Auxiliary decision-making method and decision-making system for highway bridge maintenance

Abstract

The invention provides an auxiliary decision-making method and decision-making system for highway bridge maintenance, and relates to a bridge maintenance management technology.The method comprises the steps that firstly, historical bridge maintenance data is preprocessed, including training data cleaning and input data integrity checking, so that the data quality is guaranteed; secondly, constructing a decision tree model based on a CART algorithm, selecting an optimal split feature through recursion, and generating a decision rule which is easy to understand and execute; and finally, classifying the feature data of the to-be-decided bridge by using the decision tree model, and outputting a maintenance measure decision result. Meanwhile, according to the method, the decision-making tree is visually displayed, the interpretability of the decision-making process is improved, and incremental learning is supported to adapt to new maintenance requirements. According to the invention, subjectivity and limitation of artificial experience decision making can be overcome, scientificity, adaptability and efficiency of bridge maintenance decision making are improved, and technical support is provided for intelligence of bridge maintenance management.

Description

Technical Field

[0001] The present invention relates to the technical field of bridge maintenance management, and in particular to a highway bridge maintenance auxiliary decision-making method and a decision-making system. Background Art

[0002] As an important part of transportation, the safety and stability of highway bridges are directly related to the safety of public travel and the normal operation of the social economy. With the increase in traffic volume and the impact of the natural environment, bridge structures are gradually aging and diseases are increasing. Therefore, regular bridge maintenance is essential. Traditional bridge maintenance measures often rely on expert experience and personal judgment. This approach cannot meet the growing maintenance needs and is prone to decision-making deviations due to personal differences.

[0003] In recent years, with the rapid development of information technology, a large amount of data has been accumulated in the process of highway bridge maintenance, including but not limited to bridge structural data, previous maintenance records, technical condition scores, etc. These data make it possible to realize the digitalization of bridge maintenance management. However, at present, when various bridge maintenance units formulate highway bridge maintenance measures, there is a general problem of insufficient data utilization, which makes it difficult to adjust the maintenance strategy in time and the maintenance effect is poor. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for auxiliary decision-making for highway bridge maintenance, which can reduce the uncertainty of human experience judgment through data analysis, improve the scientificity and accuracy of decision-making, and further optimize maintenance strategies and improve maintenance efficiency.

[0005] To achieve the above object, the present invention is implemented through the following technical solutions:

[0006] A method for assisting decision making in highway bridge maintenance includes the following steps:

[0007] S1: Perform data preprocessing, including cleaning training data and checking the integrity of input data;

[0008] S2: Based on the data preprocessed in S1, a decision tree model for bridge maintenance measures is constructed using the CART algorithm;

[0009] S3: Use the decision tree model constructed in S2 to output the decision results of bridge maintenance measures

[0010] Further: the specific steps of step S2 include:

[0011] S21: Select the feature with the smallest Gini impurity as the split feature of the current node;

[0012] S22: Split the data set of the current node according to the selected split feature to generate child nodes;

[0013] S23: recursively execute S21 and S22 until a preset stop condition is reached to generate a decision tree.

[0014] Further: the data preprocessing in step S1 specifically includes:

[0015] The key fields in the maintenance decision records, such as superstructure score, substructure score, bridge deck score, bridge technical condition grade, maintenance inspection grade, time since last maintenance, and maintenance decision results, were scanned, and records with missing key fields were removed to form a training data set.

[0016] Further: the data preprocessing in step S1 also includes:

[0017] Check the integrity of the bridge data to be predicted. If the key fields are completely missing, mark it as requiring manual maintenance decision-making. If only individual fields are missing, fill the missing numerical fields with the median of the field in all the data, and fill the missing categorical fields with the mode of the field in all the data.

[0018] Further: the stopping condition in step S23 is: at least one of the following: the Gini impurity of the node is less than a preset threshold, the depth of the node reaches a preset maximum depth, and the number of samples of the node is less than a preset threshold.

[0019] Further: The calculation formula of Gini impurity is:

[0020]

[0021] Where k represents the total number of categories of bridge maintenance measures;

[0022] p i Represents the proportion of the i-th category in the data set.

[0023] Further: in step S3, the decision tree is displayed in a visual manner, wherein each node represents a classification condition of bridge characteristic data, including the Gini impurity under the condition, the number of samples, the number of samples contained in each type of bridge characteristic data, and the recommended classification of bridge maintenance measures.

[0024] Furthermore: the bridge maintenance measures include preventive maintenance of the entire bridge, daily maintenance of the entire bridge, demolition and reconstruction of the entire bridge, reinforcement of the entire bridge, repair of the entire bridge, partial preventive maintenance, partial daily maintenance, partial demolition and reconstruction, partial reinforcement and partial repair.

