Crane safety performance grade evaluation method based on expert experience and deep learning

Abstract

The application discloses a crane safety performance grade evaluation method and system based on expert experience and deep learning, equipment and medium, relates to the technical field of crane safety evaluation, and the method comprises the steps that evaluation parameters in the running process of the crane are acquired, and the evaluation parameters are arranged into a feature vector sequence based on an expert scoring table; the correlation degree between each two feature vectors is calculated, and the initial weight coefficient of each feature vector is determined; a preset deep learning scoring model is used to obtain a scoring result corresponding to the two feature vectors; the scoring result is normalized, and a new feature vector is generated based on the normalized score and the feature vector sequence; the new feature vector is compared with an ideal feature vector corresponding to a reference safety grade given by an expert, and if there is a deviation, the model parameters of the deep learning scoring model are adjusted in reverse based on a loss function; and a corresponding safety performance grade is output. The application reduces the crane accident rate and ensures the safety of personnel and property.

Description

Technical Field

[0001] This invention relates to the field of crane safety assessment technology, and in particular to a method, system, equipment and medium for assessing the safety performance level of cranes based on expert experience and deep learning. Background Technology

[0002] With the continuous development of industrial production and the acceleration of urbanization, cranes have been widely used in various engineering construction, cargo loading and unloading, and industrial manufacturing. However, as a large-scale special equipment, cranes operate in complex and variable environments, involving high-altitude operations and heavy lifting, which are high-risk operations. Once a malfunction or safety accident occurs, it often results in serious casualties and property losses.

[0003] Traditional crane safety assessment methods have at least the following problems:

[0004] 1. Most of them are based on standard specifications and routine testing data. Although these methods can provide some basis for safety judgment, they have obvious limitations.

[0005] 2. Standards and specifications are often general and cannot fully cover all the complex working conditions and special situations that cranes actually operate under.

[0006] Therefore, how to more scientifically and accurately assess the safety performance level of cranes has become an urgent problem to be solved in the field of crane safety assessment. Summary of the Invention

[0007] In view of this, the purpose of this invention is to provide a method, system, equipment, and medium for assessing the safety performance level of cranes based on expert experience and deep learning, which can more accurately assess the safety performance level of cranes and reduce the incidence of crane accidents. The specific solution is as follows:

[0008] Firstly, this application discloses a method for evaluating the safety performance level of cranes based on expert experience and deep learning, including:

[0009] The evaluation parameters during crane operation are obtained, and the evaluation parameters are organized into a feature vector sequence based on an expert scoring table; wherein, the feature vector sequence includes a first feature vector that can be precisely quantified and a second feature vector that is fuzzily defined based on expert experience;

[0010] Iterate through every pair of feature vectors, calculate the degree of correlation between each pair of feature vectors, and determine the initial weight coefficient of each feature vector based on the degree of correlation.

[0011] Each pair of feature vectors is concatenated and then input into a pre-defined deep learning scoring model to obtain the scoring results for the corresponding two feature vectors.

[0012] The scoring results are normalized to obtain a normalized score, and a new feature vector is generated based on the normalized score and the feature vector sequence.

[0013] The new feature vector is compared with the ideal feature vector corresponding to the reference security level given by experts. If there is a deviation, the model parameters of the deep learning scoring model are adjusted in reverse based on the loss function.

[0014] The safety performance level of the crane is evaluated based on a trained deep learning scoring model, and the corresponding safety performance level is output.

[0015] Optionally, the step of acquiring evaluation parameters during crane operation and organizing the evaluation parameters into a feature vector sequence based on an expert scoring table includes:

[0016] The evaluation parameters during crane operation are obtained; the evaluation parameters include parameters corresponding to an expert scoring table constructed based on influencing factors S, F, and P; wherein, the influencing factor S is the degree of harm, including two levels, S1 and S2; the influencing factor F is the time of exposure to danger, including two levels, F1 and F2; and the influencing factor P is the possibility of avoiding danger, including two levels, P1 and P2.

[0017] The expert scoring table is used to distinguish between precisely quantifiable evaluation parameters and fuzzy-quantifiable evaluation parameters.

[0018] The evaluation parameters are organized into a feature vector sequence X=[x1,x2,...,x...]. n ], where n is the number of feature vectors, and the feature vector sequence includes a first feature vector that can be precisely quantified and a second feature vector that is fuzzily defined based on expert experience.

[0019] Optionally, the step of traversing every two feature vectors, calculating the correlation between every two feature vectors, and determining the initial weight coefficients of each feature vector based on the correlation includes:

[0020] Iterate through every two feature vectors X in the feature vector sequence i and X j Calculate X i With X j The degree of correlation between them;

[0021] Based on the correlation between each feature vector and all other feature vectors, the initial weight coefficients of each feature vector are determined so that each feature vector has a reasonable weight distribution when weighted summing, and the sum of all weight coefficients is 1.

