Claim settlement event detection device, method and equipment of engineering machinery and storage medium

Abstract

The invention provides a claim settlement event detection device, method and equipment for engineering machinery and a storage medium, and relates to the technical field of artificial intelligence, and the device comprises a data obtaining module which is used for obtaining multi-source data of the engineering machinery; the data processing module is used for performing data processing on the multi-source data to obtain a feature data set; the model training module is used for training a convolutional neural network model by using the feature data set to obtain an initial claim event prediction model; the federated learning module is used for constructing a federated learning framework to update gradient parameters of the initial claim settlement event prediction model to obtain a target claim settlement event prediction model; the model detection module is used for predicting an abnormal probability value of a target claim event; the data analysis module is used for comparing, analyzing and processing data to obtain the number of abnormal items; the event detection module is used for determining a claim event detection result according to the target claim event abnormal probability value and the abnormal item number. According to the invention, the authenticity identification accuracy of the claim settlement event of the engineering machinery can be improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a claim event detection device, method, equipment and storage medium for engineering machinery. Background Art

[0002] In the context of the rapid development of science and technology, artificial intelligence technology is experiencing a new stage of cross-domain and interactive development, which has brought profound opportunities for change and innovation to my country's traditional insurance industry. In this context, the intelligent cloud computing method of insurance claims based on artificial intelligence came into being, and has shown significant application and integration results in the field of Internet insurance. By using advanced technical means such as voice recognition, intelligent analysis, and face recognition, this method greatly reduces the reliance on manual operations during the life cycle of the policy, thereby effectively achieving the dual goals of cost control and improved operational efficiency. However, in the field of traditional engineering machinery insurance claims, there are still a series of technical problems that need to be solved, which seriously restrict the further development of the industry.

[0003] There are many types of construction machinery, and there are significant differences in the operating scenarios, technical parameters and vulnerable parts of different models. In addition, construction machinery accidents often occur in harsh environments such as remote construction sites and mines, where transportation is inconvenient and there are safety hazards, making it difficult for surveyors to quickly arrive at the scene, which in turn causes evidence to be easily lost or equipment to be exposed for a long time, exacerbating losses. Due to the limitations of manual review, it is difficult to quickly and accurately identify these abnormal claims behaviors, resulting in insurance companies facing greater claims risks and losses. Therefore, how to improve the accuracy of identifying the authenticity of construction machinery insurance claims has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the present invention provides a claim event detection device, method, equipment and storage medium for engineering machinery, which effectively solve the problem of low accuracy in identifying the authenticity of engineering machinery insurance claim events.

[0005] In a first aspect, the present invention provides a claim settlement event detection device for construction machinery, the device comprising: A data acquisition module, used to acquire multi-source data of engineering machinery, wherein the multi-source data at least includes engineering machinery operation data, engineering machinery maintenance data and historical claims event data; A data processing module, used for performing data cleaning, data standardization and feature extraction on the multi-source data to obtain a feature data set; A model training module, used to construct a convolutional neural network model, train the convolutional neural network model using the feature data set, and obtain an initial claims event prediction model; A federated learning module, configured to construct a federated learning framework according to the initial claim event prediction model, and update the gradient parameters of the initial claim event prediction model based on the federated learning framework to obtain a target claim event prediction model; A model detection module, configured to obtain target claim event data of a target construction machinery, perform data cleaning, data standardization, and feature extraction on the target claim event data to obtain target feature data, and input the target feature data into the target claim event prediction model for prediction to obtain an abnormal probability value of a target claim event of the target construction machinery, where the target claim event data at least includes target construction machinery operation data and target construction machinery maintenance data for which the target construction machinery applies for a claim; A data analysis module, configured to perform data comparison and analysis processing according to the target claim event data and real-time multi-source data of the target construction machinery to obtain the number of abnormal items for which the data comparison and analysis result is abnormal; An event detection module, configured to determine that the claim event detection result is an abnormal claim event if the abnormal probability value of the target claim event is greater than a preset probability threshold and the number of abnormal items is greater than a preset number of items.

[0006] In an alternative embodiment, the data acquisition module includes: A first data acquisition unit, communicatively connected to the sensor unit of the construction machinery, where the first data acquisition unit is configured to acquire the construction machinery operation data collected by the sensor unit; A second data acquisition unit, communicatively connected to a first server, where the first server is configured to store the construction machinery maintenance data of the construction machinery, and the second data acquisition unit is configured to acquire the construction machinery maintenance data stored by the first server; A third data acquisition unit, communicatively connected to a second server, where the second server is configured to store the historical claim event data, and the third data acquisition unit is configured to acquire the historical claim event data, where the historical claim event data at least includes construction machinery historical operation data, construction machinery historical maintenance data, historical claim results, and historical claim amounts.

[0007] In an alternative embodiment, the data processing module includes: A data cleaning unit, configured to perform preliminary cleaning on the multi-source data by using a spatial clustering algorithm to obtain the preliminarily cleaned multi-source data; A data comparison unit, configured to perform comparison and statistics on the preliminarily cleaned multi-source data, remove abnormal data, and obtain an initial data set; A numerical processing unit, configured to perform standardization processing on the numerical data in the initial data set by using a numerical standardization method to obtain an initial standard data set; An encoding processing unit, which is used to encode the categorical data in the initial standard dataset by using one-hot encoding technology to obtain a standard dataset; A first feature extraction unit, which is used to extract feature data from the construction machinery operation data in the standard dataset by using a sliding window algorithm to obtain an initial feature dataset; A second feature extraction unit, which is used to extract association rule features from the construction machinery maintenance data in the initial feature dataset by using an association rule mining algorithm to obtain the feature dataset.

[0008] In an optional implementation manner, the convolutional neural network model at least includes an input layer, a convolutional unit, a pooling unit, a fully connected unit, and an output layer, where: The input layer is used to splice the numerical feature data and the categorical feature data in the feature dimension to obtain a feature tensor; The convolutional unit includes multiple convolutional layers, and the multiple convolutional layers include convolutional kernels of different sizes. The multiple convolutional layers gradually extract features from the feature tensor in a stacked manner to obtain a feature vector; The pooling unit includes a max pooling layer, and the max pooling layer is used to perform feature dimensionality reduction on the feature vector output by the convolutional layer to obtain a dimensionality-reduced feature vector; The fully connected unit includes multiple fully connected layers, and the multiple fully connected layers are used to perform a linear transformation on the dimensionality-reduced feature vector to obtain a comprehensive feature vector; The output layer is used to convert the comprehensive feature vector into a prediction probability value through an activation function.

[0009] In an optional implementation manner, the model training module includes: A data partitioning unit, which is used to partition the feature dataset into a training set, a validation set, and a test set; A model training unit, which is used to train the convolutional neural network model according to the training set, and adjusts the parameters of the convolutional neural network model by using an adaptive moment estimation optimization algorithm during the training process to obtain a pre-trained convolutional neural network model; A model validation unit, which is used to perform multiple rounds of validation on the pre-trained convolutional neural network model by using K-fold cross-validation according to the validation set to obtain an initial convolutional neural network model; A model testing unit, which is used to test the initial convolutional neural network model according to the test set to obtain the initial claim event prediction model.

[0010] In an optional implementation manner, the federated learning module includes: A central control unit for distributing the initial claim event prediction model and the initial model parameters to the clients of each regional node through a central server; A regional learning unit for obtaining the local multi-source data of each regional node, training the initial claim event prediction model according to the local multi-source data and the initial model parameters, and obtaining model parameter gradient information; An aggregation calculation unit for obtaining the model parameter gradient information of each regional node, performing aggregation calculation on the model parameter gradient information, and obtaining aggregated gradient information; A model update unit for updating the global model parameters of the initial claim event prediction model according to the aggregated gradient information to obtain the target claim event prediction model.