[0025] The present invention also provides a highway bridge maintenance auxiliary decision-making system, which applies the above-mentioned method and includes:

[0026] Data preprocessing module, used to clean training data and check the integrity of input data;

[0027] The decision tree construction module is used to construct a decision tree model for bridge maintenance measures through the CART algorithm based on the data after data preprocessing;

[0028] The decision result output module is used to output the decision results of bridge maintenance measures using the constructed decision tree model.

[0029] Further: the data preprocessing module includes:

[0030] The training data cleaning unit is used to check the key data fields in the decision records and remove the records with missing key data;

[0031] The input data checking unit is used to check the key data fields of the bridges to be decided, exclude bridges with serious data missing, and fill in the missing data.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] First, by preprocessing the historical maintenance data of bridges, screening and cleaning the key feature fields, eliminating invalid data with missing key information, and checking the integrity of the input bridge data to be decided and filling in missing values, the quality of training samples and application data is guaranteed, providing a reliable data basis for subsequent decision analysis.

[0034] Second, the CART classification and regression tree algorithm is used to recursively select the optimal splitting features, continuously subdivide the tree nodes until the stopping condition is reached, and build a decision tree with the goal of minimizing the Gini impurity. This algorithm can automatically extract implicit decision rules from complex bridge condition feature data, objectively reveal the corresponding relationship between different maintenance measures and bridge characteristics, and reduce the subjective bias of human experience judgment.

[0035] 3. The constructed decision tree model can automatically recommend the most appropriate maintenance measures based on key features such as the technical condition level of the bridge, inspection scores of various parts, and the time since the last maintenance. It covers various types of measures such as preventive maintenance of the entire bridge / local parts, daily maintenance, repair, reinforcement, demolition and reconstruction, making maintenance decisions more detailed and comprehensive.

[0036] Fourth, it provides a decision tree visualization function, which presents the splitting conditions, sample distribution, Gini impurity and other information of each node through an intuitive tree diagram, making the decision process transparent. Maintenance managers can clearly review the decision basis and conduct manual review and adjustment when necessary, which improves the interpretability and credibility of maintenance decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A flowchart of a method for constructing a decision tree for highway bridge maintenance measures in one implementation method;

[0038] Figure 2 A schematic diagram of an example structure of a decision tree;

[0039] Figure 3 It is a structural diagram of a highway bridge maintenance auxiliary decision-making system; DETAILED DESCRIPTION

[0040] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. 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.

[0041] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0042] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0043] like Figure 1 As shown, one embodiment of the present invention provides a method for constructing a highway bridge maintenance measure decision tree, comprising the following steps:

[0044] Step S1, data preprocessing, which includes two sub-steps: cleaning the training data and checking the integrity of the input data.

[0045] Among them, training data cleaning is to scan the key fields in the historical maintenance decision records, including the technical condition scores of various parts of the bridge, the technical condition level of the bridge, the type of maintenance measures, etc., and eliminate those records that are missing key information to form a complete and available training sample set. Input data integrity check refers to reviewing the bridge data currently waiting for decision-making to determine whether it has the complete information required for decision-making. If it is found that several fields are completely missing, indicating that the information is insufficient to support the decision, the bridge data is marked as requiring manual judgment: if only individual fields are missing, they can be filled with reasonable values, such as replacing the missing numerical field with the median of the field of the same type of bridge, and replacing the missing categorical field with the mode. Through these two preprocessing links, unusable dirty data is screened out to ensure the quality of data entering the decision analysis.

[0046] Step S2: construct a decision tree model using the CART algorithm based on the processed training data.

[0047] Among them, CART, or classification and regression tree, recursively selects the best splitting feature to split the sample set into different child nodes. Each split makes the child node more pure, that is, it contains as many samples of the same type as possible. The quality of its split is measured by calculating the Gini impurity. The smaller the Gini impurity, the smaller the internal difference of the child node after the split, and the better the classification effect. The calculation formula of this indicator is:

[0048]

[0049] Where k represents the total number of categories of bridge maintenance measures;

[0050] p i Represents the proportion of the i-th category in the data set.

[0051] In some other embodiments, the process of recursive growth of the decision tree is as follows:

[0052] S21: Evaluate various splitting schemes for all features under the current node, calculate the Gini impurity of the child nodes after the split, and select the feature with the smallest Gini impurity as the optimal splitting feature;

[0053] S22: Use the optimal split feature to divide the data of the current node into two child nodes, so that the samples of the same child node have the same or similar values ​​on the feature, and generate child nodes;

[0054] S23: Determine whether the preset stop condition is met. If so, stop growing. Otherwise, take the child node as the new current node and recursively execute S21 and S22 to generate a decision tree.

