Elevator claim settlement anomaly detection method, device and equipment and storage medium
By combining the Internet of Things and feature encoders, elevator operation data is processed automatically, solving the problems of high operating costs and low efficiency in traditional elevator claims processing, and enabling fast and accurate fault diagnosis and claims decision-making.
Patent Information
- Application Number
- CN202511708015.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-19
- Publication Date
- 2026-02-24
AI Technical Summary
Traditional elevator claims processing suffers from high operating costs, low efficiency, inaccurate judgments, and susceptibility to human error, especially in preventing insurance fraud, which requires extensive manual verification and multiple confirmations.
By acquiring elevator detection data and reference datasets through the Internet of Things, a baseline vector is generated using a pre-trained feature encoder. The distance between the detection vector and the baseline vector is calculated. If the distance exceeds a threshold, an anomaly is identified, and anomaly analysis is performed to provide analysis results to guide claims requests.
It enables rapid and accurate elevator malfunction diagnosis, reduces manual verification time and costs, improves claims efficiency and accuracy, avoids human error, and ensures the fairness of claims decisions.
Smart Images
Figure CN121553787A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of artificial intelligence technology, specifically to a method, device, equipment, and storage medium for detecting elevator claims anomalies. Background Technology
[0002] In elevator claims processing, to prevent insurance fraud, insurance companies often need to dispatch a large number of staff to the site for verification. This process is not only time-consuming and labor-intensive, but also often requires multiple confirmations of the specific circumstances of the elevator malfunction, such as whether a malfunction actually exists, what the cause of the malfunction is, and whether it meets the compensation conditions of the insurance policy. Due to the need for on-site inspection, review of repair records, and equipment diagnosis, the entire claims cycle is often extended, resulting in significant operating costs for insurance companies. At the same time, manual verification is easily limited by the on-site environment and the experience of the staff, which may affect the accuracy and efficiency of the judgment, further increasing the uncertainty of the claims process. Therefore, the traditional claims process is not only costly, but also prone to delays or incorrect compensation judgments due to inefficiency or human error. Summary of the Invention
[0003] To address the aforementioned issues, this application provides a method, apparatus, equipment, and storage medium for detecting elevator claims anomalies. This method can reduce operating costs during the claims process, increase claims processing speed, and enable insurance companies to process claims requests more efficiently while preventing insurance fraud.
[0004] The embodiments of this application disclose the following technical solutions:
[0005] A method for detecting abnormalities in elevator claims, the method comprising:
[0006] In response to a claim request from the first target elevator, the system obtains the test data and the comparison dataset of the first target elevator from the Internet of Things. The test data includes the latest raw test data of the first target elevator. The comparison dataset includes multiple raw normal comparison data of the first target elevator. The raw data in the test data and the comparison dataset include the first target elevator's running time, number of door openings and closings, running distance, number of wire rope bends, and number of passengers carried per day.
[0007] A baseline vector is obtained based on the aforementioned comparison dataset, and the data to be detected is input into a pre-trained feature encoder to obtain the vector to be detected.
[0008] If the distance between the vector to be detected and the reference vector is greater than the distance threshold, then it is determined that the first target elevator is abnormal;
[0009] Based on the control dataset and the data to be detected, anomaly analysis is performed on the first target elevator that exhibits anomalies, and the analysis results are obtained.
[0010] The analysis results are sent to the user, who is then reminded that the claim for the first target elevator is reasonable.
[0011] In one possible implementation, the training process of the feature encoder includes:
[0012] Training samples are constructed, comprising multiple positive sample pairs. Each positive sample pair includes original sample data and augmented sample data. The original sample data includes the normal operation data of the second target elevator on a certain day, including the duration of each run, the number of door openings and closings, the distance traveled, the number of wire rope bends, and the number of passengers carried each time. The augmented sample data is obtained by augmenting the original sample data.
[0013] The training samples are used as input, and the parameters of the initial model are iteratively optimized using the NT-Xent loss function until the training cutoff condition is met, thus obtaining the optimized model; the initial model includes an initial encoder and an initial projection head;
[0014] Remove the trained projection head from the optimized model to obtain the feature encoder.
[0015] In one possible implementation, the method further includes:
[0016] When the Internet of Things updates the new normal operation data of the second target elevator, the new normal operation data is added to the training sample to obtain the latest training sample;
[0017] The parameters of the optimization model are re-optimized based on the latest training samples to obtain the latest optimization model;
[0018] The step of removing the trained projection head from the optimized model to obtain the feature encoder includes:
[0019] The feature encoder is obtained by removing the trained projection head from the latest optimized model.
[0020] In one possible implementation, obtaining the benchmark vector based on the comparison dataset includes:
[0021] The comparison dataset is input into the feature encoder to obtain multiple feature embedding vectors;
[0022] The reference vector is obtained by averaging the multiple feature embedding vectors.