[0011] In an optional implementation manner, the global model parameters at least include an input weight matrix, a hidden layer weight matrix, an output weight matrix, and a bias term, and the model update unit is used to update the input weight matrix, the hidden layer weight matrix, the output weight matrix, and the bias term according to the aggregated gradient information by using a stochastic gradient descent algorithm.

[0012] In a second aspect, the present invention provides a method for detecting claim events of construction machinery, and the method is applied to the claim event detection device of construction machinery in the first aspect of the present invention. The method includes: Obtaining multi-source data of construction machinery, where the multi-source data at least includes construction machinery operation data, construction machinery maintenance data, and historical claim event data; Performing data cleaning, data standardization, and feature extraction on the multi-source data to obtain a feature data set; Constructing a convolutional neural network model, training the convolutional neural network model by using the feature data set to obtain an initial claim event prediction model; Constructing a federated learning framework according to the initial claim event prediction model, and performing gradient parameter update on the initial claim event prediction model based on the federated learning framework to obtain a target claim event prediction model; Obtaining target claim event data of target construction machinery, performing data cleaning, data standardization, and feature extraction on the target claim event data to obtain target feature data, inputting the target feature data into the target claim event prediction model for prediction to obtain a target claim event abnormal probability value of the target construction machinery, where the target claim event data at least includes target construction machinery operation data and target construction machinery maintenance data for which the target construction machinery applies for a claim; Performing data comparison and analysis processing based on the target claim event data and the real-time multi-source data of the target construction machinery, and obtaining the number of abnormal items with abnormal data comparison and analysis results; If the abnormal probability value of the target claim event is greater than a preset probability threshold and the number of abnormal items is greater than a preset number of items, then determine that the claim event detection result is an abnormal claim event.

[0013] In a third aspect, the present invention provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the computer program to implement the claim event detection method for construction machinery as described in the second aspect of the present invention.

[0014] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the claim event detection method for construction machinery as described in the second aspect of the present invention.

[0015] The claim event detection device, method, equipment, and storage medium for construction machinery provided by the present invention perform data preprocessing by obtaining multi-dimensional construction machinery operation data, construction machinery maintenance data, and historical claim event data, construct a convolutional neural network model, and train the convolutional neural network model according to the data set after data preprocessing to obtain a target claim event prediction model, which can comprehensively and accurately identify claim abnormal behaviors of various construction machinery and improve the accuracy of identifying the authenticity of claim events. By constructing a federated learning framework, the target claim event prediction model can continuously learn new claim event data, timely discover new types of claim abnormal behavior patterns and characteristics, and as the data continues to accumulate and the model continues to be optimized, the device's ability to identify claim abnormal behaviors will continue to increase, effectively coping with the increasingly complex and changeable abnormal claim risks. The device realizes automatic data collection, analysis, and event detection, greatly shortening the claim review cycle, reducing the manual review workload, reducing the investment in labor costs, and more importantly, by accurately identifying claim abnormal behaviors, avoiding unnecessary claim expenditures, saving a large amount of funds for insurance companies, and improving the overall operating efficiency of the company. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the embodiments. It should be understood that the following drawings only show some embodiments of the present invention, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0017] Figure 1 is the first schematic diagram of the structure of the claim event detection device for construction machinery provided by the embodiment of the present invention; Figure 2 It is the second schematic diagram of the structure of the claim event detection device for construction machinery provided by the embodiment of the present invention; Figure 3 It is the third schematic diagram of the structure of the claim event detection device for construction machinery provided by the embodiment of the present invention; Figure 4 It is the schematic diagram of the structure of the convolutional neural network model in the embodiment of the present invention; Figure 5 It is the fourth schematic diagram of the structure of the claim event detection device for construction machinery provided by the embodiment of the present invention; Figure 6 It is the fifth schematic diagram of the structure of the claim event detection device for construction machinery provided by the embodiment of the present invention; Figure 7 It is the schematic diagram of the process flow of the claim event detection method for construction machinery provided by the embodiment of the present invention; Figure 8 It is the schematic diagram of the structure of an electronic device provided by the embodiment of the present invention.

[0018] Main element symbol description: 100, claim event detection device for construction machinery; 110, data acquisition module; 111, first data acquisition unit; 112, second data acquisition unit; 113, third data acquisition unit; 120, data processing module; 121, data cleaning unit; 122, data comparison unit; 123, numerical processing unit; 124, encoding processing unit; 125, first feature extraction unit; 126, second feature extraction unit; 130, model training module; 131, data division unit; 132, model training unit; 133, model verification unit; 134, model testing unit; 140, federated learning module; 141, central control unit; 142, regional learning unit; 143, aggregation calculation unit; 144, model update unit; 150, model detection module; 160, data analysis module; 170, event detection module; 200, sensor unit; 300, first server; 400, second server; 800, electronic device; 810, processor; 820, communication interface; 830, memory; 840, communication bus. Detailed implementation manners

[0019] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be further described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention. It should be noted that the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0020] In the field of traditional engineering machinery insurance claims settlement, there are a series of technical problems that urgently need to be solved, which seriously restrict the further development of the industry. There are a wide variety of engineering machinery, and there are significant differences in the operation scenarios, technical parameters, and vulnerable parts of different models. In addition, engineering machinery accidents mostly occur in remote construction sites, mines and other harsh environments, where transportation is inconvenient and there are safety hazards, making it difficult for surveyors to quickly reach the scene, resulting in the easy loss of evidence or the long-term exposure of equipment, which exacerbates the loss. Due to the limitations of manual review, it is difficult to quickly and accurately identify these abnormal claims settlement behaviors, resulting in the insurance company facing greater claims settlement risks and losses. Therefore, how to improve the accuracy of identifying the authenticity of engineering machinery insurance claims settlement events has become a technical problem that urgently needs to be solved.

[0021] Embodiment 1 The embodiment of the present invention provides a detection device for claims settlement events of engineering machinery, which effectively solves the problem of low accuracy in identifying the authenticity of engineering machinery insurance claims settlement events. Figure 1 It is the first schematic diagram of the structure of the detection device for claims settlement events of engineering machinery provided by the embodiment of the present invention. As Figure 1 shown, the detection device 100 for claims settlement events of engineering machinery includes a data acquisition module 110, a data processing module 120, a model training module 130, a federated learning module 140, a model detection module 150, a data analysis module 160, and an event detection module 170.

[0022] The data acquisition module 110 is used to acquire multi-source data of engineering machinery, and the multi-source data includes but is not limited to engineering machinery operation data, engineering machinery maintenance data, and historical claims settlement event data. Figure 2 It is the second schematic diagram of the structure of the detection device for claims settlement events of engineering machinery provided by the embodiment of the present invention. As Figure 2 shown, the data acquisition module 110 specifically includes a first data acquisition unit 111, a second data acquisition unit 112, and a third data acquisition unit 113, where: The first data acquisition unit 111 is connected to the sensor unit 200 of the engineering machinery in communication, and is used to acquire the engineering machinery operation data collected by the sensor unit 200. In an embodiment of the present invention, the sensor unit 200 includes, but is not limited to, high-precision sensors such as a GPS sensor with a real-time positioning function, a fuel consumption sensor that accurately measures fuel consumption, and an engine speed sensor that accurately monitors the engine speed. By installing these high-precision sensors at key parts of the engineering machinery, the engineering machinery operation data such as the working time, working position, fuel consumption and engine speed of the engineering machinery can be collected in real time. For example, on a certain excavator, the longitude and latitude position data fed back by the GPS sensor in real time is [116.38, 39.90], the fuel consumption measured by the fuel consumption sensor at a certain moment is 5.2 liters / hour, and the engine speed sensor monitors the engine speed to be 1800 rpm. The collected engineering machinery operation data is quickly transmitted to the first data acquisition unit 111 via 4G or 5G wireless communication.