[0022] Optionally, the step of concatenating every two feature vectors and inputting them into a preset deep learning scoring model to obtain a scoring result for the corresponding two feature vectors includes:

[0023] Take every two feature vectors X i and X j Concatenate them into a new vector Z=[X i ;X j ];

[0024] The new vector Z is input into a deep learning scoring model containing L hidden layers, and the scoring result is calculated using the following expression. :

[0025] (1),

[0026] In expression (1), L is the total number of hidden layers. The weight matrix of the output layer. This is the output of the hidden layer of layer L. This is the bias vector for the output layer.

[0027] Optionally, the step of normalizing the scoring result to obtain a normalized score, and generating a new feature vector based on the normalized score and the feature vector sequence, includes:

[0028] The scoring results are evaluated using the softmax function. Normalization is performed to convert the scores into normalized scores in probability distribution form. The expression is:

[0029] (2),

[0030] In expression (2), It is an exponential function. This represents the exponential sum of the scores of the i-th eigenvector and all n eigenvectors;

[0031] Based on the normalized score A new feature vector is generated by weighted summation of the corresponding feature vectors. The expression is:

[0032] (3),

[0033] In expression (3), This represents the j-th eigenvector in the eigenvector sequence.

[0034] Optionally, comparing the new feature vector with the ideal feature vector corresponding to the reference security level given by experts, and if there is a deviation, adjusting the model parameters of the deep learning scoring model in reverse based on the loss function, includes:

[0035] Obtain the reference safety level corresponding to the current crane as determined by experts based on experience, and determine the ideal feature vector corresponding to the reference safety level. ;

[0036] The generated new feature vector y is compared with the ideal feature vector, and the loss function value is calculated. The expression is:

[0037] (4),

[0038] Where the loss function is and m is the number of samples, For the new feature vector predicted by the model, The ideal feature vector corresponding to the security level for expert reference;

[0039] If the loss function value exceeds a preset threshold, backpropagation is used to adjust the model parameters of the deep learning scoring model in order to optimize the weight allocation for fuzzy quantization evaluation.

[0040] Optionally, the security level is divided into five levels, from PL a to PL e, each corresponding to a different level of security risk.

[0041] Secondly, this application discloses a crane safety performance level evaluation system based on expert experience and deep learning, comprising:

[0042] The feature extraction module is used to obtain evaluation parameters during the operation of the crane and organize the evaluation parameters into a feature vector sequence based on the expert scoring table; wherein, the feature vector sequence includes a first feature vector that can be precisely quantified and a second feature vector that is fuzzily defined based on expert experience;

[0043] The correlation calculation module is used to traverse every two feature vectors, calculate the correlation degree between every two feature vectors, and determine the initial weight coefficient of each feature vector based on the correlation degree.

[0044] The scoring module is used to concatenate every two feature vectors and input them into a preset deep learning scoring model to obtain the scoring results for the corresponding two feature vectors.

[0045] The normalization module is used to normalize the scoring results to obtain a normalized score, and to generate a new feature vector based on the normalized score and the feature vector sequence.

[0046] The optimization module is used to compare the new feature vector with the ideal feature vector corresponding to the reference security level given by experts. If there is a deviation, the model parameters of the deep learning scoring model are adjusted in reverse based on the loss function.

[0047] The evaluation module is used to evaluate the safety performance level of the crane based on a trained deep learning scoring model, and output the corresponding safety level.

[0048] Thirdly, this application discloses an electronic device characterized by comprising:

[0049] Memory, used to store computer programs;

[0050] A processor is used to execute the computer program to implement the steps of the aforementioned disclosed method for evaluating the safety performance level of cranes based on expert experience and deep learning.

[0051] Fourthly, this application discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the steps of the aforementioned disclosed method for evaluating the safety performance level of cranes based on expert experience and deep learning.

[0052] As can be seen, this application discloses the acquisition of evaluation parameters during crane operation, and the organization of these evaluation parameters into a feature vector sequence based on an expert scoring table. The feature vector sequence includes a first feature vector that can be precisely quantified and a second feature vector that is fuzzily defined based on expert experience. Each pair of feature vectors is traversed, the correlation between them is calculated, and the initial weight coefficients of each feature vector are determined based on the correlation. Each pair of feature vectors is concatenated and input into a preset deep learning scoring model to obtain a scoring result for the corresponding two feature vectors. The scoring result is normalized to obtain a normalized score, and a new feature vector is generated based on the normalized score and the feature vector sequence. The new feature vector is compared with the ideal feature vector corresponding to the reference safety level given by the experts. If a deviation exists, the model parameters of the deep learning scoring model are adjusted in reverse based on a loss function. The crane safety performance level is evaluated based on the trained deep learning scoring model to output the corresponding safety performance level.