[0023] In one possible implementation, the step of performing anomaly analysis on the first target elevator exhibiting anomalies based on the control dataset and the data to be detected, and obtaining analysis results, includes:
[0024] The runtime of each run in the data to be detected is compared with the runtime of each run in the control dataset. If the runtime of each run in the data to be detected is greater than or less than any runtime of each run in the control dataset, it is determined that the first target elevator has an abnormal runtime.
[0025] The number of door openings and closings in the data to be detected is compared with each number of door openings and closings in the control dataset. If the number of door openings and closings in the data to be detected is greater than or less than any number of door openings and closings in the control dataset, then it is determined that the first target elevator has an abnormal number of door openings and closings.
[0026] The running distance of each time in the data to be detected is compared with the running distance of each time in the control dataset. If the running distance of each time in the data to be detected is greater than or less than any running distance of each time in the control dataset, it is determined that the first target elevator has an abnormal running distance.
[0027] The number of times each door opens and closes in the data to be detected is compared with the number of times each door opens and closes in the control dataset. If the runtime of each run in the data to be detected is greater than or less than any one of the runtimes in the control dataset, then it is determined that the first target elevator has an abnormal runtime.
[0028] The number of times the wire rope is bent in each test data is compared with the number of times the wire rope is bent in each test data in the control dataset. If the number of times the wire rope is bent in each test data is greater than or less than any one of the numbers of times the wire rope is bent in the control dataset, it is determined that the first target elevator has an abnormal number of wire rope bends.
[0029] The number of passengers in each trip in the data to be detected is compared with the number of passengers in each trip in the control dataset. If the number of passengers in each trip in the data to be detected is greater than or less than any number of passengers in each trip in the control dataset, it is determined that the first target elevator has an abnormal running time.
[0030] An elevator claims anomaly detection device, the device comprising:
[0031] The first acquisition unit is configured to, in response to a claim request from the first target elevator, acquire the test data and the comparison dataset of the first target elevator from the Internet of Things; the test data includes the latest original test data of the first target elevator; the comparison dataset includes multiple original normal comparison data of the first target elevator; the original data in the test data and the comparison dataset include the first target elevator's running time, number of door openings and closings, running distance, number of wire rope bends, and number of passengers carried per day.
[0032] The second acquisition unit is used to acquire a reference vector based on the comparison dataset and input the data to be detected into a pre-trained feature encoder to obtain the vector to be detected.
[0033] An anomaly detection unit is used to determine that the first target elevator is abnormal if the distance between the vector to be detected and the reference vector is greater than a distance threshold.
[0034] An anomaly analysis unit is used to perform anomaly analysis on the first target elevator that has anomalies based on the control dataset and the data to be detected, and to obtain analysis results.
[0035] The notification unit is used to send the analysis results to the user and remind the user that the claim for the first target elevator is reasonable.
[0036] In one possible implementation, the device further includes:
[0037] A construction unit is used to construct training samples, which include multiple positive sample pairs. Each positive sample pair includes original sample data and augmented sample data. The original sample data includes the normal operation data of the second target elevator on a certain day. The normal operation data includes the duration of each run, the number of door openings and closings, the distance traveled each time, the number of wire rope bends each time, and the number of passengers carried each time. The augmented sample data is obtained by data augmentation based on the original sample data.
[0038] The first parameter optimization unit is used to take the training samples as input and iteratively optimize the parameters of the initial model using the NT-Xent loss function until the training cutoff condition is met, thus obtaining the optimized model; the initial model includes an initial encoder and an initial projection head;
[0039] The removal unit is used to remove the trained projection head from the optimized model to obtain the feature encoder.
[0040] In one possible implementation, the device further includes:
[0041] The sample update unit is used to add the new normal operation data of the second target elevator to the training sample when the Internet of Things updates the new normal operation data of the second target elevator to obtain the latest training sample.
[0042] The second parameter optimization unit is used to re-optimize the parameters of the optimization model based on the latest training samples to obtain the latest optimization model;
[0043] The removal unit is specifically used for:
[0044] The feature encoder is obtained by removing the trained projection head from the latest optimized model.
[0045] An elevator claim anomaly detection device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the elevator claim anomaly detection method as described above.
[0046] A computer-readable storage medium storing instructions that, when executed on a terminal device, cause the terminal device to perform the elevator claim anomaly detection method as described above.
[0047] Compared with the prior art, this application has the following advantages:
[0048] This application provides a method, apparatus, device, and storage medium for detecting elevator claim anomalies. Specifically, when executing the elevator claim anomaly detection method provided in this application, the system first responds to elevator claim requests by obtaining the latest data of the target elevator and its normal operation comparison dataset from an Internet of Things (IoT) system. This data includes various elevator operating indicators, such as the duration of each run, the number of door openings and closings, the running distance, the number of wire rope bends, and the number of passengers carried. A baseline vector is generated based on the comparison dataset, and the data to be detected is processed by a feature encoder to obtain a vector to be detected. If the distance between the vector to be detected and the baseline vector exceeds a set threshold, the elevator is determined to be abnormal, and further anomaly analysis is performed. The analysis results not only help to quickly determine whether the elevator has a malfunction, but also help to judge the rationality of the claim request based on the actual situation. Finally, the analysis results are fed back to the user, providing clear guidance on whether the claim request is reasonable.