[0023] The second data acquisition unit 112 is connected to the first server 300 for communication. The first server 300 is used to store the engineering machinery maintenance data of the engineering machinery. The second data acquisition unit 112 is used to acquire the engineering machinery maintenance data stored by the first server 300. In an embodiment of the present invention, the first server 300 may be a data management server of the maintenance management system of the engineering machinery enterprise. The second data acquisition unit 112 establishes a secure and stable data communication connection with the first server 300 through the data interface technology, and automatically extracts the engineering machinery maintenance data such as maintenance time, maintenance items, maintenance costs, and maintenance plant information at a preset time interval. Exemplarily, the engineering machinery maintenance data extracted at a certain time is as follows: the maintenance time is "2025-03-05 10:30:00", the maintenance item is "replacement of hydraulic pump", the maintenance cost is 5,000 yuan, and the name of the maintenance plant is "XX Engineering Machinery Maintenance Plant".

[0024] The third data acquisition unit 113 is communicatively connected to the second server 400, and the second server 400 is used to store historical claims event data. The third data acquisition unit 113 is used to acquire historical claims event data, and the historical claims event data includes but is not limited to historical operation data of engineering machinery, historical maintenance data of engineering machinery, historical claims results, and historical claims amounts. These historical claims event data have been labeled and are the cornerstone of subsequent training of convolutional neural network models. At the same time, when the subsequent target claims event prediction model performs multi-dimensional data analysis, the historical claims event data can serve as an important reference. In an embodiment of the present invention, the second server 400 can be a server equipped with an insurance company's claims database, and the historical claims event data in the claims database is synchronized to the third data acquisition unit 113 through a professional data synchronization tool.

[0025] Optionally, the data synchronization tool may be an ETL (Extract Transform Load) tool. When using the ETL tool for data synchronization, data transformation and cleaning tasks can be set according to pre-established data mapping rules. At the same time, after the data synchronization is completed, a data comparison algorithm is used to perform data consistency checks to ensure that the historical claim event data synchronized to the third data acquisition unit 113 is consistent with the data in the insurance company's claim database. Exemplarily, the synchronized historical claim event data is as follows: the reason for the claim application is "collision accident", the claim application time is "2024-12-15", the claim amount is 8,000 yuan, the claim result is "paid", and the historical operation data and historical maintenance data of the construction machinery at the time of the collision accident.

[0026] The data processing module 120 is used to perform data cleaning, data standardization, and feature extraction on multi-source data to obtain a feature data set. Figure 3 This is the third schematic diagram of the structure of the claim event detection device for construction machinery provided by the embodiments of the present invention. As Figure 3 shown, the data processing module 120 includes a data cleaning unit 121, a data comparison unit 122, a numerical processing unit 123, a coding processing unit 124, a first feature extraction unit 125, and a second feature extraction unit 126, where: The data cleaning unit 121 comprehensively cleans the collected multi-source data using a spatial clustering algorithm to obtain the preliminarily cleaned multi-source data. In the embodiments of the present invention, the spatial clustering algorithm uses the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm. This algorithm determines clusters based on the tightness of the sample distribution, can discover clusters of any shape, and can effectively identify noise points.

[0027] Exemplarily, for data cleaning of the fuel consumption value in the construction machinery operation data, first collect multi-dimensional data containing the fuel consumption value and related information, integrate and select the relevant feature data of the fuel consumption value as the algorithm input, standardize the data and process missing values to make the relevant feature data more suitable for algorithm processing. Then set the neighborhood radius and minimum number of points parameters, identify the core points, generate clusters based on the core points, and identify the points that do not belong to any cluster as noise points. Then set the normal fuel consumption threshold according to historical data and data of similar equipment, determine the fuel consumption data exceeding the threshold in the noise points as abnormal data, and correct the abnormal data according to the data characteristics. Those that cannot be corrected are directly deleted. Finally, select indicators such as accuracy and recall to evaluate the cleaning effect, and verify the results through methods such as comparison with known correct data or cross-validation. If the cleaning effect is not ideal, adjust the parameters or improve the preprocessing steps and then re-clean and evaluate.

[0028] The data comparison unit 122 is used to compare and count the multi-source data after preliminary cleaning, eliminate abnormal data, and obtain an initial data set. In order to further ensure the accuracy of the data, a statistical method is used to perform in-depth comparison on the multi-source data after preliminary cleaning to accurately determine whether it is abnormal data, and further correct and delete the abnormal data to obtain an initial data set.

[0029] For example, by making an in-depth comparison with the historical fuel consumption data of the target engineering machinery and the fuel consumption data of the same type of engineering machinery, first collect the detailed historical fuel consumption data of the target engineering machinery, covering information such as different working conditions and time periods, and collect the fuel consumption data of multiple engineering machinery of the same type under similar working conditions for integration to ensure the consistency and integrity of the data. Then determine the key characteristic data related to fuel consumption, such as equipment operating speed, load weight and working hours, and standardize or normalize these key characteristic data to eliminate the dimensional effect for subsequent comparative analysis. Then group the key characteristic data according to factors such as the working mode, working environment and service life of the engineering machinery, or classify them according to time periods such as days, weeks and months, and compare them in more comparable subsets. Finally, we compare the data from multiple dimensions such as mean, distribution and trend, calculate the average fuel consumption of the target engineering machinery and the same type of engineering machinery under different groups and compare the differences. We analyze the distribution of the fuel consumption data of the two by drawing charts, observe the trend of the target engineering machinery's fuel consumption over time or working conditions, and compare it with the average trend of the same type of engineering machinery to find abnormal fluctuations. If it is determined to be abnormal, it will be corrected or deleted directly according to the data characteristics. For example, the normal fuel consumption range of an excavator is 3-7 liters / hour after statistical analysis. If the fuel consumption data collected at a certain time is 10 liters / hour, which exceeds the normal fuel consumption range, it will be marked as abnormal data, and can be further corrected based on historical data or data of the same type. If it cannot be corrected, it will be deleted directly.

[0030] The numerical processing unit 123 is used to standardize the numerical data in the initial data set using a numerical standardization method to obtain an initial standard data set. In an embodiment of the present invention, numerical standardization methods such as Z-score standardization, deviation standardization, and decimal calibration standardization can be used to process the numerical data, and the numerical data such as the maintenance cost and the claim amount are converted into a range of 0~1 to obtain numerical feature data. For example, assuming that the minimum value of the maintenance cost is 1,000 yuan and the maximum value is 10,000 yuan, when the maintenance cost of a certain time is 3,000 yuan, the standard value after deviation standardization is (3000-1000) / (10000-1000)=0.222.

[0031] The encoding processing unit 124 is used to encode the categorical data in the initial standard dataset by using the one-hot encoding technique to obtain the standard dataset. In the embodiments of the present invention, for categorical data such as repair shop names and claim reasons in the initial standard dataset, the one-hot encoding technique is used for encoding processing, and the categorical data is converted into a binary vector form that is easy for a computer to process, obtaining categorical feature data.

[0032] Exemplarily, assume that a simple repair claim data is as shown in Table 1 below, including two columns of categorical data: repair shop name and claim reason.

[0033] Table 1. Schematic table of repair claim data

[0034] Processing the above two columns of categorical data by using the one-hot encoding technique includes the following steps: First, determine all possible values of each categorical variable. The possible values of the repair shop name may be Factory A and Factory B, and the possible values of the claim reason may be collision and natural damage.

[0035] Then, create a new binary column for each possible value of each categorical variable. For the repair shop name, create two new binary columns: Repair shop name - Factory A and Repair shop name - Factory B. For the claim reason, create two new binary columns: Claim reason - collision and Claim reason - natural damage.

[0036] Finally, if the original value of a certain record's categorical variable is the same as the value corresponding to the new binary column, the value of this new binary column is 1, otherwise it is 0. The encoded result is as shown in Table 2 below.