[0053] Therefore, by combining expert experience and deep learning, this method can effectively integrate precisely quantifiable and fuzzily defined feature vectors to conduct a more scientific and accurate assessment of crane safety performance. In feature vector processing, it considers both the objective basis provided by precise data and the judgment of expert experience regarding complex working conditions and special situations. Determining initial weight coefficients through correlation calculations allows each feature vector to play a reasonable role in the assessment, avoiding the excessive influence of a single factor. In the scoring stage, the application of a deep learning scoring model makes the assessment of feature vectors more intelligent and accurate. Normalization transforms the scoring results into a more comparable form, and the generation of new feature vectors further integrates the information from each feature vector. The optimization module continuously optimizes the assessment process and improves the accuracy of the assessment by comparing with ideal feature vectors and adjusting model parameters in reverse. Safety performance level assessment based on the trained model provides a reliable reference for crane safety management. The classification of different safety levels, such as five levels from PLa to PLe, allows managers to clearly understand the safety risk level of the crane, thereby taking corresponding measures to reduce the crane accident rate and ensure the safety of personnel and property. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0055] Figure 1 This is a flowchart of a crane safety performance level evaluation method based on expert experience and deep learning disclosed in this application;

[0056] Figure 2 This is a schematic diagram of the structure of a crane safety performance level evaluation system based on expert experience and deep learning disclosed in this application;

[0057] Figure 3 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0058] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0059] With the continuous development of industrial production and the acceleration of urbanization, cranes have been widely used in various engineering construction, cargo loading and unloading, and industrial manufacturing. However, as a large-scale special equipment, cranes operate in complex and variable environments, involving high-altitude operations and heavy lifting, which are high-risk operations. Once a malfunction or safety accident occurs, it often results in serious casualties and property losses.

[0060] Traditional crane safety assessment methods have at least the following problems:

[0061] 1. Most of them are based on standard specifications and routine testing data. Although these methods can provide some basis for safety judgment, they have obvious limitations.

[0062] 2. Standards and specifications are often general and cannot fully cover all the complex working conditions and special situations that cranes actually operate under.

[0063] Therefore, the purpose of this invention is to provide a method, system, equipment, and medium for evaluating the safety performance level of cranes based on expert experience and deep learning, which can more accurately evaluate the safety performance level of cranes and reduce the incidence of crane accidents.

[0064] like Figure 1 As shown, this invention discloses a method for evaluating the safety performance level of cranes based on expert experience and deep learning, including:

[0065] Step S100: Obtain the evaluation parameters during the operation of the crane, and organize the evaluation parameters into a feature vector sequence based on the expert scoring table; wherein, the feature vector sequence includes a first feature vector that can be precisely quantified and a second feature vector that is fuzzily defined based on expert experience;

[0066] In this embodiment, evaluation parameters during crane operation are acquired and organized into a feature vector sequence based on an expert scoring table. Evaluation parameters can refer to physical quantities or operational indicators that are collected in real-time or periodically recorded during crane operation, reflecting its structural state, operating conditions, and safety risk level. Examples include lifting load, lifting height, operating speed, braking response time, wire rope wear, limit switch trigger frequency, wind speed, and track inclination. The expert scoring table can be a structured scoring tool compiled based on crane safety technical specifications, industry accident statistics, and the practical experience of senior engineers. It sets graded descriptions and corresponding score ranges for each parameter based on influencing factors. The first feature vector, which can be precisely quantified, can be a numerical vector obtained through direct sensor measurement, database query, or standardized calculation. Each dimension of this vector has a clear unit, dimension, and repeatability. For example... The first feature vector represents a ternary vector consisting of the current lifting load, height, and braking response time. The second feature vector, fuzzily defined based on expert experience, can be a vector generated through expert subjective judgment, semantic description transformation, or membership function mapping. Its dimensions do not have unique numerical solutions but are expressed as intervals, levels, linguistic variables, or probability distributions, for example... The ternary vectors formed after encoding; in this embodiment, the process of organizing the evaluation parameters into a sequence of feature vectors based on the expert scoring table is to classify all evaluation parameters into the category of influencing factors defined by the expert scoring table according to their type, and then to assign quantitative or fuzzy values ​​to each parameter according to the scoring rules, and finally arrange them in a unified dimensional order to form an ordered vector set; this process ensures that objective data and expert experience are aligned and integrated in the same mathematical space, providing a structurally consistent input basis for subsequent association modeling and deep learning processing.

[0067] Understandably, this application may, for example, categorize the collected assessment parameters according to a pre-defined classification system of three influencing factors in an expert scoring table: S (Depth of Injury: S1 - Minor injury, recoverable minor injury; S2 - Serious injury, irreversible serious injury or death), F (Time of Exposure to Danger: F1 - Short-term exposure, occasional entry into the danger zone; F2 - Long-term exposure, frequent or continuous exposure to the danger zone), and P (Possibility of Avoiding Danger: P1 - Possibly avoidable, sufficient time and conditions to avoid the danger; P2 - Difficult to avoid, sudden danger or insufficient conditions for avoidance). The mapping is performed, and then numerical calibration or linguistic variable encoding is performed respectively to generate a sequence of feature vectors. For example, this application can also cross-score the performance of the same evaluation parameter under multiple influence dimensions based on the weight coefficients and scoring scales corresponding to different influencing factors in the expert scoring table, and concatenate the scores of each dimension into sub-vectors, and then combine them into a complete sequence of feature vectors. Furthermore, this application can also store the expert scoring table in the form of a relational table in a local database. After the system obtains the evaluation parameters, it automatically matches the corresponding row records and calls the built-in mapping function to complete the conversion of parameters to feature vectors.