[0049] This application utilizes IoT technology to acquire elevator operating data in real time and automates the processing using a pre-trained feature encoder, eliminating the need for manual verification in the claims process. By comparing the data to be detected with normal data in a control dataset, it can accurately determine whether the elevator is malfunctioning, thus quickly assessing the reasonableness of the claim request. This avoids multiple on-site inspections and repeated confirmations required by traditional methods, significantly shortening the claims cycle, reducing the labor costs of on-site verification, and improving claims efficiency and accuracy. Furthermore, the detection results of this application are not only more objective but also avoid the subjectivity and errors inherent in manual judgment, reducing the risk of claims errors. Attached Figure Description
[0050] To more clearly illustrate the technical solutions in this embodiment or the prior art, the drawings used in the description of the embodiment or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0051] Figure 1 A flowchart of an elevator claims anomaly detection method provided in this application embodiment;
[0052] Figure 2 A flowchart of a feature encoder training and detection method provided in this application embodiment;
[0053] Figure 3 This is a schematic diagram of the structure of an elevator claims anomaly detection device provided in an embodiment of this application. Detailed Implementation
[0054] To facilitate understanding of the technical solutions provided in the embodiments of this application, the background technology involved in the embodiments of this application will be described below.
[0055] In the elevator claims process, to prevent insurance fraud, insurance companies often need to dispatch a large number of staff for on-site verification. This process is not only time-consuming but also very manpower-intensive. During on-site verification, staff need to confirm the specific circumstances of the elevator malfunction multiple times, such as whether the malfunction actually exists, what the cause of the malfunction is, and whether these malfunctions meet the compensation conditions of the insurance policy.
[0056] The existence of on-site inspections, repair record reviews, and equipment diagnostics often prolongs the claims process, resulting in significant operating costs for insurance companies. Furthermore, manual verification is susceptible to the influence of the on-site environment and staff experience, potentially leading to decreased accuracy and efficiency in judgment, further increasing the uncertainty and complexity of claims processing.
[0057] Therefore, traditional claims processes are not only costly, but also prone to delays or incorrect payouts due to inefficiency or human error. This not only affects the operational efficiency of insurance companies, but also reduces customer satisfaction.
[0058] To address this issue, this application provides a method, apparatus, device, and storage medium for detecting elevator claim anomalies. Upon receiving an elevator claim request for a target elevator, the system automatically retrieves the elevator's detection data and reference dataset from the Internet of Things (IoT). The detection data includes the elevator's latest raw data, while the reference dataset includes multiple raw data records of the elevator during normal operation. This data covers key indicators such as runtime, number of door openings and closings, travel distance, number of wire rope bends, and number of passengers. Next, the distance between the detection vector and the reference vector is calculated. If this distance exceeds a set threshold, the elevator is deemed to be abnormal. For elevators with detected anomalies, further anomaly analysis is performed. The analysis results help accurately assess the cause of the elevator malfunction and the reasonableness of the claim request. Finally, the analysis results are fed back to the user, reminding them that the claim request is reasonable. This application not only significantly reduces labor costs and on-site verification time but also improves the accuracy and efficiency of judgment. Through automated and intelligent data processing, elevator malfunctions can be quickly and accurately identified and claims assessed, thereby shortening the claims process, reducing operating costs, and improving customer satisfaction. Furthermore, because the entire process is based on data analysis and machine learning technologies, human error can be effectively avoided, ensuring the fairness and accuracy of claims decisions.
[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.
[0060] See Figure 1 The figure is a flowchart of an elevator claims anomaly detection method provided in an embodiment of this application. Figure 1 As shown, the elevator claim anomaly detection method may include steps S101-S105:
[0061] S101: In response to the claim request of the first target elevator, obtain the detection data and comparison dataset of the first target elevator from the Internet of Things.
[0062] In this method, the first step is to respond to the claim request of the first target elevator by obtaining the elevator's test data and reference dataset from the Internet of Things (IoT). Specifically, the test data includes the latest raw test data of the first target elevator, while the reference dataset contains multiple raw data records of the elevator during normal operation. The test data includes the latest raw data of the elevator, while the reference dataset contains multiple historical data records of the elevator during normal operation.
[0063] These data cover several key metrics for the elevator during a given day, such as runtime per trip, number of door openings and closings per trip, distance traveled per trip, number of wire rope bends per trip, and number of passengers carried per trip. These parameters are calculated by dividing the elevator's total runtime, number of door openings and closings, distance traveled, number of wire rope bends, and number of passengers carried per day by the corresponding number of trips.