[0037] Table 2. Schematic table of one-hot encoding result

[0038] Feature engineering is carried out on the standard dataset through a feature extraction algorithm, and the original data in the standard dataset is converted into a feature form suitable for input into a machine learning model.

[0039] The first feature extraction unit 125 is used to extract feature data from the construction machinery operation data in the standard dataset by using the sliding window algorithm to obtain the initial feature dataset. Exemplarily, the sliding window algorithm is used to extract feature data such as the continuous working duration of the construction machinery and the moving trajectory feature of the working position from the construction machinery operation data. First, define the size and sliding step of the sliding window, then calculate the sum of the working durations within each sliding window as the continuous working duration, calculate the Euclidean distance between the starting position and the ending position within each sliding window as the moving trajectory feature of the working position, and finally integrate the continuous working durations and the moving trajectory features of the working positions of each sliding window to obtain the feature data.

[0040] The second feature extraction unit 126 is configured to extract association rule features from the construction machinery maintenance data in the initial feature dataset by using an association rule mining algorithm, so as to obtain a feature dataset. In an embodiment of the present invention, the association rule mining algorithm may be the Apriori algorithm. For the construction machinery maintenance data in the initial feature dataset, the Apriori algorithm is used to extract the association rule features of the construction machinery maintenance data, and potential rules that certain maintenance items frequently appear simultaneously are obtained.

[0041] In an embodiment of the present invention, first, the target maintenance record is converted into a transaction dataset suitable for processing by the Apriori algorithm. Each transaction represents a maintenance record, which contains the items involved in the target maintenance. Then, all possible item sets, that is, combinations of different maintenance items, are generated. According to a preset minimum support threshold, frequent item sets, that is, item sets whose occurrence frequencies meet the requirements, are filtered out. Based on the frequent item sets, association rules are generated according to the minimum confidence threshold, so as to find out which maintenance items often appear simultaneously. Finally, corresponding numerical feature data or categorical feature data are obtained based on the corresponding maintenance items in the association rule features. Exemplarily, for a series of maintenance item records such as ["Replace filter element", "Replace engine oil"], ["Replace filter element", "Check circuit"], ["Replace engine oil", "Check tire"], and ["Replace filter element", "Replace engine oil", "Check circuit"], Apriori algorithm analysis finds that the two maintenance items "Replace filter element" and "Replace engine oil" often appear simultaneously, thus forming an association rule feature, and the categorical feature data corresponding to "Replace filter element" and "Replace engine oil" are associated to obtain corresponding feature data.

[0042] In an embodiment of the present invention, after feature extraction from the construction machinery operation data and the construction machinery maintenance data, the extracted various feature data can be organized into the form of a feature matrix. For example, for the continuous working duration and the moving trajectory features of the working position of the construction machinery and the association rule features of the maintenance items, each sample (such as each construction machinery or each time period, etc.) can be used as a row, and different features can be used as columns to form a two-dimensional feature matrix. Such a feature matrix can clearly display the feature structure of the data, which is convenient for subsequent machine learning algorithms to process and analyze.

[0043] Optionally, the association rule mining algorithm can also be used to deeply analyze the historical claim event data, so as to discover potential abnormal claim patterns and rules hidden behind the historical claim event data, which is beneficial to improving the accuracy of claim event detection.

[0044] The model training module 130 is used to build a convolutional neural network model, and train the convolutional neural network model using the feature dataset to obtain an initial claim event prediction model.

[0045] In the embodiment of the present invention, a mature and efficient deep learning framework, such as TensorFlow or PyTorch, is selected to build a convolutional neural network model. Figure 4 It is a schematic structural diagram of the convolutional neural network model in the embodiment of the present invention. As Figure 4 shown, the convolutional neural network model includes an input layer, a convolutional unit, a pooling unit, a fully connected unit, and an output layer, where: The input layer is used to splice the numerical feature data and the categorical feature data in the feature dimension to obtain a feature tensor. There are differences in the tensor form between the numerical feature data and the categorical feature data in the feature matrix. Numerical feature data are generally continuous real values. For example, repair costs and claim amounts exist in the tensor in the form of floating-point numbers. The numerical size has practical physical significance. For example, the higher the repair cost, the larger the possible scale of repair. During model training, the numerical changes of the numerical feature data will directly affect the calculation of the model. Categorical feature data represents different categories, such as repair shop names and claim reasons. After encoding, it is presented in binary form, with only one position being 1 and the rest being 0. The position where 1 is located represents the category to which it belongs. There is no comparison relationship of numerical size between categories, only a discrete representation. The input layer concatenates the numerical feature data and the categorical feature data after one-hot encoding in a certain dimension. Exemplarily, assuming there is a batch of data, and each sample has 3 numerical feature data and 5 categorical feature data (after one-hot encoding, it is a vector of length 5), then they can be concatenated in the feature dimension into a feature tensor of length 8.

[0046] The convolutional unit includes multiple convolutional layers. The multiple convolutional layers include convolutional kernels of different sizes. The multiple convolutional layers gradually extract features from the feature tensor in a stacked manner to obtain a feature vector. In an embodiment of the present invention, the convolutional unit includes two convolutional layers. The first convolutional layer and the second convolutional layer are arranged in sequence by stacking. The output of the first convolutional layer serves as the input of the second convolutional layer, gradually extracting the key features in the feature tensor to obtain a feature vector. In the first convolutional layer, 16 3×3 convolutional kernels and 16 5×5 convolutional kernels are respectively used. The smaller 3×3 convolutional kernels can capture the local detailed features in the feature data, while the larger 5×5 convolutional kernels can extract more extensive context features. In the second convolutional layer, 32 3×3 convolutional kernels and 32 5×5 convolutional kernels are respectively used, so as to better extract features of different scales to adapt to the complex and diverse characteristics of the multi-source data of construction machinery. Optionally, the number of convolutional layers can be increased according to the complexity of the multi-source data. As the convolutional layer deepens, the degree of abstraction of the features gradually increases. It is also possible to select ways such as skip connection stacking and multi-branch stacking to stack the convolutional layers to adapt to different situations of the multi-source data.

[0047] The pooling unit includes a max pooling layer. The max pooling layer is used to perform feature dimensionality reduction on the feature vector output by the convolutional layer to obtain a dimensionality-reduced feature vector. The pooling unit receives the feature vector output by the convolutional unit, performs a downsampling operation on it, and outputs the dimensionality-reduced feature vector with reduced dimensions to the subsequent fully connected unit. In an embodiment of the present invention, a 2×2 max pooling layer is used to reduce the data dimension of the feature vector without losing key information, reduce the computational amount and the number of parameters of the model, and prevent overfitting. Specifically, the max pooling operation selects the maximum value in each pooling window as the output. In a 2×2 pooling window, the maximum value among the four elements is taken as the output value of the window. Optionally, the number of layers of the max pooling layer can be determined comprehensively according to specific task requirements, the characteristics of the multi-source data, and the overall architecture of the convolutional neural network model.

[0048] The fully connected unit includes multiple fully connected layers. The multiple fully connected layers are used to perform a linear transformation on the dimensionality-reduced feature vector to obtain a comprehensive feature vector. In an embodiment of the present invention, the fully connected unit flattens the dimensionality-reduced feature vector output by the pooling unit. The fully connected unit includes a first fully connected layer and a second fully connected layer. The first fully connected layer includes 128 neurons, and the second fully connected layer includes 256 neurons. All neurons of the first fully connected layer are fully connected to all neurons of the second fully connected layer. The fully connected layer comprehensively processes the previously extracted dimensionality-reduced feature vector, performs a linear transformation on the dimensionality-reduced feature vector through a weight matrix and a bias term to obtain a comprehensive feature vector, and maps the comprehensive feature vector to the final output layer. Optionally, the number of layers of the fully connected layer and the number of neurons can be set according to the complexity of the model and task requirements.