[0068] For example, this application could be based on the following evaluation parameters collected during the daily inspection of a tower crane: lifting load of 18.6t (rated maximum load 22.7t), lifting height of 72.3m (maximum design height 80m), braking response time of 0.38s (upper limit of safety threshold), slight broken wires visible on the surface of the wire rope (expert judgment: slight wear), intermittent abnormal noise in the slewing mechanism (expert judgment: occasional abnormality), and the intervals between the three most recent maintenance records being 14 days, 16 days, and 15 days respectively (expert overall judgment: good maintenance timeliness); based on the SFP expert scoring table, the system categorizes the first three parameters into the precisely quantifiable category, and after normalization, forms the first feature vector. The normalization process specifically involves dividing the collected physical quantity value by the preset safety threshold or maximum design value of that type of parameter. For example, the lifting height is divided by the maximum design height of the crane, and the braking response time is divided by the maximum response time allowed by safety regulations, thus obtaining a dimensionless feature value between 0 and 1. The latter three parameters are then encoded using linguistic variables (e.g., slight wear → 0.3, occasional abnormality → 0.4, good maintenance timeliness → 0.85) to form the second feature vector. The final result is a sequence containing two feature vectors. .

[0069] In this way, by using an expert scoring table to structure and semantically map the multi-source heterogeneous evaluation parameters, the objectivity and verifiability of sensor data are preserved, while also incorporating the comprehensive judgment capabilities of senior engineers on dynamic risk scenarios. This provides a high-quality input foundation that combines numerical rigidity and semantic flexibility for subsequent modeling, ensuring the engineering interpretability and industry credibility of the model output.

[0070] Step S200: Traverse every two feature vectors, calculate the degree of correlation between every two feature vectors, and determine the initial weight coefficient of each feature vector based on the degree of correlation;

[0071] In this embodiment, every pair of feature vectors is traversed, the correlation between each pair of feature vectors is calculated, and the initial weight coefficient of each feature vector is determined based on the correlation. Here, each pair of feature vectors can refer to a sequence of feature vectors. Any ordered combination ,in and The degree of correlation can refer to a metric that characterizes whether two feature vectors have a synergistic enhancement, mutual exclusion, or redundant substitution relationship in the dimension of security risk representation. Its calculation does not depend on the specific numerical value, but focuses on the directional consistency, distribution similarity, or semantic relevance between the vectors; the initial weight coefficient can be assigned to each feature vector. nonnegative real numbers This is used to adjust its contribution ratio to the overall security assessment result during subsequent weighted fusion, and satisfies... In this embodiment, the process of determining the initial weight coefficients of each feature vector based on the degree of correlation is to first calculate the average degree of correlation between each feature vector and all other feature vectors, then use this value as its initial importance measure, and finally obtain the final weight coefficients through normalization. This mechanism enables feature vectors that frequently participate in strong correlations to obtain higher weights, avoids isolated parameters or low-correlation parameters from dominating the evaluation results, and improves the robustness of the model.

[0072] Understandably, this application could, for example, calculate the directional consistency between every two feature vectors based on cosine similarity, use it as a correlation value, and apply it to each... Calculate its relationship with the rest The average cosine similarity of the vectors is then normalized using softmax to obtain the initial weight coefficients, providing initialization guidance that aligns with engineering principles for subsequent deep learning scoring models.

[0073] For example, this application can be for the above-mentioned feature vector sequence System calculation and The cosine similarity is 0.88, and since it contains only two vectors, therefore... The average correlation was 0.88. The average correlation coefficient was also 0.88; after normalization, the initial weight coefficients were obtained. If the sequence is subsequently expanded to include five feature vectors, then each vector will be correlated with the other four vectors, and a differential weight allocation will be generated accordingly.

[0074] In this way, by explicitly modeling the semantic / statistical coupling relationship between feature vectors and generating normalized initial weights, the subjective bias caused by manually setting weights is avoided, and the weight imbalance caused by differences in feature dimensions, fuzzy quantization distortion and the subjectivity of expert experience is alleviated. This provides a physically interpretable and engineering-reasonable initial parameter configuration for subsequent deep learning scoring models, and significantly enhances the model's adaptability and robustness to multi-source heterogeneous parameters under complex working conditions.

