducted on the identified high-risk factor combinations, including risk causes, possible consequences, effectiveness of existing control measures, etc., and these high-risk combinations and their priorities are clearly listed in the risk assessment report.
[0195] Through the above steps, the high-risk intersection items of primary risk factors and secondary risk factors in the construction site of the extra-long tunnel can be effectively identified, providing a scientific basis for formulating targeted risk control measures. This method not only helps managers understand the specific distribution of risks, but also helps to achieve effective allocation of resources and effective management of risks.
[0196] like Figure 5 As shown, on the other hand, the present invention designs a comprehensive risk assessment system based on the above detailed extra-long tunnel construction site risk assessment method. The system aims to comprehensively assess the risks of the extra-long tunnel construction site through a series of steps such as data collection, processing, analysis and model prediction, and provide targeted risk control suggestions. The system consists of five main parts: data collection module, data processing and analysis module, risk assessment model training module, risk assessment module and control suggestion generation module.
[0197] The data collection module is configured to collect raw data and historical data from the construction site of the extra-long tunnel, including construction logs, safety accident records, etc. In implementation, data is automatically collected using technical means such as sensors and log recording systems, and users are allowed to manually input or upload data.
[0198] The data processing and analysis module is configured to process and analyze the collected raw data to identify a set of risk factors; during its implementation, natural language processing technology is used to extract accident keywords, and a clustering algorithm is used to classify risk factors.
[0199] The risk assessment model training module is configured to build a risk assessment model, and train the risk assessment model in combination with an optimization algorithm to obtain a trained risk assessment model; design a neural network structure, including an input layer, a hidden layer, and an output layer, and design an activation function and a loss function. The model is trained using a sample set, and the network weights are optimized through repeated iterations until the network loss function converges or reaches a preset number of iterations.
[0200] The risk assessment module is configured to use the trained risk assessment model to evaluate the original data, predict the risk probability and the probability of risk level distribution, output the first probability matrix and the second probability matrix, combine the first probability matrix and the second probability matrix to form a risk matrix, and use the assessment method to quantitatively evaluate the risk matrix to obtain the risk assessment result; the model will predict the risk level based on the input real-time data, and generate a comprehensive risk assessment report in combination with the risk factor analysis results. The report not only includes the risk level, but also lists the main risk factors and their possible consequences in detail.
[0201] The control suggestion generation module is configured to select risk control measures based on the risk assessment results to form risk control suggestions. The control suggestion generation module selects appropriate risk control measures based on the risk level and probability output by the risk assessment model, combined with the pre-set risk control strategy and measure library. Based on the selected risk control measures, a detailed risk control suggestion is generated, including the specific content of the control measures, implementation methods, responsible persons, time nodes and other information.
[0202] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for risk assessment of extra-long tunnel construction site, characterized in that: The following steps are involved: S1 collects the original data of the extra-long tunnel construction site and obtains the historical accident records related to the extra-long tunnel construction site; S2 processes and analyzes historical accident records to identify a set of risk factors; Step S2 includes the following steps: S21 constructs a risk factor index system for the construction site of a super-long tunnel, including primary risk factor indicators and secondary risk factor indicators, and uses the analytic hierarchy process to assign initial weights to the primary risk factor indicators and secondary risk factor indicators; S22 determines the k value according to the number of the first-level risk factor indicators, and determines the m value according to the number of the second-level risk factor indicators; S23 extracts keywords from historical accident records to obtain accident keywords, wherein the historical accident records include accident frequencies; S24 clusters the accident keywords using a clustering algorithm to form k accident sets and k first cluster centers, and performs secondary clustering on the k accident sets respectively, and obtains m second cluster centers and m accident subsets for each accident set; S25 selects a first cluster center and matches it with the first-level risk factor indicator. If the match is successful, directly proceed to step S26; otherwise, the first cluster center is added as a new first-level risk factor indicator and then proceeds to step S26; S26 selects a second cluster center to which the first cluster center belongs to match the secondary risk factor indicator. If the match is successful, the accident frequency corresponding to the second cluster center is quantified as a weighting factor, and the initial weights of the primary risk factor indicator and the secondary risk factor indicator are updated according to the weighting factor; if the match fails, the second cluster center is added as a new secondary risk factor indicator, and the initial weight of its secondary risk factor indicator is formed according to the accident frequency corresponding