[0049] The output layer is used to convert the comprehensive feature vector into a predicted probability value through an activation function. In the embodiment of the present invention, the output layer uses a neuron and the Softmax function to convert the comprehensive feature vector output by the fully connected unit into a predicted probability value, and this predicted probability value represents the probability that the construction machinery claim application is an abnormal claim application. The closer the output predicted probability value is to 1, the higher the possibility that the construction machinery claim application is an abnormal claim application; the closer the predicted probability value is to 0, the higher the possibility that the construction machinery claim application is normal. The result of the output layer will be used as an important basis for judging the claim event detection.

[0050] Figure 5 It is the fourth schematic diagram of the structure of the claim event detection device for construction machinery provided by the embodiment of the present invention, as Figure 5 shown, the model training module 130 includes a data partitioning unit 131, a model training unit 132, a model verification unit 133, and a model testing unit 134, where: The data partitioning unit 131 is used to partition the feature data set into a training set, a verification set, and a test set. In the embodiment of the present invention, after the feature data set is labeled, the feature data set is partitioned into a training set, a verification set, and a test set according to the ratio of 70%, 20%, and 10%.

[0051] The model training unit 132 is used to train the convolutional neural network model according to the training set, and during the training process, the adaptive moment estimation optimization algorithm is used to adjust the parameters of the convolutional neural network model to obtain a pre-trained convolutional neural network model. In the embodiment of the present invention, the initial model parameters of the convolutional neural network model are set, and these initial model parameters include weight parameters, bias parameters, and hyperparameters. The training set is input into the convolutional neural network model for training. During the training process, the adaptive moment estimation optimization algorithm is used to adjust the model parameters, and the specific steps are as follows: First, initialize the first-order moment estimation m t and the second-order moment estimation v t , the first-order moment estimation m t represents the mean of the gradient, and the second-order moment estimation represents the uncentered variance of the gradient. Set the initial first-order moment estimation m 0 and the initial second-order moment estimation v 0 to 0. At the same time, set hyperparameters such as the learning rate and the exponential decay rate of the moment estimation. The learning rate can be set to a relatively small value, such as 0.001. The exponential decay rate β 1 of the first-order moment estimation is set to 0.9, and the exponential decay rate β 2 of the second-order moment estimation is set to 0.999.

[0052] During the model training process, for each training sample or mini-batch of samples, the gradient of the model loss function with respect to the model parameters is calculated. In the embodiments of the present invention, according to the cross-entropy loss function, the gradient of the model parameters on the current sample is calculated using the backpropagation algorithm, where the true label of the cross-entropy loss function uses 1 to represent the abnormal category and 0 to represent the normal category. Since the construction machinery operation data is the real operation data collected by the sensor unit 200, and the construction machinery maintenance data is the real maintenance data obtained from the data management server of the construction machinery enterprise's maintenance management system, both belong to normal data. Therefore, the construction machinery operation data and the construction machinery maintenance data are labeled as the normal category. The historical claim event data determines the true label according to the historical claim result. The historical claim event data with the historical claim result of "paid" is labeled as the normal category, and the historical claim event data with the historical claim result of "not paid" is labeled as the abnormal category.

[0053] Update the first moment estimate according to the calculated gradient. The update formula for the first moment estimate is as follows:

[0054] In the above formula, m t represents the first moment estimate at the current time step t , β 1 represents the exponential decay rate of the first moment estimate, m t-1 represents the first moment estimate at the previous time step t- 1, g t represents the current time step t 's gradient.

[0055] At the same time, update the second moment estimate. The update formula for the second moment estimate is as follows:

[0056] In the above formula, v t represents the second moment estimate at the current time step t , β 2 represents the exponential decay rate of the second moment estimate, v t-1 represents the second moment estimate at the previous time step t- 1.

[0057] Since there will be biases in the first moment estimate and the second moment estimate in the initial stage, bias correction is required. The bias-corrected first moment estimate is as follows:

[0058] In the above formula, represents the bias-corrected first moment estimate,t Denotes the current time step, i.e., the number of steps in the current training.

[0059] The second - moment estimate after bias correction is as follows:

[0060] In the above formula, Denotes the second - moment estimate after bias correction.

[0061] Finally, according to the first - moment estimate after bias correction and the second - moment estimate after bias correction, update the model parameters. The model parameter update formula is as follows:

[0062] In the above formula, Denotes the model parameters at the current time step t of the model, Denotes the model parameters at the previous time step t -1, Denotes the learning rate, Denotes a constant used to prevent the denominator from being zero.

[0063] By continuously repeating the above model parameter update process, during the model training process, gradually adjust the model parameters to make the cross - entropy loss function of the model gradually decrease, achieving the purpose of optimizing the model performance and obtaining a pre - trained convolutional neural network model.

[0064] The model verification unit 133 is used to perform multiple rounds of verification on the pre - trained convolutional neural network model using the K - fold cross - validation method based on the validation set, and obtain an initial convolutional neural network model. In the embodiments of the present invention, by dividing the validation set into K subsets and alternately using them as the validation subset and the training subset, perform multiple rounds of verification on the pre - trained convolutional neural network model to prevent overfitting and improve the generalization ability of the model, and obtain an initial convolutional neural network model.

[0065] The model testing unit 134 is used to test the initial convolutional neural network model according to the test set and obtain an initial claim event prediction model. In the embodiments of the present invention, use the test set to evaluate the initial convolutional neural network model, monitor indicators such as accuracy and recall rate during the model testing process through the TensorBoard visualization tool. If the accuracy and recall rate of the initial convolutional neural network model on the test set reach the preset threshold, it indicates that the model has good performance, and an initial claim event prediction model is obtained.

[0066] The federated learning module 140 is used to construct a federated learning framework based on the initial claim event prediction model, and perform gradient parameter update on the initial claim event prediction model based on the federated learning framework to obtain a target claim event prediction model. Figure 6It is the fifth schematic diagram of the structure of the claim event detection device for construction machinery provided by the embodiments of the present invention. As Figure 6 shown, the federated learning module 140 specifically includes a central control unit 141, a regional learning unit 142, an aggregation calculation unit 143, and a model update unit 144, where: The central control unit 141 is used to distribute the initial claim event prediction model and the initial model parameters to the clients of each regional node through the central server. In the embodiments of the present invention, the initial model parameters are randomly initialized, and each regional node is a branch of an insurance company or a regional department of a construction machinery enterprise in different regions. Each regional node performs initialization settings for the client to ensure smooth communication channels between nodes, and at the same time prepares local multi-source data, which includes local construction machinery operation data, local construction machinery maintenance data, and local historical claim event data of the regional node.

[0067] The regional learning unit 142 is used to obtain the local multi-source data of each regional node, and train the initial claim event prediction model according to the local multi-source data and the initial model parameters to obtain model parameter gradient information. In the embodiments of the present invention, the regional node trains the received initial claim event prediction model based on the local multi-source data. The training process also uses the adaptive moment estimation optimization algorithm to adjust the model parameters, and performs multiple rounds of training with the goal of minimizing the cross-entropy loss function. During the training process, the regional learning unit 142 only calculates the model parameter gradient information and does not directly transmit the original data. After the local training of each regional node is completed, the calculated model parameter gradient information is encrypted and uploaded to the aggregation calculation unit 143. The model parameter gradient information represents the update direction and amplitude of the model parameters on the local multi-source data. Optionally, the encryption method includes one or more of encryption methods such as homomorphic encryption, secure multi-party encryption, and differential privacy.

[0068] The aggregation calculation unit 143 is used to obtain the model parameter gradient information of each regional node, and perform aggregation calculation on the model parameter gradient information to obtain aggregated gradient information. The specific steps of the aggregation calculation are as follows: If the model parameter gradient information uses an encryption method, it is necessary to decrypt the model parameter gradient information before performing the aggregation calculation for subsequent calculations.