[0075] Step S300: Concatenate every two feature vectors and input them into a preset deep learning scoring model to obtain the scoring results for the corresponding two feature vectors;

[0076] In this embodiment, every two feature vectors are concatenated and then input into a preset deep learning scoring model to obtain the scoring results for the corresponding two feature vectors; wherein, concatenating every two feature vectors can refer to combining the feature vectors... and Perform vector concatenation operations either column-wise or row-wise to form a new input vector. The dimension of the deep learning scoring model is the sum of the dimensions of the original vectors. The preset deep learning scoring model can be a feedforward neural network with a known structure and parameters to be trained. Its input layer receives the concatenated feature vectors, the hidden layer mines the interaction relationship between features through nonlinear transformation, and the output layer generates a single real value, representing the joint contribution strength of the feature to the combination in the safety risk dimension. The scoring result can be a continuous numerical value output by the model. Its absolute value has no physical meaning, but its relative size can be used to characterize the strength of the risk synergy effect of different features on the combination. In this embodiment, the process of concatenating every two feature vectors and inputting them into the preset deep learning scoring model is to explicitly construct pairwise interactive input samples while maintaining the original semantic structure of the features. This allows the model to learn implicit coupling patterns that are difficult for expert experience to exhaustively enumerate (e.g., high load + moderate wear may cause sudden fracture, while low load + moderate wear is controllable risk), thereby making up for the insufficient generalization ability of pure rule methods.

[0077] It is understandable that this application may, for example, be a concatenation of vectors. The input is fed into a deep learning scoring model containing two fully connected hidden layers, each followed by a ReLU activation function. The output layer uses linear activation to output the scoring result. ;

[0078] (1),

[0079] In expression (1), L is the total number of hidden layers. The weight matrix of the output layer. This is the output of the hidden layer of layer L. This is the bias vector for the output layer.

[0080] For example, this application may be the application obtained above. and spliced ​​as Input the data into a preset deep learning scoring model and output the scoring results. Similarly, we can obtain (Due to the asymmetric design of the model, directional sensitivity is allowed); the results indicate that the combination of the two exhibits a strong risk synergy effect under the current operating conditions.

[0081] In this way, by concatenating the original feature vectors pairwise and inputting them into a deep learning scoring model, higher-order interaction relationships between features are explicitly modeled. While preserving the semantic integrity of expert experience and objective data, potential synergistic or inhibitory effects are explored, effectively making up for the shortcomings of traditional rule-based methods in identifying complex coupling risks. This lays a solid data-driven foundation for the subsequent generation of context-aware new feature vectors.

[0082] Step S400: Normalize the scoring result to obtain a normalized score, and generate a new feature vector based on the normalized score and the feature vector sequence;

[0083] In this embodiment, the scoring result is normalized to obtain a normalized score, and a new feature vector is generated based on the normalized score and the feature vector sequence; wherein, normalization processing may refer to normalizing the same feature vector... All corresponding scores Normalized scores mapped to probability distribution form To satisfy and Normalized scores can reflect... and The soft attention weights, ranked by relative importance, are used to guide the subsequent weighted fusion process; the new feature vector can be obtained by weighting and summing all vectors in the feature vector sequence according to their normalized scores.

[0084] (3),

[0085] In expression (3), This represents the j-th eigenvector in the eigenvector sequence;

[0086] In this embodiment, the process of generating new feature vectors based on normalized scores and feature vector sequences realizes the leap from local pairwise scoring to global feature reconstruction, enabling each original feature vector to absorb information from other features, forming a more discriminative and robust higher-order representation, and effectively alleviating the evaluation bias caused by single-point parameter distortion.

[0087] It is understood that this application may, for example, employ a softmax function for each The corresponding set of scores is normalized, i.e. Then perform a weighted summation to generate a new feature vector;

[0088] (2),

[0089] In expression (2), It is an exponential function. This represents the exponential sum of the scores of the i-th eigenvector and all n eigenvectors;

[0090] For example, this application may be based on the above-described results. (Self-related) Calculate the softmax value. , This leads to the generation of new feature vectors. This vector preserves both The core of precise operating parameters, and also incorporates The experiential risk tendency it carries has a stronger comprehensive characterization ability.

[0091] In this way, by weighting and fusing the original feature vectors with normalized scores, each feature vector can be re-represented in the global context. This not only inherits the objectivity of quantifiable parameters but also incorporates the semantic flexibility of expert experience parameters, thereby generating new feature vectors that are discriminative, robust, and interpretable. This provides high-quality intermediate representation support for subsequent model optimization using expert knowledge as a supervision signal.

[0092] Step S500: Compare the new feature vector with the ideal feature vector corresponding to the reference security level given by the expert. If there is a deviation, adjust the model parameters of the deep learning scoring model in reverse based on the loss function.

[0093] In this embodiment, the new feature vector is compared with the ideal feature vector corresponding to the reference safety level given by the expert. If there is a deviation, the model parameters of the deep learning scoring model are adjusted in reverse based on the loss function. The reference safety level given by the expert can refer to the safety level to which the current crane should belong, determined by a qualified crane safety assessment expert based on on-site inspection, historical data, and industry experience, such as PL c level. The ideal feature vector can be a standard feature template predefined by the expert for each safety level (PL a–PL e), consisting of several typical parameter combinations and their reasonable value ranges, representing the ideal safety state of the crane at that level. The deviation can refer to the new feature vector... With the corresponding ideal feature vector The numerical difference is measured in terms of Euclidean distance, cosine distance, or weighted MSE; the loss function can be an objective function used in supervised learning to measure the degree of inconsistency between the predicted output and the true label, and its gradient can drive the model parameters to be updated in the direction of reducing error; in this embodiment, the process of comparing the new feature vector with the ideal feature vector corresponding to the reference safety level given by the expert constructs a supervision signal with expert knowledge as the gold standard, so that the learning objective of the deep learning model is no longer limited to data fitting, but approximates the ideal safety state representation under the expert's cognitive system; this mechanism ensures that the model output not only has statistical rationality, but also engineering interpretability and industry acceptance.