[0064] For example, suppose the first target elevator ran 301 times in a certain day, then...
[0065] The duration of each run of the target elevator on a certain day is 103.2 minutes / 301 times = 0.343 minutes / run.
[0066] The number of times the target elevator opens and closes its doors on a given day is 364 times / 301 times = 1.209.
[0067] The target elevator's travel distance per trip on a given day is 2840 meters / 301 trips = 9.435 meters per trip.
[0068] The number of times the target elevator's steel cable bends on a given day is 150 times / 301 times = 0.498.
[0069] The number of passengers carried by the target elevator on a given day is 560 people / 301 times = 1.86 people per trip.
[0070] In this way, the system can obtain detailed operating parameters of the elevator and compare them with normal operating data to determine if any abnormalities exist. This precise data collection and processing method can greatly improve the accuracy and efficiency of claims processing.
[0071] S102: Obtain a baseline vector based on the reference dataset, and input the data to be detected into a pre-trained feature encoder to obtain the vector to be detected.
[0072] To detect elevator claim anomalies, a baseline vector can be obtained based on a comparison dataset. The comparison dataset includes multiple raw data records of the elevator during normal operation, including runtime, number of door openings and closings, travel distance, number of wire rope bends, and number of passengers carried per trip. This normal operation data is input into a pre-trained feature encoder to generate the baseline vector. The baseline vector, calculated based on the normal operation data, represents the feature representation of the elevator during normal operation.
[0073] The data to be detected includes the latest raw data of the elevator, including parameters such as the duration of each run, the number of door openings and closings, the distance traveled per run, the number of bends in the steel cable, and the number of passengers carried per run. This data can be input into a pre-trained feature encoder, which transforms the complex raw data into a low-dimensional vector representation, i.e., the vector to be detected.
[0074] A feature encoder is a pre-trained model that can efficiently transform raw high-dimensional data into dimensional feature vectors.
[0075] S103: If the distance between the vector to be detected and the reference vector is greater than the distance threshold, then it is determined that the first target elevator is abnormal.
[0076] To determine if there is an anomaly in the elevator, the distance between the vector to be detected and the reference vector can be calculated to measure the degree of deviation between the current operating state and the normal state.
[0077] Specifically, if the distance between the vector to be detected and the reference vector (such as Euclidean distance, or other similarity metrics) is greater than a preset distance threshold (such as 0.1 or 0.2), the first target elevator is considered to be abnormal. This distance threshold is a predefined value used to determine whether the difference between vectors exceeds the normal range. The distance threshold is usually set based on historical data and statistical analysis to ensure that anomalies can be effectively detected while avoiding false alarms.
[0078] S104: Based on the control dataset and the data to be detected, perform anomaly analysis on the first target elevator that has anomalies, and obtain the analysis results.
[0079] After identifying an anomaly in the primary target elevator, further analysis can be conducted by comparing the data to be tested with historical normal data in a control dataset to determine if there are any significant abnormal behaviors or operational deviations. In this process, the data to be tested represents the elevator's actual operating data over a specific period, while the control dataset provides reference values and patterns for the elevator's long-term normal operation. By comparing the data to be tested with the control data, it is possible to detect indicators that deviate from normal standards, such as the duration of each run, the number of door openings and closings, and the frequency of wire rope bending, to determine if these values exceed normal limits.
[0080] S105: Send the analysis results to the user and remind the user that the claim for the first target elevator is reasonable.
[0081] Finally, the system can send the results of the previous analysis to the user for their reference. The analysis results will list detailed data on the elevator's deviation from normal operating conditions, helping the user to fully understand the elevator's current status. Based on this data, the system will also clearly remind the user that the claim for the primary target elevator is reasonable. The reminder may include key information supporting the claim, such as the severity of the malfunction, the specific values of the abnormal indicators, and their impact on the elevator's normal operation, thereby ensuring that the user can make decisions based on accurate data and provide the valid evidence required for the claim.
[0082] Based on the content of S101-S105, the process first responds to the claim request from the target elevator, obtains the target elevator's detection data and reference dataset from the Internet of Things (IoT), and obtains a baseline vector based on the reference dataset. The detection data is then input into a feature encoder to obtain the detection vector. Next, the distance between the detection vector and the reference vector is compared with a distance threshold to determine if the target elevator is abnormal. If the target elevator is abnormal, anomaly analysis is performed based on the anomalies found in the reference dataset and the detection data to obtain the analysis results. Finally, the analysis results are sent to the user, reminding them that the claim request for the target elevator is reasonable. This application can reduce operating costs in the claims process, improve claims speed, and enable insurance companies to process claims more efficiently while preventing insurance fraud.