[0069] Then, according to the performance, data volume, or other factors of different regional nodes, corresponding weights can be assigned to the model parameter gradient information of each regional node. For example, a regional node with a larger data volume may contribute more to the update of the global model, so a higher weight can be given. The weight assignment can be achieved through a weight function W i to implement, i represents different regional nodes.

[0070] Next, aggregate the model parameter gradient information of each regional node according to the set aggregation method. In the embodiment of the present invention, the aggregation method adopts average aggregation, that is, sum the model parameter gradient information of all regional nodes and divide by the number of regional nodes. If there is weight assignment, calculate the weighted average. The aggregation formula is as follows:

[0071] In the above formula, represents the aggregated gradient information, i represents different regional nodes, N represents the total number of regional nodes, represents the i weight of the th regional node, i represents the model parameter gradient information of the

[0072] The model update unit 144 is used to update the global model parameters of the initial claim event prediction model according to the aggregated gradient information to obtain the target claim event prediction model. In the embodiment of the present invention, the model update unit 144 is used to update the global model parameters by using the stochastic gradient descent algorithm according to the aggregated gradient information. The global model parameters include, but are not limited to, the input weight matrix, the hidden layer weight matrix, the output weight matrix, and the bias term.

[0073] Input weight matrix W x is used to map the input data to the hidden state, and its dimension is the input feature dimension and the hidden layer dimension. The hidden layer weight matrix W h is used to calculate the transformation between hidden states, and its dimension is two hidden layer dimensions. The output weight matrix W o is used to map the hidden state to the output space, and its dimension is the hidden layer dimension and the number of output categories. The bias term includes the bias vector b h of the hidden layer b o and the bias vector W x of the output layer. In the embodiment of the present invention, the update method of the global model parameters adopts the stochastic gradient descent algorithm. The update formula of the input weight matrix

[0074] In the above formula, represents the updated input weight matrix, represents the input weight matrix at the current moment, a represents the learning rate, Represents the aggregated gradient information of the input weight matrix.

[0075] Similarly, the hidden layer weight matrix can be processed in the same way as the input weight matrix W h , the output weight matrix W o and the bias term are updated to obtain the target claim event prediction model. Optionally, the updated global model parameters are distributed to each regional node for the next round of local training. The federated learning framework breaks the dependence of machine learning on data centralization and has significant advantages in data privacy protection, model training efficiency improvement, and resource saving.

[0076] The model detection module 150 is used to obtain the target claim event data of the target construction machinery, perform data cleaning, data standardization, and feature extraction on the target claim event data to obtain target feature data, and input the target feature data into the target claim event prediction model for prediction to obtain the abnormal probability value of the target claim event of the target construction machinery. The target claim event data includes at least the target construction machinery operation data and the target construction machinery maintenance data of the target construction machinery. In the embodiment of the present invention, when a new target claim event arrives for the target construction machinery, the target claim event data is input into the model detection module 150 through the data interface of the device. The target claim event data includes the operation data and maintenance data for which the target construction machinery applies for a claim. The model detection module 150 performs data cleaning, data standardization, and feature extraction on the target claim event data to obtain target feature data, where the methods of data cleaning, data standardization, and feature extraction are the same as those of the data processing module 120 for multi-source data. The target feature data is input into the target claim event prediction model, and the target claim event prediction model quickly calculates the abnormal probability value of the target claim event of the target construction machinery according to the trained parameters and algorithms. Exemplarily, if a set of target claim event data is input, after data processing and input into the target claim event prediction model, and the target claim event prediction model outputs an abnormal probability value of 75% for the target claim event, it means that the probability of the target claim event being abnormal is 75%.

[0077] The data analysis module 160 is used to perform data comparison and analysis processing based on the target claim event data and the real-time multi-source data of the target construction machinery to obtain the number of abnormal items with the data comparison and analysis result being abnormal. In the embodiment of the present invention, the data analysis module 160 performs comparison and analysis processing on the operation data and maintenance data for which a claim is applied in the target claim event data and the real-time multi-source data of the target construction machinery based on a data analysis tool. The comparison and analysis processing include, but are not limited to, time comparison analysis, space comparison analysis, motion trajectory comparison analysis, maintenance item comparison analysis, and maintenance cost comparison analysis, etc.

[0078] Optionally, it is possible to compare and analyze whether the time and location in the target claim event data exactly match the actual working time and location in the target construction machinery operation data, or to compare and analyze whether the damage situation in the target claim event data is logically consistent with the repair items and costs in the target construction machinery repair data. It is also possible to compare the relevance between the faulty parts of the target construction machinery in the target claim event data and the repair items in the target construction machinery repair data, and evaluate whether the repair cost is within a reasonable range. Exemplarily, assuming that the target claim event data is "2025-03-10 14:00-16:00" with a fault occurring at "[116.402211, 39.912345]", while the actual construction machinery operation data shows that the target construction machinery was at another location [116.354433, 39.887654] during this time period, it indicates that the result of the spatial comparison analysis is abnormal. If the repair items shown in the target construction machinery repair data do not match the damage situation in the target claim event data, it indicates that the result of the repair item comparison analysis is abnormal.

[0079] The event detection module 170 is used to determine that the claim event detection result is an abnormal claim event if the abnormal probability value of the target claim event is greater than the preset probability threshold and the number of abnormal items is greater than the preset number of items. In the embodiment of the present invention, the target claim event is comprehensively judged as an abnormal claim event based on the abnormal probability value of the target claim event and the number of abnormal items, that is, whether there is an abnormal claim behavior.

[0080] Exemplarily, the preset probability threshold is set to 0.8, and the preset number of items is set to 2. If the abnormal probability value of the target claim event output by the target claim event prediction model is greater than 0.8, and the number of abnormal items with an abnormal data comparison analysis result is greater than 2, it is determined that the target claim event is an abnormal claim event, and there is an abnormal claim behavior. If the abnormal probability value of the target claim event output by the target claim event prediction model is less than or equal to 0.8, and the number of abnormal items with an abnormal data comparison analysis result does not exceed 2, it is determined that the target claim event is a normal claim event, and there is no abnormal claim behavior. If the abnormal probability value of the target claim event output by the target claim event prediction model is less than or equal to 0.8, but the number of abnormal items with an abnormal data comparison analysis result is greater than 2, it indicates that further manual review is required for determination.

[0081] As a further implementation manner of the embodiments of the present invention, a detection report may also be generated and sent to the terminal device of the claims settlement reviewer. The content of the detection report includes, but is not limited to, the abnormal probability value of the target claims event, the data comparison and analysis result, and the claims event detection result. The detection report can provide a comprehensive and accurate reference basis for the reviewer to carry out further investigation or processing work. If it is determined as an abnormal claims application, the claims settlement reviewer may reject the claims application according to the content of the report and timely feedback the relevant information to the insurance company and the construction machinery enterprise. At the same time, if the claims settlement reviewer finds that there is a misjudgment event in the target claims event prediction model, the historical claims event data of the misjudgment event will be re-annotated and incorporated into the training set to retrain the target claims event prediction model, thereby improving the prediction accuracy of the target claims event prediction model.

[0082] The claims event detection device for construction machinery provided by the present invention preprocesses data by acquiring multi-dimensional construction machinery operation data, construction machinery maintenance data, and historical claims event data, constructs a convolutional neural network model, and trains the convolutional neural network model according to the data set after data preprocessing to obtain a target claims event prediction model, which can comprehensively and accurately identify various claims abnormal behaviors of construction machinery and improve the accuracy of identifying the authenticity of claims events. By constructing a federated learning framework, the target claims event prediction model can continuously learn new claims event data, timely discover new types of claims abnormal behavior patterns and characteristics. With the continuous accumulation of data and the continuous optimization of the model, the device's ability to identify claims abnormal behaviors will continue to increase, effectively coping with the increasingly complex and changeable claims abnormal risks. The device realizes automatic data collection, analysis, and event detection, greatly shortening the claims settlement review cycle, reducing the manual review workload, and reducing the investment in labor costs.