[0094] It is understood that this application may, for example, use mean squared error (MSE) as the loss function, i.e. After taking the partial derivative of the loss function with respect to the model parameters, the weights and biases are updated using stochastic gradient descent, where:

[0095] (4),

[0096] Where the loss function is and m is the number of samples, For the new feature vector predicted by the model, The ideal feature vector corresponding to the security level for expert reference;

[0097] For example, this application can be based on expert determination that the current crane belongs to PL c level, and its corresponding ideal feature vector is: The above results Substituting into the MSE loss function, we can calculate... If the value is greater than the preset threshold Then, backpropagation is initiated to update the weight matrices of layers 1 and 2 in the deep learning scoring model. , and bias vector , .

[0098] Step S600: Evaluate the safety performance level of the crane based on the trained deep learning scoring model, and output the corresponding safety performance level;

[0099] Here, a well-trained deep learning scoring model can refer to a deep learning scoring model that has completed multiple rounds of parameter updates, whose loss function has converged to a stable region, and which has achieved a predetermined accuracy requirement on the validation set; the security performance level can be a PL (Performance Level) level according to the ISO13849-1 standard, including five levels from PLa to PLe, corresponding to the lowest to the highest security risk control capabilities; the output corresponding security performance level can be the newly generated feature vector. Each of the five levels of ideal feature vectors The distance is calculated, and the level corresponding to the smallest distance is selected as the final output result. In this embodiment, the process of evaluating the safety performance level of the crane based on the trained deep learning scoring model completes a complete closed loop from original operating parameters → feature vector → association modeling → context enhancement → expert alignment → level determination. This ensures that the evaluation results are both rooted in real-time data and anchored to expert consensus, thus possessing both objectivity and authority.

[0100] Understandably, this application could, for example, calculate the Euclidean distance between the newly generated feature vector and each of the five ideal feature vectors, and select the level to which the ideal feature vector with the smallest distance belongs as the output; alternatively, it could input the new feature vector into a pre-trained lightweight classifier (such as a Softmax output layer), directly output the probability distribution of each level, and take the level corresponding to the highest probability as the result; furthermore, this application could combine a confidence threshold and a distance outlier detection mechanism to label the evaluation reliability level (such as high confidence, requiring verification) while outputting the level. Based on any of the above methods, this application obtains a clear, interpretable, and internationally compliant security performance level output, facilitating maintenance personnel to quickly identify risk levels and take tiered control measures.

[0101] For example, this application may be a new feature vector obtained after the above optimization. The Euclidean distances are calculated with the ideal eigenvectors of Pa to Pe respectively, and the results are as follows: The minimum value of 0.023 corresponds to the PLC level, therefore the system outputs the current crane safety performance level as PLC.

[0102] In this way, by constructing an end-to-end learning loop with expert knowledge as the supervised target, the training objective of deep learning models is elevated from simply fitting data to approximating the ideal safety state under the expert cognitive system. This enables the model output to not only have statistical generalization ability, but also engineering authority and decision support value. At the same time, the final output PL level is directly mapped to internationally accepted safety standards, which can be seamlessly integrated with existing industrial operation and maintenance processes and risk management strategies, significantly improving the feasibility and management adaptability of the technical solution in real-world scenarios.

[0103] As can be seen, this application discloses the acquisition of evaluation parameters during crane operation, and the organization of these evaluation parameters into a feature vector sequence based on an expert scoring table. The feature vector sequence includes a first feature vector that can be precisely quantified and a second feature vector that is fuzzily defined based on expert experience. Each pair of feature vectors is traversed, the correlation between them is calculated, and the initial weight coefficients of each feature vector are determined based on the correlation. Each pair of feature vectors is concatenated and input into a preset deep learning scoring model to obtain a scoring result for the corresponding two feature vectors. The scoring result is normalized to obtain a normalized score, and a new feature vector is generated based on the normalized score and the feature vector sequence. The new feature vector is compared with the ideal feature vector corresponding to the reference safety level given by the experts. If a deviation exists, the model parameters of the deep learning scoring model are adjusted in reverse based on a loss function. The crane safety performance level is evaluated based on the trained deep learning scoring model to output the corresponding safety performance level. Therefore, this approach achieves a structured alignment between objective data and expert experience through expert scoring tables, assigns reasonable initial weights to features using association analysis, models complex interactions between features using deep learning models, generates context-aware new feature vectors through normalization mechanisms, and continuously optimizes model parameters using expert-defined ideal feature vectors as the supervised target, ultimately outputting a safety performance level that meets international standards. This solution organically integrates expert experience with data-driven models, retaining the flexibility and adaptability of human judgment while leveraging the generalization and stability of machine learning, significantly improving the scientific rigor, accuracy, and practicality of crane safety assessments.