to the second cluster center, and the initial weight of its primary risk factor indicator is formed according to the accident frequency of the first cluster center to which it belongs; S27: Repeat steps S25-S26 until all first cluster centers and second cluster centers are matched, and obtain the final first-level risk factor index, second-level risk factor index, and corresponding index weights to form a risk factor set; When quantizing the weighting factors and updating the initial weights, the steps are as follows: Assume that the accident frequency corresponding to the second cluster center is f, the weighting factor is w, the initial weight of the first-level risk factor indicator is w1, and the initial weight of the second-level risk factor indicator is w2; Quantify the accident frequency as a weighting factor: Among them, f i represents the accident frequency corresponding to each second cluster center, and k×m represents the total number of second cluster centers; Update the initial weight of the first-level risk factor indicator: w'1 = w1 + γ × w, where γ is the learning rate; Update the initial weights of the secondary risk factor indicators: w'2 = w2 + ξ × w, where ξ is the learning rate; S3 establishes a risk assessment model for the extra-long tunnel construction site and performs pre-training in combination with a risk factor set, inputs the original data into the trained risk assessment model, and obtains a first probability matrix and a second probability matrix, wherein the first probability matrix describes the probability of occurrence under each risk factor, and the second probability matrix describes the probability of risk level distribution under each risk factor; Step S3 includes the following steps: S31 constructs a risk assessment model, which is a BP neural network model. The BP neural network model is divided into an input layer, a hidden layer and an output layer, and contains multiple neurons; S32 prepares a sample set, uses the sample set in combination with the risk factor set to train the risk assessment model, and uses an optimization algorithm to optimize the number of neurons during the training process to obtain a trained risk assessment model, the number of neurons of which is N; S33 inputs the original data into the trained risk assessment model for prediction, and the model outputs the probability of occurrence and the probability of risk level distribution under each risk factor; S34 generates a first probability matrix and a second probability matrix according to the result of the model prediction, wherein the first probability matrix describes the occurrence probability of each risk factor, and the second probability matrix describes the risk level distribution probability of each risk factor; Step S32 includes the following steps: S321 collects historical data of the construction site of the extra-long tunnel to prepare a sample set, and pre-processes the sample set, the sample set including a reference value; S322 determines the number of neurons in each layer and the network weight in the risk assessment model, wherein the number of neurons in the input layer is the same as the number of risk factors, and the number of neurons in the output layer is the dimension of the risk matrix; S323 encodes the number of neurons in the hidden layer of the risk assessment model and uses it as an optimization target of the optimization algorithm; S324 randomly generates a set of different numbers of hidden layer neurons as the initial population of the optimization algorithm; S325 defines the network loss function of the risk assessment model as a fitness function; S326 selects individuals to determine the network structure of the risk assessment model, inputs the sample set into the risk assessment model, passes through the input layer, the hidden layer and finally the output layer in sequence, the neurons of each layer calculate the output according to the input and the network weight, and pass it to the next layer, in the output layer, calculates the value of the network loss function according to the actual output value and the reference value, calculates the gradient of the network loss function to each network weight according to the value of the network loss function, and passes the gradient back to the network, and updates the network weight according to the update formula; S327 evaluates the performance of the individual according to the value of the network loss function, and iteratively updates the population according to the evaluation result; S328 repeats steps S326-S327 until a preset number of iterations is reached or the network loss function converges, and a trained risk assessment model is obtained, the number of neurons of which is N; The network loss function is: Where L is the network loss function, the first term is the loss function of the occurrence probability under each risk factor, the second term is the loss function of the risk level distribution probability under each risk factor, α and β are the weight parameters of the first and second terms respectively, M is the number of samples, is the true label of the i2th sample, which is 0 or 1. is the probability that the model predicts that the i2th sample category is 1, C is the total number of risk levels, is a binary indicator whose value is 0 or 1, indicating whether sample i2 belongs to category c. is the probability that the model predicts whether sample i2 belongs to category c; In step S326, the update formula is: In the formula, t is the current iteration number, t+1 represents the next iteration, Represents the new network weight value for the next iteration, used to connect neuron h2 and neuron h1. Represents the network weight value in the current iteration, and also connects neuron h2 and neuron h1. represents the error gradient of the h2th neuron for the i2th sample, represents the output value of the i2th sample passing through the h2th neuron, and ε is the learning speed; S4 combines the first probability matrix and the second probability matrix to form a risk matrix, and uses an evaluation method to quantitatively evaluate the risk matrix to obtain a risk evaluation result; S5 formulates risk control recommendations based on risk assessment results.