[0083] Embodiment 2 Based on the same technical concept as in Embodiment 1 above, an embodiment of the present invention provides a method for detecting claims events of construction machinery, and this method is applied to the claims event detection device 100 in Embodiment 1. Figure 7 It is a schematic flowchart of the method for detecting claims events of construction machinery provided by an embodiment of the present invention, as Figure 7 shown, and this method includes the following steps: S100. Acquire multi-source data of construction machinery, and the multi-source data includes at least construction machinery operation data, construction machinery maintenance data, and historical claims event data; S200. Perform data cleaning, data standardization, and feature extraction on the multi-source data to obtain a feature data set; S300. Construct a convolutional neural network model, and use the feature data set to train the convolutional neural network model to obtain an initial claims event prediction model; S400. Construct a federated learning framework based on the initial claim event prediction model, and update the gradient parameters of the initial claim event prediction model based on the federated learning framework to obtain the target claim event prediction model; S500. Obtain the target claim event data of the target construction machinery, perform data cleaning, data standardization and feature extraction on the target claim event data to obtain the target feature data, and input the target feature data into the target claim event prediction model for prediction to obtain the target claim event anomaly probability value of the target construction machinery. The target claim event data includes at least the target construction machinery operation data and the target construction machinery maintenance data for which the target construction machinery applies for claims; S600. Perform data comparison and analysis processing based on the target claim event data and the real-time multi-source data of the target construction machinery to obtain the number of abnormal items with the data comparison and analysis result being abnormal; S700. If the target claim event anomaly probability value is greater than the preset probability threshold and the number of abnormal items is greater than the preset number of items, determine that the claim event detection result is an abnormal claim event.

[0084] The claim event detection method for construction machinery provided by the embodiments of the present invention can comprehensively and accurately identify the claim abnormal behaviors of various construction machinery, improve the accuracy of identifying claim abnormal events, avoid unnecessary claim expenditures through accurate identification of claim abnormal behaviors, save a large amount of funds for insurance companies, and improve the overall operation efficiency of the company.

[0085] It can be understood that the implementation manners in the claim event detection device for construction machinery described in the above Embodiment 1 are equally applicable to this embodiment and can achieve the same technical effects, so they will not be repeated here.

[0086] Embodiment 3 Based on the same concept, the embodiments of the present invention also provide an electronic device, Figure 8 which is a schematic structural diagram of an electronic device provided by the embodiments of the present invention. As Figure 8 shown, the electronic device 800 may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840. Among them, the processor 810, the communication interface 820, and the memory 830 communicate with each other through the communication bus 840. The processor 810 can call the logical instructions in the memory 830 to execute the steps of the claim event detection method for construction machinery as described in the above embodiments, for example, including: S100. Obtain the multi-source data of the construction machinery. The multi-source data includes at least the construction machinery operation data, the construction machinery maintenance data, and the historical claim event data; S200. Clean, standardize, and extract features from multi-source data to obtain a feature dataset; S300. Build a convolutional neural network model and train the convolutional neural network model using the feature dataset to obtain an initial claim event prediction model; S400. Build a federated learning framework based on the initial claim event prediction model, and update the gradient parameters of the initial claim event prediction model based on the federated learning framework to obtain a target claim event prediction model; S500. Obtain the target claim event data of the target construction machinery, clean, standardize, and extract features from the target claim event data to obtain target feature data, input the target feature data into the target claim event prediction model for prediction to obtain the target claim event anomaly probability value of the target construction machinery. The target claim event data includes at least the target construction machinery operation data and the target construction machinery maintenance data for which the target construction machinery applies for a claim; S600. Perform data comparison and analysis processing based on the target claim event data and the real-time multi-source data of the target construction machinery to obtain the number of abnormal items for which the data comparison and analysis result is abnormal; S700. If the target claim event anomaly probability value is greater than a preset probability threshold and the number of abnormal items is greater than a preset number of items, determine that the claim event detection result is an abnormal claim event.

[0087] Among them, the processor 810 can be a central processing unit (CPU). The processor can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. chips, or a combination of the above types of chips.

[0088] In addition, when the logical instructions in the above-mentioned memory 830 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0089] The memory 830 may include a program storage area and a data storage area. Among them, the program storage area can store an operating system and application programs required for at least one function; the data storage area can store data created by the processor, etc. In addition, the memory may include a high-speed random access memory and may also include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some embodiments, the memory may optionally include a memory remotely set relative to the processor, and these remote memories can be connected to the processor through a network. Examples of the above-mentioned network include but are not limited to the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0090] Embodiment 4 Based on the same concept, an embodiment of the present invention further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and the computer program includes at least one segment of code. The at least one segment of code can be executed by a master control device to control the master control device to implement the steps of the claim event detection method for construction machinery as described in the above-mentioned various embodiments, for example, including: S100. Obtain multi-source data of construction machinery. The multi-source data at least includes construction machinery operation data, construction machinery maintenance data, and historical claim event data; S200. Perform data cleaning, data standardization, and feature extraction on the multi-source data to obtain a feature data set; S300. Construct a convolutional neural network model, and use the feature data set to train the convolutional neural network model to obtain an initial claim event prediction model; S400. Construct a federated learning framework according to the initial claim event prediction model, and update the gradient parameters of the initial claim event prediction model based on the federated learning framework to obtain a target claim event prediction model; S500. Obtain the target claim event data of the target construction machinery, perform data cleaning, data standardization, and feature extraction on the target claim event data to obtain target feature data, input the target feature data into the target claim event prediction model for prediction, and obtain the abnormal probability value of the target claim event of the target construction machinery. The target claim event data includes at least the target construction machinery operation data and the target construction machinery maintenance data for which the target construction machinery applies for claims. S600. Perform data comparison and analysis processing based on the target claim event data and the real-time multi-source data of the target construction machinery to obtain the number of abnormal items with abnormal data comparison and analysis results. S700. If the abnormal probability value of the target claim event is greater than the preset probability threshold and the number of abnormal items is greater than the preset number of items, determine that the claim event detection result is an abnormal claim event.

[0091] Based on the same technical concept, an embodiment of the present invention further provides a computer program, which, when executed by a main control device, is used to implement the above method embodiment.

[0092] The computer program can be stored in whole or in part on a computer-readable storage medium packaged together with the processor, or can be stored in whole or in part on a memory not packaged together with the processor.

[0093] Based on the same technical concept, an embodiment of the present invention further provides a processor, which is used to implement the above method embodiment. The above processor can be a chip.

[0094] In summary, the claim event detection device, method, equipment, and storage medium for construction machinery provided by the present invention perform data preprocessing by obtaining multi-dimensional construction machinery operation data, construction machinery maintenance data, and historical claim event data, construct a convolutional neural network model, and train the convolutional neural network model according to the data set after data preprocessing to obtain a target claim event prediction model, which can comprehensively and accurately identify various claim abnormal behaviors of construction machinery and improve the accuracy of identifying the authenticity of claim events. By constructing a federated learning framework, the target claim event prediction model can continuously learn new claim event data, timely discover new types of claim abnormal behavior patterns and characteristics. With the continuous accumulation of data and the continuous optimization of the model, the device's ability to identify claim abnormal behaviors will continue to increase, effectively coping with the increasingly complex and changeable abnormal claim risks. The device realizes automatic data collection, analysis, and event detection, greatly shortening the claim review cycle, reducing the manual review workload, reducing the investment in labor costs. More importantly, by accurately identifying claim abnormal behaviors, it avoids unnecessary claim expenditures, saves a large amount of funds for insurance companies, and improves the overall operation efficiency of the company.