[0104] like Figure 2 As shown, the present invention also discloses a crane PL safety level assessment system based on expert experience and deep learning, comprising:

[0105] The feature extraction module 11 is used to acquire evaluation parameters during the operation of the crane and organize the evaluation parameters into a feature vector sequence based on the expert scoring table; wherein, the feature vector sequence includes a first feature vector that can be precisely quantified and a second feature vector that is fuzzily defined based on expert experience;

[0106] The correlation calculation module 12 is used to traverse every two feature vectors, calculate the correlation degree between every two feature vectors, and determine the initial weight coefficient of each feature vector based on the correlation degree.

[0107] The scoring module 13 is used to concatenate every two feature vectors and input them into a preset deep learning scoring model to obtain the scoring results of the corresponding two feature vectors.

[0108] The normalization module 14 is used to normalize the scoring result to obtain a normalized score, and generate a new feature vector based on the normalized score and the feature vector sequence.

[0109] The optimization module 15 is used to compare the new feature vector with the ideal feature vector corresponding to the reference security level given by the expert. If there is a deviation, the model parameters of the deep learning scoring model are adjusted in reverse based on the loss function.

[0110] Evaluation module 16 is used to evaluate the safety performance level of the crane based on a trained deep learning scoring model, and output the corresponding safety level.

[0111] Therefore, the feature extraction module completes the structured mapping of multi-source heterogeneous parameters through an expert scoring table, ensuring that the input data is both objective and semantically interpretable; the association calculation module generates initial weights based on the coupling relationship between features, providing the model with a reasonable starting configuration; the scoring module mines the implicit interaction effect between feature pairs through deep neural networks, making up for the insufficient generalization ability of rule-based methods; the normalization module realizes the leap from local scoring to global context representation, improving the robustness of the model; the optimization module uses expert knowledge as a supervision signal to drive the model to continuously approach the ideal safety state, ensuring the authority of the output; and the evaluation module maps the continuous model output to the international standard PL level, realizing an effective closed loop from evaluation results to operation and maintenance decisions.

[0112] Furthermore, embodiments of this application also disclose an electronic device, Figure 3 This is a structural diagram of an electronic device according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application.

[0113] Figure 3 This is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of this application. Specifically, the electronic device 20 may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the crane safety performance level assessment method based on expert experience and deep learning disclosed in any of the foregoing embodiments. Alternatively, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0114] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0115] The processor 21 may include one or more processing cores, such as a quad-core processor or an octa-core processor. The processor 21 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 21 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, the processor 21 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, the processor 21 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0116] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon can include operating system 221, computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0117] The operating system 221 manages and controls the various hardware devices and computer programs 222 on the electronic device 20 to enable the processor 21 to perform calculations and processing on the massive amounts of data 223 in the memory 22. It can be Windows Server, Netware, Unix, Linux, etc. The computer program 222, in addition to including a computer program capable of performing the crane safety performance level assessment method based on expert experience and deep learning, which is executed by the electronic device 20 according to any of the aforementioned embodiments, may further include computer programs capable of performing other specific tasks. The data 223 may include data received by the electronic device from external devices, as well as data collected by its own input / output interface 25.

[0118] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned disclosed method for evaluating the safety performance level of cranes based on expert experience and deep learning. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0119] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0120] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application. The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly in hardware, software modules executed by a processor, or a combination of both. The software module may be located in random access memory (RAM), memory, read-only memory (ROM), electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, removable disks, CD-ROMs (Compact Disc-Read Only Memory), or any other form of storage medium known in the art.

[0121] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0122] The solution provided by the present invention has been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for evaluating the safety performance level of cranes based on expert experience and deep learning, characterized in that, include: The evaluation parameters during crane operation are obtained, and the evaluation parameters are organized into a feature vector sequence based on an expert scoring table; wherein, the feature vector sequence includes a first feature vector that can be precisely quantified and a second feature vector that is fuzzily defined based on expert experience; Iterate through every pair of feature vectors, calculate the degree of correlation between each pair of feature vectors, and determine the initial weight coefficient of each feature vector based on the degree of correlation. Each pair of feature vectors is concatenated and then input into a pre-defined deep learning scoring model to obtain the scoring results for the corresponding two feature vectors. The scoring results are normalized to obtain a normalized score, and a new feature vector is generated based on the normalized score and the feature vector sequence. The new feature vector is compared with the ideal feature vector corresponding to the reference security level given by experts. If there is a deviation, the model parameters of the deep learning scoring model are adjusted in reverse based on the loss function. The safety performance level of the crane is evaluated based on a trained deep learning scoring model, and the corresponding safety performance level is output.