[0055] The stopping conditions here can be: the impurity of the node is less than a certain threshold, the depth of the node reaches a predetermined maximum depth, the number of samples contained in the node is less than or equal to a certain threshold, etc. If any of the conditions are met, the current node is considered to be sufficiently "pure" and does not need to be further split, so it is marked as a leaf node.

[0056] Step S3, using the constructed decision tree model, classify and predict the input bridge data to be classified, and output the maintenance measure decision result.

[0057] Specifically, starting from the root node, the feature values ​​of the samples to be classified are matched with the decision tree nodes one by one. According to the splitting conditions of each node, the samples are divided into corresponding child nodes and finally fall into a leaf node. The category to which the leaf node belongs is the maintenance measure of the model decision.

[0058] This process can be used Figure 2 The decision tree diagram shown is used to visually represent it. Each node in the diagram is represented by a polygon, and the following contents are noted in the node:

[0059] Judgment conditions: conditions used to determine whether the threshold of a specific feature is met, including superstructure score, substructure score, time since last maintenance, bridge deck score, and maintenance inspection level;

[0060] Gini coefficient: The Gini coefficient is used to measure the impurity of a node, ranging from 0 to 1. The smaller the value, the higher the purity of the node, that is, the samples in the node tend to belong to the same category. The Gini coefficient of the root node is 0.789, indicating that the sample categories in the node are relatively mixed;

[0061] Number of samples: indicates the number of samples contained in the node, where the root node contains 16 samples;

[0062] Sample distribution: indicates the number of samples in each measure category in this node;

[0063] Measure category: The measure category represents the category with the most samples in the node, which serves as the measure category of the node.

[0064] When the decision tree node no longer splits, a leaf node is generated. In the leaf node, the decision tree takes the category with the largest number of samples in the node as the prediction result and assigns it to the node. When the Gini coefficient of a leaf node = 0, it means that the sample purity in the node is very high, all samples belong to the same category, and the measure category of the node is the recommended maintenance measure.

[0065] Figure 2 The process of generating a decision tree can be divided into the following steps:

[0066] The first is the optimal feature selection. At each split, the Gini coefficient under different features and their threshold combinations is evaluated to select the feature-threshold combination that can minimize the node impurity. The second is data partitioning. Based on the selected features and thresholds, the data is divided into two subsets, which constitute the left (lower) and right (upper) child nodes respectively. Each internal node represents a binary decision point.

[0067] Next is the recursive construction process, repeating the feature selection and data partitioning process for each child node. The termination conditions include reaching the preset maximum depth, the number of node samples is lower than the minimum threshold, and the direct termination of the split when the Gini coefficient is 0 in special cases. Finally, the leaf node is determined. When the node meets the termination condition, the leaf node is formed, and the measure category with the most samples is used as the recommended maintenance strategy for the node. The leaf node is used as the terminal node of the decision tree for classification prediction of new samples.

[0068] Through this progressive decision-making process, a decision tree model that can effectively classify and predict maintenance measures is finally constructed. This model not only has strong interpretability, but also can intuitively display decision rules, which is convenient for practical application and result analysis.

[0069] It should be noted that Figure 2 The decision tree in Figure 1 The data samples shown in the figure are more complex than those in actual applications. We can prune the trees appropriately, limit the maximum depth of the tree, the minimum number of samples in a leaf node, and the minimum reduction in the Gini index for splitting, etc., to avoid overfitting and improve decision efficiency.

[0070] In addition, as new bridge data is generated, we can also regularly use the accumulated samples to retrain the decision tree so that it can adapt to changes in data distribution. At the same time, maintenance managers can modify the suggestions given by the decision tree as appropriate based on actual conditions. This human-machine combination can give full play to the advantages of expert experience and machine calculations, and continuously improve the scientificity and accuracy of maintenance decisions.

[0071] Figure 3 Another embodiment of the present invention is given, which discloses the structural composition of the decision support system for completing the above method. The system mainly includes three functional modules: data preprocessing module, decision tree construction module and decision result output module. Each module is linked through a data interface to form a complete data processing and decision-making process.

[0072] The data preprocessing module includes a training data cleaning unit and an input data checking unit, which are responsible for extracting historical maintenance records from the bridge maintenance management database, cleaning them, removing invalid data with missing key information, and generating a training sample set. At the same time, the module also reviews the input bridge data to be decided, determines its completeness, and fills in missing fields when necessary to ensure that it is suitable for decision analysis.