[0095] Reference to "embodiments" in this document means that the specific features, structures, or characteristics described in connection with the embodiments can be included in at least one embodiment of the present application. The phrase appears in various places in the specification and does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art will explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0096] The above-described embodiments merely represent several implementation manners of the present invention. The description thereof is relatively specific and detailed, but should not be construed as a limitation on the scope of the patent for the present invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can be made, and these all fall within the protection scope of the present invention. Therefore, the protection scope of the patent for the present invention shall be subject to the appended claims.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A claim event detection device for construction machinery, characterized in that: The device comprises: A data acquisition module, used to acquire multi-source data of engineering machinery, wherein the multi-source data at least includes engineering machinery operation data, engineering machinery maintenance data and historical claims event data; A data processing module, used for performing data cleaning, data standardization and feature extraction on the multi-source data to obtain a feature data set; A model training module, used to construct a convolutional neural network model, train the convolutional neural network model using the feature data set, and obtain an initial claims event prediction model; A federated learning module, used to construct a federated learning framework according to the initial claim event prediction model, and update the gradient parameters of the initial claim event prediction model based on the federated learning framework to obtain a target claim event prediction model; A model detection module is used to obtain target claim event data of a target engineering machinery, perform data cleaning, data standardization and feature extraction on the target claim event data to obtain target feature data, input the target feature data into the target claim event prediction model for prediction, and obtain an abnormal probability value of the target claim event of the target engineering machinery, wherein the target claim event data at least includes the target engineering machinery operation data and target engineering machinery maintenance data for which the target engineering machinery applies for a claim; A data analysis module, used to perform data comparison and analysis based on the target claim event data and the real-time multi-source data of the target engineering machinery, and obtain the number of abnormal items with abnormal data comparison and analysis results; The event detection module is used to determine that the claim event detection result is an abnormal claim event if the abnormal probability value of the target claim event is greater than a preset probability threshold and the number of abnormal items is greater than a preset number of items.

2. The claim settlement event detection device for construction machinery according to claim 1, characterized in that: The data acquisition module comprises: A first data acquisition unit is connected to the sensor unit of the engineering machinery for communication, and the first data acquisition unit is used to acquire the operation data of the engineering machinery collected by the sensor unit; a second data acquisition unit, which is in communication connection with the first server, the first server being used to store the engineering machinery maintenance data of the engineering machinery, and the second data acquisition unit being used to acquire the engineering machinery maintenance data stored in the first server; The third data acquisition unit is communicatively connected to the second server, the second server is used to store the historical claims event data, and the third data acquisition unit is used to acquire the historical claims event data, wherein the historical claims event data at least includes historical operation data of engineering machinery, historical maintenance data of engineering machinery, historical claims results and historical claims amounts.

3. The claim settlement event detection device for construction machinery according to claim 1, characterized in that: The data processing module comprises: A data cleaning unit, used to perform preliminary cleaning on the multi-source data using a spatial clustering algorithm to obtain the multi-source data after preliminary cleaning; A data comparison unit is used to compare and count the multi-source data after the preliminary cleaning, remove abnormal data, and obtain an initial data set; A numerical processing unit, used for performing standardization processing on the numerical data in the initial data set by using a numerical standardization method to obtain an initial standard data set; An encoding processing unit, used for encoding the categorized data in the initial standard data set by using a one-hot encoding technique to obtain a standard data set; A first feature extraction unit is used to extract feature data from the engineering machinery operation data in the standard data set by using a sliding window algorithm to obtain an initial feature data set; The second feature extraction unit is used to extract association rule features from the engineering machinery maintenance data in the initial feature data set by using an association rule mining algorithm to obtain the feature data set.

4. The claim settlement event detection device for construction machinery according to claim 1, characterized in that: The convolutional neural network model at least includes an input layer, a convolution unit, a pooling unit, a fully connected unit and an output layer, wherein: The input layer is used to concatenate the numerical feature data and the categorical feature data in the feature dimension to obtain a feature tensor; The convolution unit includes a plurality of convolution layers, the plurality of convolution layers include convolution kernels of different sizes, and the plurality of convolution layers are stacked to gradually extract features from the feature tensor to obtain a feature vector; The pooling unit includes a maximum pooling layer, and the maximum pooling layer is used to perform feature dimension reduction on the feature vector output by the convolution layer to obtain a reduced dimension feature vector; The fully connected unit includes a plurality of fully connected layers, and the plurality of fully connected layers are used to perform a linear transformation on the dimension-reduced feature vector to obtain a comprehensive feature vector; The output layer is used to convert the comprehensive feature vector into a predicted probability value through an activation function.

5. The claim settlement event detection device for construction machinery according to claim 4, characterized in that: The model training module includes: A data division unit, used for dividing the feature data set into a training set, a validation set and a test set; A model training unit, used to train the convolutional neural network model according to the training set, and to adjust the parameters of the convolutional neural network model using an adaptive moment estimation optimization algorithm during the training process to obtain a pre-trained convolutional neural network model; A model verification unit, used to perform multiple rounds of verification on the pre-trained convolutional neural network model using a K-fold cross-validation method according to the verification set to obtain an initial convolutional neural network model; A model testing unit is used to test the initial convolutional neural network model according to the test set to obtain the initial claims event prediction model.

6. The claim settlement event detection device for construction machinery according to claim 1, characterized in that: The federated learning module includes: A central control unit, used to distribute the initial claim event prediction model and initial model parameters to the clients of each regional node through the central server; A regional learning unit, used for acquiring local multi-source data of each regional node, training the initial claim event prediction model according to the local multi-source data and the initial model parameters, and obtaining model parameter gradient information; An aggregation calculation unit, used to obtain the model parameter gradient information of each of the regional nodes, perform aggregation calculation on the model parameter gradient information, and obtain aggregated gradient information; A model updating unit is used to update the global model parameters of the initial claim event prediction model according to the aggregated gradient information to obtain the target claim event prediction model.

7. The claim settlement event detection device for construction machinery according to claim 6, characterized in that: The global model parameters include at least an input weight matrix, a hidden layer weight matrix, an output weight matrix and a bias term, and the model updating unit is used to update the input weight matrix, the hidden layer weight matrix, the output weight matrix and the bias term using a stochastic gradient descent algorithm according to the aggregated gradient information.

8. A method for detecting claims events of construction machinery, characterized in that: The method is applied to the claim settlement event detection device for engineering machinery according to any one of claims 1 to 7, and the method comprises: Acquire multi-source data of the construction machinery, wherein the multi-source data at least includes construction machinery operation data, construction machinery maintenance data, and historical claims event data; Performing data cleaning, data standardization and feature extraction on the multi-source data to obtain a feature data set; Constructing a convolutional neural network model, and using the feature data set to train the convolutional neural network model to obtain an initial claims event prediction model; Building a federated learning framework according to the initial claim event prediction model, and updating the gradient parameters of the initial claim event prediction model based on the federated learning framework to obtain a target claim event prediction model; Obtain target claim event data of a target engineering machinery, perform data cleaning, data standardization and feature extraction on the target claim event data to obtain target feature data, input the target feature data into the target claim event prediction model for prediction, and obtain an abnormal probability value of the target claim event of the target engineering machinery, wherein the target claim event data at least includes the target engineering machinery operation data and target engineering machinery maintenance data for which the target engineering machinery applies for a claim; Performing data comparison and analysis based on the target claim event data and the real-time multi-source data of the target engineering machinery to obtain the number of abnormal items with abnormal data comparison and analysis results; If the abnormal probability value of the target claim event is greater than the preset probability threshold, and the number of abnormal items is greater than the preset number of items, the claim event detection result is determined to be an abnormal claim event.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: The processor executes the computer program to implement the method for detecting claims events for construction machinery as described in claim 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, the method for detecting claims events of construction machinery as described in claim 8 is implemented.

Images