[0083] In one possible implementation, such as Figure 2 As shown, the training process of the feature encoder specifically includes S201-S203:
[0084] S201: Construct training samples.
[0085] In training the feature encoder, the first step is to construct training samples, which consist of multiple positive sample pairs. Each positive sample pair contains two parts: original sample data and augmented sample data. The original sample data comes from the normal operation data of the second target elevator (which can be any elevator), recording the elevator's specific operation on a given day, including the duration of each run, the number of door openings and closings, the travel distance, the number of wire rope bends, and the number of passengers carried. This data reflects the elevator's operational characteristics under normal conditions. To enhance the model's robustness, data augmentation is performed on each feature value based on the original sample data, such as adding Gaussian noise to each feature, with the standard deviation of the noise set to 0.02. This method generates one or more augmented samples for each original sample data, forming multiple positive sample pairs. For example, for original sample data A1, its original data and the augmented data after adding noise are as follows:
[0086] A1_view1 (original): [0.343, 1.000, 1.209, 9.435, 0.498, 1.860];
[0087] A1_view2 (enhanced): [0.348, 0.995, 1.215, 9.445, 0.503, 1.868].
[0088] These two samples form a positive sample pair. The same augmentation process is applied to other original sample data (such as A2, A3, etc.) to generate corresponding positive sample pairs, for example:
[0089] A2_view1 (original): [0.331, 1.000, 1.256, 8.719, 0.488, 1.478];
[0090] A2_view2 (enhanced): [0.336, 0.995, 1.262, 8.729, 0.493, 1.486].
[0091] In this way, we ultimately created multiple augmented views for various elevator samples and combined these views into multiple positive sample pairs for use in training the feature encoder. These positive sample pairs, by providing both the original data and the noise-enhanced data, help the encoder learn more robust feature representations, thereby improving the accuracy and stability of the model in practical applications.
[0092] S202: Using the training samples as input, iteratively optimize the parameters of the initial model using the NT-Xent loss function until the training cutoff condition is met, and then stop to obtain the optimized model.
[0093] During model training, the previously constructed training samples can be input into the initial model for training. The model is iteratively optimized using the NT-Xent (Normalized Temperature-scaled Cross Entropy) loss function, with the aim of minimizing the difference in distance between samples, until the training cutoff condition is met, at which point training stops and the optimized model is obtained.
[0094] Specifically, the initial model consists of two main parts: the initial encoder and the initial projection head.
[0095] The initial encoder structure is as follows: The input layer contains 6 neurons, corresponding to standardized data of elevator operation features (including runtime, number of door openings and closings, etc.). Next, the first hidden layer contains 16 neurons, employing a linear transformation combined with the ReLU activation function to enhance the model's non-linear expressive power; the second hidden layer contains 12 neurons, also using a linear transformation and ReLU activation function; the third hidden layer is further reduced to 8 neurons to continue processing higher-level feature representations. Finally, the output layer consists of 4 neurons, used to generate the final feature embedding vector.
[0096] The initial projection head receives the 4D feature embedding vector output by the encoder and performs further mapping and transformation. Its input layer also consists of 4 neurons, and its output layer has 4 neurons to output normalized contrastive features. The hidden layer contains 2 neurons using the ReLU activation function. The purpose of this projection head is to further transform the features output by the encoder into a form suitable for contrastive learning, thereby making similar samples as close as possible in the feature space and dissimilar samples as far apart as possible.
[0097] Through this training process, the model will gradually adjust its parameters and optimize the feature representation of elevator operation data, thereby improving its performance in subsequent tasks, such as anomaly detection and fault diagnosis.
[0098] S203: Remove the trained projection head from the optimized model to obtain the feature encoder.
[0099] During model optimization, the trained projector head is only used to project the features output by the encoder into the contrastive learning task for optimization within the contrastive loss function. Once the model training is complete and the feature embedding space has been effectively optimized using the NT-Xent loss function, the projector head's role is fulfilled. Therefore, after obtaining the optimized model, the projector head is removed, retaining only the trained feature encoder. The feature encoder is responsible for extracting and representing useful features from the original input data, eliminating the need for further mapping by the projector head. The resulting feature encoder, obtained after removing the projector head, is the core of the model and can be used in practical tasks. This approach not only preserves the optimized feature representation but also simplifies the model structure, improving its efficiency and scalability in real-world applications.
[0100] In one possible implementation, the method further includes:
[0101] When the IoT updates the new normal operation data of the second target elevator, the new normal operation data is added to the training samples to obtain the latest training samples. Then, based on the latest training samples, the parameters of the optimization model are re-optimized to obtain the latest optimization model.
[0102] Specifically, when the IoT system updates the normal operating data of the second target elevator, this new data can be added to the training samples to form a training set containing the latest normal operating data. In this way, the training samples can better reflect the current operating status of the elevator, thereby improving the adaptability and accuracy of the model.