2. The crane safety performance level assessment method based on expert experience and deep learning according to claim 1, characterized in that, The process of acquiring evaluation parameters during crane operation and organizing these parameters into a feature vector sequence based on an expert scoring table includes: The evaluation parameters during crane operation are obtained; the evaluation parameters include parameters corresponding to an expert scoring table constructed based on influencing factors S, F, and P; wherein, the influencing factor S is the degree of harm, including two levels, S1 and S2; the influencing factor F is the time of exposure to danger, including two levels, F1 and F2; and the influencing factor P is the possibility of avoiding danger, including two levels, P1 and P2. The expert scoring table is used to distinguish between precisely quantifiable evaluation parameters and fuzzy-quantifiable evaluation parameters. The evaluation parameters are organized into a feature vector sequence X=[x1,x2,...,x...]. n ], where n is the number of feature vectors, and the feature vector sequence includes a first feature vector that can be precisely quantified and a second feature vector that is fuzzily defined based on expert experience.

3. The crane safety performance level assessment method based on expert experience and deep learning according to claim 1, characterized in that, The process of traversing every pair of feature vectors, calculating the correlation between each pair of feature vectors, and determining the initial weight coefficients for each feature vector based on the correlation includes: Iterate through every two feature vectors X in the feature vector sequence i and X j Calculate X i With X j The degree of correlation between them; Based on the correlation between each feature vector and all other feature vectors, the initial weight coefficients of each feature vector are determined so that each feature vector has a reasonable weight distribution when weighted summing, and the sum of all weight coefficients is 1.

4. The crane safety performance level evaluation method based on expert experience and deep learning according to claim 2, characterized in that, The step of concatenating every two feature vectors and inputting them into a preset deep learning scoring model to obtain a scoring result for the corresponding two feature vectors includes: Take every two feature vectors X i and X j Concatenate them into a new vector Z=[X i ;X j ]; The new vector Z is input into a deep learning scoring model containing L hidden layers, and the scoring result is calculated using the following expression. : (1), In expression (1), L is the total number of hidden layers. The weight matrix of the output layer. This is the output of the hidden layer of layer L. This is the bias vector for the output layer.

5. The crane safety performance level assessment method based on expert experience and deep learning according to claim 4, characterized in that, The step of normalizing the scoring result to obtain a normalized score, and generating a new feature vector based on the normalized score and the feature vector sequence, includes: The scoring results are evaluated using the softmax function. Normalization is performed to convert the scores into normalized scores in probability distribution form. The expression is: (2), In expression (2), It is an exponential function. This represents the exponential sum of the scores of the i-th eigenvector and all n eigenvectors; Based on the normalized score A new feature vector is generated by weighted summation of the corresponding feature vectors. The expression is: (3), In expression (3), This represents the j-th eigenvector in the eigenvector sequence.

6. The crane safety performance level assessment method based on expert experience and deep learning according to claim 5, characterized in that, The step of comparing the new feature vector with the ideal feature vector corresponding to the reference security level given by experts, and if there is a deviation, adjusting the model parameters of the deep learning scoring model in reverse based on the loss function, includes: Obtain the reference safety level corresponding to the current crane as determined by experts based on experience, and determine the ideal feature vector corresponding to the reference safety level. ; The generated new feature vector y is compared with the ideal feature vector, and the loss function value is calculated. The expression is: （4）， Where the loss function is and m is the number of samples, For the new feature vector predicted by the model, The ideal feature vector corresponding to the security level for expert reference; If the loss function value exceeds a preset threshold, backpropagation is used to adjust the model parameters of the deep learning scoring model in order to optimize the weight allocation for fuzzy quantization evaluation.

7. The crane safety performance level assessment method based on expert experience and deep learning according to claim 1, characterized in that, The security levels are divided into five levels, from PL a to PL e, each corresponding to a different level of security risk.

8. A crane safety performance level evaluation system based on expert experience and deep learning, characterized in that, include: The feature extraction module is used to obtain evaluation parameters during the operation of the crane and organize the evaluation parameters into a feature vector sequence based on the expert scoring table; wherein, the feature vector sequence includes a first feature vector that can be precisely quantified and a second feature vector that is fuzzily defined based on expert experience; The correlation calculation module is used to traverse every two feature vectors, calculate the correlation degree between every two feature vectors, and determine the initial weight coefficient of each feature vector based on the correlation degree. The scoring module is used to concatenate every two feature vectors and input them into a preset deep learning scoring model to obtain the scoring results for the corresponding two feature vectors. The normalization module is used to normalize the scoring results to obtain a normalized score, and to generate a new feature vector based on the normalized score and the feature vector sequence. The optimization module is used to compare the new feature vector with the ideal feature vector corresponding to the reference security level given by experts. If there is a deviation, the model parameters of the deep learning scoring model are adjusted in reverse based on the loss function. The evaluation module is used to evaluate the safety performance level of the crane based on a trained deep learning scoring model, and output the corresponding safety level.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor is configured to execute the computer program to implement the steps of the crane safety performance level assessment method based on expert experience and deep learning as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store a computer program; wherein, when the computer program is executed by a processor, it implements the steps of the crane safety performance level evaluation method based on expert experience and deep learning as described in any one of claims 1 to 7.

Images