[0073] The decision tree construction module receives the preprocessed training data, calls the CART algorithm, and generates a decision tree through recursive splitting. During this process, the module calculates the Gini index of each candidate split feature and selects the optimal splitting scheme until a complete decision tree is generated. The module outputs a data structure that represents the decision tree, including the topological structure of the tree, the splitting basis of each node, and the category distribution information of the leaf nodes.

[0074] The decision result output module receives the decision tree generated by the decision tree construction module on the one hand, and obtains the characteristic parameters of the bridge to be decided input by the user or collected by the system on the other hand. Then, it classifies the bridge to be decided into a leaf node according to the node rules of the decision tree, and outputs the category of the leaf node as the decision result. At the same time, in order to enhance the interpretability of the decision process, the module can also visualize the decision tree to provide intuitive decision-making basis for maintenance managers.

[0075] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Those skilled in the art can adjust the division and implementation of each functional module according to the needs of practical applications to implement the present invention in the best possible way. For example, when data is insufficient or the decision tree structure is too complex, integrated learning can be introduced to construct combination models such as random forests to improve the robustness of decision-making. For another example, the types of measures for bridge maintenance can be further refined to provide targeted decision support for the maintenance of local components. Any modifications, equivalent substitutions and improvements made to this solution based on the technical essence of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for assisting decision making in highway bridge maintenance, characterized in that: The following steps are involved: S1: Perform data preprocessing, including cleaning training data and checking the integrity of input data; S2: Based on the data preprocessed in S1, a decision tree model for bridge maintenance measures is constructed using the CART algorithm; S3: Use the decision tree model constructed in S2 to output the decision results of bridge maintenance measures.

2. The method for auxiliary decision-making for highway bridge maintenance according to claim 1 is characterized in that: The specific steps of step S2 include: S21: Select the feature with the smallest Gini impurity as the split feature of the current node; S22: Split the data set of the current node according to the selected split feature to generate child nodes; S23: recursively execute S21 and S22 until a preset stop condition is reached to generate a decision tree.

3. The method for auxiliary decision-making for highway bridge maintenance according to claim 1 is characterized in that: The data preprocessing in step S1 specifically includes: The key fields in the maintenance decision records, such as superstructure score, substructure score, bridge deck score, bridge technical condition grade, maintenance inspection grade, time since last maintenance, and maintenance decision results, were scanned, and records with missing key fields were removed to form a training data set.

4. The method for auxiliary decision-making for highway bridge maintenance according to claim 1, characterized in that: The data preprocessing in step S1 also includes: Check the integrity of the bridge data to be predicted. If the key fields are completely missing, mark it as requiring manual maintenance decision-making. If only individual fields are missing, fill the missing numerical fields with the median of the field in all the data, and fill the missing categorical fields with the mode of the field in all the data.

5. The method for auxiliary decision-making for highway bridge maintenance according to claim 2 is characterized in that: The stopping condition in step S23 is: at least one of the following: the Gini impurity of the node is less than a preset threshold, the depth of the node reaches a preset maximum depth, and the number of samples of the node is less than a preset threshold.

6. The method for auxiliary decision making for highway bridge maintenance according to claim 2, characterized in that: The calculation formula of Gini impurity is: Where k represents the total number of categories of bridge maintenance measures; p i Represents the proportion of the i-th category in the data set.

7. The method for auxiliary decision making for highway bridge maintenance according to claim 1, characterized in that: In step S3, the decision tree is displayed in a visual manner, where each node represents a classification condition of bridge characteristic data, including the Gini impurity under the condition, the number of samples, the number of samples contained in each type of bridge characteristic data, and the recommended classification of bridge maintenance measures.

8. A method for auxiliary decision making for highway bridge maintenance according to any one of claims 1 to 7, characterized in that: The bridge maintenance measures include preventive maintenance of the entire bridge, daily maintenance of the entire bridge, demolition and reconstruction of the entire bridge, reinforcement of the entire bridge, repair of the entire bridge, partial preventive maintenance, partial daily maintenance, partial demolition and reconstruction, partial reinforcement and partial repair.

9. A highway bridge maintenance auxiliary decision-making system, characterized in that: The method according to any one of claims 1 to 7 comprises: Data preprocessing module, used to clean training data and check the integrity of input data; The decision tree construction module is used to construct a decision tree model for bridge maintenance measures through the CART algorithm based on the data after data preprocessing; The decision result output module is used to output the decision results of bridge maintenance measures using the constructed decision tree model.

10. The highway bridge maintenance auxiliary decision-making system according to claim 9, characterized in that: The data preprocessing module comprises: The training data cleaning unit is used to check the key data fields in the decision records and remove the records with missing key data; The input data checking unit is used to check the key data fields of the bridges to be decided, exclude bridges with serious data missing, and fill in the missing data.

Images