[0103] Next, based on the latest training samples, the parameters of the optimized model can be readjusted and optimized. By re-optimizing the model parameters, it can be ensured that the model can effectively capture the feature changes of the elevator under different operating states, thereby generating a more accurate feature representation.
[0104] Through this process, the resulting optimized model will better reflect the current normal operation of the elevator and adapt to new data changes, thereby improving the model's effectiveness and stability in practical applications.
[0105] In one possible implementation, removing the trained projection head from the optimized model to obtain the feature encoder includes:
[0106] The feature encoder is obtained by removing the trained projection head from the latest optimized model.
[0107] In one possible implementation, the presence of anomalies in the first target elevator can be determined by comparing each indicator in the data to be detected with those in the control dataset one by one. The specific steps are as follows:
[0108] Runtime anomaly detection: The runtime of each instance in the test data is compared with the runtime of each instance in the control dataset. If any runtime in the test data is significantly greater than or less than any runtime in the control dataset, the first target elevator is considered to have a runtime anomaly.
[0109] Anomaly detection of door opening / closing frequency: The number of door openings / closings in the test data is compared with the number of door openings / closings in the control dataset. If the number of door openings / closings in the test data is greater than or less than any number of door openings / closings in the control dataset, the elevator is considered to have an anomaly in the number of door openings / closings.
[0110] Anomaly detection in running distance: The running distance of each instance in the test data is compared with each running distance in the control dataset. If any running distance in the test data is significantly greater or less than any running distance in the control dataset, the elevator is considered to have an anomaly in running distance.
[0111] Detection of abnormal wire rope bending counts: The number of bends in each wire rope in the test data is compared with the number of bends in each wire rope in the control dataset. If the number of bends in the test data is greater than or less than any bend count in the control dataset, the elevator is considered to have an abnormal number of wire rope bends.
[0112] Anomaly detection of passenger numbers: The passenger numbers in each instance of the data to be detected are compared with the passenger numbers in the control dataset. If the passenger numbers in the data to be detected are greater than or less than any passenger number in the control dataset, the elevator is considered to have an anomaly in passenger numbers.
[0113] By using this method of item-by-item comparison and analysis, the system can meticulously identify whether there are any abnormalities in multiple operating indicators of the elevator.
[0114] See Figure 3 , Figure 3 This is a schematic diagram of the structure of an elevator claims anomaly detection device provided in an embodiment of this application. Figure 3 As shown, the elevator claims anomaly detection device includes:
[0115] The first acquisition unit 301 is used to acquire, in response to a claim request from the first target elevator, the test data and the comparison dataset of the first target elevator from the Internet of Things; the test data includes the latest original test data of the first target elevator; the comparison dataset includes multiple original normal comparison data of the first target elevator; the original data in the test data and the comparison dataset include the first target elevator's running time, number of door openings and closings, running distance, number of wire rope bends, and number of passengers carried each time on a certain day;
[0116] The second acquisition unit 302 is used to acquire a benchmark vector based on the comparison dataset;
[0117] The input unit 303 is used to input the data to be detected into a pre-trained feature encoder to obtain the vector to be detected;
[0118] Anomaly identification unit 304 is used to determine that the first target elevator is abnormal if the distance between the vector to be detected and the reference vector is greater than a distance threshold.
[0119] The anomaly analysis unit 305 is used to perform anomaly analysis on the first target elevator that has anomalies based on the control dataset and the data to be detected, and to obtain the analysis results.
[0120] The reminder sending unit 306 is used to send the analysis results to the user and remind the user that the claim for the first target elevator is reasonable.
[0121] In one possible implementation, the device further includes:
[0122] A construction unit is used to construct training samples, which include multiple positive sample pairs. Each positive sample pair includes original sample data and augmented sample data. The original sample data includes the normal operation data of the second target elevator on a certain day. The normal operation data includes the duration of each run, the number of door openings and closings, the distance traveled each time, the number of wire rope bends each time, and the number of passengers carried each time. The augmented sample data is obtained by data augmentation based on the original sample data.
[0123] The first parameter optimization unit is used to take the training samples as input and iteratively optimize the parameters of the initial model using the NT-Xent loss function until the training cutoff condition is met, thus obtaining the optimized model; the initial model includes an initial encoder and an initial projection head;
[0124] The removal unit is used to remove the trained projection head from the optimized model to obtain the feature encoder.
[0125] In one possible implementation, the device further includes:
[0126] The sample update unit is used to add the new normal operation data of the second target elevator to the training sample when the Internet of Things updates the new normal operation data of the second target elevator to obtain the latest training sample.
[0127] The second parameter optimization unit is used to re-optimize the parameters of the optimization model based on the latest training samples to obtain the latest optimization model.
[0128] In one possible implementation, the removal unit is specifically used for:
[0129] The feature encoder is obtained by removing the trained projection head from the latest optimized model.