2. A risk assessment method for a super-long tunnel construction site as claimed in claim 1, characterized in that: The clustering algorithm includes a clustering loss function, which is expressed as follows: Where, L k is the clustering loss function, n1 is the number of data, n2 is the number of clusters, n2 is k or m, s i is the i-th data, z j is the cluster center of the jth cluster, λ is the regularization parameter, and ‖·‖ represents the distance between data.
3. A risk assessment method for a super-long tunnel construction site as claimed in claim 1, characterized in that: Step S4 includes the following steps: The S41 risk factor set includes p first-level risk factor indicators and q second-level risk factor indicators. The size of the first probability matrix P is p*k, and the size of the second probability matrix L is p*k. The risk matrix R is calculated based on the first probability matrix and the second probability matrix: in, represents element-wise multiplication, that is: Among them, g1 is the g1th first-level risk factor indicator, and g2 is the g2th second-level risk factor indicator; S42 calculates the first comprehensive risk value of each first-level risk factor indicator: In the formula, It represents the first comprehensive risk value of the g1th first-level risk factor indicator; S43 sorts the first-level risk factor indicators from high to low according to the first comprehensive risk value to obtain a first sequence, and determines the risk priority of the first sequence, and the first first-level risk factor indicator has the highest risk priority; S44 calculates the second comprehensive risk value of each secondary risk factor indicator: In the formula, It represents the second comprehensive risk value of the g2-th secondary risk factor indicator; S45 sorts the secondary risk factor indicators from high to low according to the second comprehensive risk value to obtain a second sequence, and determines the risk priority of the second sequence, with the first secondary risk factor indicator being the highest risk priority; S46 According to The cross term of the first-level risk factor indicator and the second-level risk factor indicator is found in the first sequence and the second sequence, which is taken as a high-risk risk factor combination. The risk factor combination is analyzed to form a risk assessment result.
4. A risk assessment method for a super-long tunnel construction site as claimed in claim 3, characterized in that: Step S46 includes the following steps: S461 sets the risk threshold and divides the risk into three levels: low risk, medium risk and high risk according to the risk threshold; S462 Calculate the risk value of each risk factor combination Comparing the risk value with the risk threshold, and classifying the risk factor combination as low risk, medium risk or high risk according to the comparison result; S463 screens out all risk factor combinations classified as high risk from the classification results; S464 sorts the risk factor combinations with high risks from high to low according to the risk values to obtain a third sequence, and marks the third sequence with a final risk priority; S465 analyzes the high-risk risk factors in the third sequence in turn, determines the risk causes, risk consequences, control measures and their effectiveness, and forms a risk assessment report as the risk assessment result.
5. A risk assessment system for extra-long tunnel construction sites, characterized in that: The system is used to perform the method according to any one of claims 1 to 4, and the system comprises: A data acquisition module configured to collect raw data and historical data from a construction site of a super-long tunnel; A data processing and analysis module is configured to process and analyze the collected raw data to identify a set of risk factors; A risk assessment model training module is configured to construct a risk assessment model and train the risk assessment model in combination with an optimization algorithm to obtain a trained risk assessment model; A risk assessment module is configured to evaluate the original data using the trained risk assessment model, predict the risk probability and the risk level distribution probability, output a first probability matrix and a second probability matrix, combine the first probability matrix and the second probability matrix to form a risk matrix, and use an assessment method to quantitatively evaluate the risk matrix to obtain a risk assessment result; The control suggestion generation module is configured to select risk control measures based on the risk assessment results to form risk control suggestions.