[0130] In one possible implementation, the second acquisition unit 302 is specifically used for:
[0131] The comparison dataset is input into the feature encoder to obtain multiple feature embedding vectors;
[0132] The reference vector is obtained by averaging the multiple feature embedding vectors.
[0133] In one possible implementation, the step of performing anomaly analysis on the first target elevator exhibiting anomalies based on the control dataset and the data to be detected, and obtaining analysis results, includes:
[0134] The runtime of each run in the data to be detected is compared with the runtime of each run in the control dataset. If the runtime of each run in the data to be detected is greater than or less than any runtime of each run in the control dataset, it is determined that the first target elevator has an abnormal runtime.
[0135] The number of door openings and closings in the data to be detected is compared with each number of door openings and closings in the control dataset. If the number of door openings and closings in the data to be detected is greater than or less than any number of door openings and closings in the control dataset, then it is determined that the first target elevator has an abnormal number of door openings and closings.
[0136] The running distance of each time in the data to be detected is compared with the running distance of each time in the control dataset. If the running distance of each time in the data to be detected is greater than or less than any running distance of each time in the control dataset, it is determined that the first target elevator has an abnormal running distance.
[0137] The number of times each door opens and closes in the data to be detected is compared with the number of times each door opens and closes in the control dataset. If the runtime of each run in the data to be detected is greater than or less than any one of the runtimes in the control dataset, then it is determined that the first target elevator has an abnormal runtime.
[0138] The number of times the wire rope is bent in each test data is compared with the number of times the wire rope is bent in each test data in the control dataset. If the number of times the wire rope is bent in each test data is greater than or less than any one of the numbers of times the wire rope is bent in the control dataset, it is determined that the first target elevator has an abnormal number of wire rope bends.
[0139] The number of passengers in each trip in the data to be detected is compared with the number of passengers in each trip in the control dataset. If the number of passengers in each trip in the data to be detected is greater than or less than any number of passengers in each trip in the control dataset, it is determined that the first target elevator has an abnormal running time.
[0140] In addition, this application embodiment also provides an elevator claim anomaly detection device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the elevator claim anomaly detection method as described above.
[0141] In addition, this application embodiment also provides a computer-readable storage medium storing instructions that, when executed on a terminal device, cause the terminal device to perform the elevator claim anomaly detection method as described above.
[0142] This application, by combining IoT technology and a pre-trained feature encoder, achieves real-time automated processing of elevator operation data, thus eliminating the need for manual verification in the claims process. By comparing the data to be tested with a normal control dataset, this application can accurately identify whether there are abnormal malfunctions in the elevator and quickly determine the reasonableness of the claim request, avoiding the cumbersome on-site inspections and multiple confirmation steps in traditional claims methods. This not only significantly shortens the claims cycle and reduces the time and labor costs associated with manual verification, but also improves the efficiency and accuracy of claims processing. Simultaneously, automated anomaly analysis using data from the IoT avoids biases and errors in human judgment, ensuring more objective and fair test results and significantly reducing the probability of claims errors.
[0143] It should be noted that the various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for the device embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiments. The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separate. The components indicated as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment solution according to actual needs. Those skilled in the art can understand and implement this without creative effort.
[0144] The above description is merely one specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A method for detecting abnormalities in elevator claims, characterized in that, The method includes: In response to a claim request from the first target elevator, the system obtains the test data and comparison dataset of the first target elevator from the Internet of Things (IoT). The test data includes the latest raw test data of the first target elevator. The comparison dataset includes multiple raw normal comparison data of the first target elevator. The raw data in the test data and the comparison dataset include the first target elevator's running time, number of door openings and closings, running distance, number of wire rope bends, and number of passengers carried per day. A baseline vector is obtained based on the aforementioned comparison dataset, and the data to be detected is input into a pre-trained feature encoder to obtain the vector to be detected. If the distance between the vector to be detected and the reference vector is greater than the distance threshold, then it is determined that the first target elevator is abnormal; Based on the comparison dataset and the data to be detected, an anomaly analysis is performed on the first target elevator that exhibits anomalies, and the analysis results are obtained. The analysis results are sent to the user, who is then reminded that the claim for the first target elevator is reasonable.
2. The method according to claim 1, characterized in that, The training process of the feature encoder includes: Training samples are constructed, comprising multiple positive sample pairs. Each positive sample pair includes original sample data and augmented sample data. The original sample data includes the normal operation data of the second target elevator on a certain day, including the duration of each run, the number of door openings and closings, the distance traveled, the number of wire rope bends, and the number of passengers carried each time. The augmented sample data is obtained by augmenting the original sample data. The training samples are used as input, and the parameters of the initial model are iteratively optimized using the NT-Xent loss function until the training cutoff condition is met, thus obtaining the optimized model; the initial model includes an initial encoder and an initial projection head; Remove the trained projection head from the optimized model to obtain the feature encoder.
3. The method according to claim 2, characterized in that, The method further includes: When the Internet of Things updates the new normal operation data of the second target elevator, the new normal operation data is added to the training sample to obtain the latest training sample; The parameters of the optimization model are re-optimized based on the latest training samples to obtain the latest optimization model; The step of removing the trained projection head from the optimized model to obtain the feature encoder includes: The feature encoder is obtained by removing the trained projection head from the latest optimized model.
4. The method according to claim 1, characterized in that, The process of obtaining the baseline vector based on the reference dataset includes: The comparison dataset is input into the feature encoder to obtain multiple feature embedding vectors; The reference vector is obtained by averaging the multiple feature embedding vectors.
5. The method according to claim 1, characterized in that, Based on the control dataset and the data to be detected, anomaly analysis is performed on the first target elevator exhibiting anomalies, and the analysis results are obtained, including: The runtime of each run in the data to be detected is compared with the runtime of each run in the control dataset. If the runtime of each run in the data to be detected is greater than or less than any runtime of each run in the control dataset, it is determined that the first target elevator has an abnormal runtime. The number of door openings and closings in the data to be detected is compared with each number of door openings and closings in the control dataset. If the number of door openings and closings in the data to be detected is greater than or less than any number of door openings and closings in the control dataset, then it is determined that the first target elevator has an abnormal number of door openings and closings. The running distance of each time in the data to be detected is compared with the running distance of each time in the control dataset. If the running distance of each time in the data to be detected is greater than or less than any running distance of each time in the control dataset, it is determined that the first target elevator has an abnormal running distance. The number of times each door opens and closes in the data to be detected is compared with the number of times each door opens and closes in the control dataset. If the runtime of each run in the data to be detected is greater than or less than any one of the runtimes in the control dataset, then it is determined that the first target elevator has an abnormal runtime. The number of times the wire rope is bent in each test data is compared with the number of times the wire rope is bent in each test data in the control dataset. If the number of times the wire rope is bent in each test data is greater than or less than any one of the numbers of times the wire rope is bent in the control dataset, it is determined that the first target elevator has an abnormal number of wire rope bends. The number of passengers in each trip in the data to be detected is compared with the number of passengers in each trip in the control dataset. If the number of passengers in each trip in the data to be detected is greater than or less than any number of passengers in each trip in the control dataset, it is determined that the first target elevator has an abnormal running time.
6. An elevator claims anomaly detection device, characterized in that, The device includes: The first acquisition unit is configured to, in response to a claim request from the first target elevator, acquire the test data and the comparison dataset of the first target elevator from the Internet of Things; the test data includes the latest original test data of the first target elevator; the comparison dataset includes multiple original normal comparison data of the first target elevator; the original data in the test data and the comparison dataset include the first target elevator's running time, number of door openings and closings, running distance, number of wire rope bends, and number of passengers carried per day. The second acquisition unit is used to acquire a reference vector based on the comparison dataset and input the data to be detected into a pre-trained feature encoder to obtain the vector to be detected. An anomaly detection unit is used to determine that the first target elevator is abnormal if the distance between the vector to be detected and the reference vector is greater than a distance threshold. An anomaly analysis unit is used to perform anomaly analysis on the first target elevator that has an anomaly based on the control dataset and the data to be detected, and to obtain the analysis results. The notification unit is used to send the analysis results to the user and remind the user that the claim for the first target elevator is reasonable.
7. The apparatus according to claim 6, characterized in that, The device further includes: A construction unit is used to construct training samples, which include multiple positive sample pairs. Each positive sample pair includes original sample data and augmented sample data. The original sample data includes the normal operation data of the second target elevator on a certain day. The normal operation data includes the duration of each run, the number of door openings and closings, the distance traveled each time, the number of wire rope bends each time, and the number of passengers carried each time. The augmented sample data is obtained by data augmentation based on the original sample data. The first parameter optimization unit is used to take the training samples as input and iteratively optimize the parameters of the initial model using the NT-Xent loss function until the training cutoff condition is met, thus obtaining the optimized model; the initial model includes an initial encoder and an initial projection head; The removal unit is used to remove the trained projection head from the optimized model to obtain the feature encoder.
8. The apparatus according to claim 7, characterized in that, The device further includes: The sample update unit is used to add the new normal operation data of the second target elevator to the training sample when the Internet of Things updates the new normal operation data of the second target elevator to obtain the latest training sample. The second parameter optimization unit is used to re-optimize the parameters of the optimization model based on the latest training samples to obtain the latest optimization model; The removal unit is specifically used for: The feature encoder is obtained by removing the trained projection head from the latest optimized model.
9. An elevator claims anomaly detection device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the elevator claim anomaly detection method as described in any one of claims 1-5.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed on a terminal device, cause the terminal device to perform the elevator claim anomaly detection method as described in any one of claims 1-5.