Internet of things based method and system for monitoring whole life cycle of rental equipment

Abstract

The application provides a kind of based on Internet of Things lease equipment whole life cycle monitoring method and system, belong to equipment operation monitoring technical field, this method obtains the real-time state data of lease equipment in the running process by multi-source sensor;Real-time state data is extracted for features, obtain a variety of feature vectors, and construct multidimensional feature vector set;Multidimensional feature vector set is input into random forest model, and whether lease equipment is in abnormal state is judged in combination with space-time clustering algorithm;If yes, then in combination with abnormal detection result, historical operation data and baseline model, the fault of lease equipment is predicted, and the fault probability value of lease equipment is obtained;Based on position data, the future position of lease equipment is predicted, and the electronic fence boundary is set for lease equipment, if the future position exceeds the electronic fence boundary, the early warning notice of track deviation is generated, and the system finally realizes the visual presentation of track and fault by heat map mode, improves the intelligentization and management accuracy of equipment operation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of equipment operation and maintenance monitoring, in particular to a leasing equipment whole life cycle monitoring method and system based on Internet of Things, and belongs to the technical field of industrial Internet of Things, intelligent fault detection and track management system. BACKGROUND

[0002] With the development of industrial digitization and intelligent operation and maintenance, the demand for remote monitoring of leasing assets such as engineering machinery, energy equipment and logistics equipment is increasing. Traditional equipment state monitoring methods rely on single physical quantity (such as temperature, vibration or pressure) for fixed-point detection, which cannot fully reflect the evolution trend of equipment operation state and is prone to false positives or false negatives. In recent years, the application of Internet of Things and edge computing has promoted the development of multi-source heterogeneous sensor technology, and equipment state data can be continuously collected in time and space dimensions. However, how to fuse and analyze multi-dimensional time series data such as vibration, temperature, pressure and position is still a core technical difficulty in current engineering practice.

[0003] In terms of anomaly detection, traditional methods rely on static threshold setting and cannot adapt to different equipment types and dynamic working conditions, and cannot adjust the judgment criteria in real time. In terms of fault prediction, existing technologies mostly use a single model to perform trend regression on historical data, ignoring the long-term and short-term dependencies of state data, resulting in insufficient prediction accuracy. In terms of track management, a static electronic fence is often used for boundary crossing judgment, but this method cannot dynamically reflect the adjustment of the working boundary and cannot identify the impassable areas in the working area, which has the risk of misjudgment. In addition, existing monitoring systems generally lack the ability to spatially visualize the warning results, making it difficult for operation and maintenance personnel to timely perceive the abnormal areas and fault trends of the equipment.

[0004] Therefore, how to construct a monitoring method that combines multi-source data fusion, dynamic anomaly identification, intelligent fault prediction and spatial deviation judgment for leasing equipment application scenarios has become a technical problem to be solved by the present application. SUMMARY

[0005] The present application provides a leasing equipment whole life cycle monitoring method and system based on Internet of Things, which aims to solve the technical problems of existing technologies, such as difficulty in fusing and analyzing multi-source sensor data, static anomaly detection, inaccurate fault prediction, distorted track deviation judgment and insufficient result visualization.

[0006] In a first aspect, the present application provides a leasing equipment whole life cycle monitoring method based on Internet of Things, which comprises:

[0007] S1, monitoring the operation process of the leasing equipment by a multi-source sensor to obtain real-time state data, the real-time state data comprising position data, vibration data, pressure data and temperature data;

[0008] S2, feature extraction is performed on the real-time state data to obtain a plurality of feature vectors of the rental equipment, and a multi-dimensional feature vector set is constructed;

[0009] S3, inputting the multi-dimensional feature vector set into a random forest model and combining a space-time clustering algorithm to determine whether the rental equipment is in an abnormal state;

[0010] S4, if yes, combining an abnormal detection result, historical operation data and a preset baseline model to perform fault prediction on the rental equipment to obtain a fault probability value of the rental equipment;

[0011] S5, predicting a future position of the rental equipment based on the position data, setting an electronic fence boundary for the rental equipment, if the future position exceeds the electronic fence boundary, generating a warning notification of track deviation, and pushing the warning information to a management platform.

[0012] Further, the feature extraction on the real-time state data comprises:

[0013] The feature extraction on the real-time state data is based on a convolutional neural network model, and the convolutional neural network model comprises a deep feature fusion layer composed of a plurality of convolution and pooling structures, which is used to extract a frequency mode of the vibration data, a mutation mode of the pressure data and a periodic mode of the temperature data.

[0014] Further, before inputting the multi-dimensional feature vector set into the random forest model, it comprises:

[0015] S31, continuously sampling each type of feature data in the multi-dimensional feature vector set based on a sliding time window;

[0016] S32, calculating the mean and standard deviation of each type of feature data within a preset window to construct a dynamic statistical feature set;

[0017] S33, when the deviation of newly collected feature data relative to the mean within the corresponding window exceeds a predetermined abnormal triggering threshold, determining that the feature data is an abnormal warning point;

[0018] S34, inputting the abnormal warning result as a priori judgment of the random forest model.

[0019] Further, the determination of whether the rental equipment is in an abnormal state comprises:

[0020] S41, inputting the multi-dimensional feature vector set into a trained random forest model, and the trained random forest model is trained by a training data set containing historical operation data and corresponding state labels of the rental equipment at each use stage;

[0021] S42: The random forest model performs preliminary classification processing based on the input feature data and outputs the predicted probability of each state category;

[0022] S43: Label the data at the time points where anomalies occur based on the predicted probability;

[0023] S44: Based on the spatiotemporal clustering algorithm, aggregate analysis is performed on the data points marked as anomalies in the spatial and temporal dimensions to identify whether there are group, regional or phased abnormal behaviors and obtain clustering results;

[0024] S45: Based on the clustering results, anomaly patterns are formed in spatially continuous regions or time periods;

[0025] S46: Compare the abnormal pattern with the common behavior patterns of the rental equipment in historical data, calculate the similarity, and if the similarity is greater than a preset first threshold, mark it as an abnormal state.

[0026] Furthermore, the baseline model records the health parameter ranges of the leased equipment at different operating stages, which are used to identify trend changes in the time series.

[0027] Furthermore, fault prediction for leased equipment includes:

[0028] S61: Use the anomaly detection results as explicit state variables in the input features of the LSTM model to identify equipment malfunctions in the short term;

[0029] S62: Slice the historical operation data of the leased equipment into segments using a fixed time window to construct a multi-dimensional time-series input matrix. Each row in the matrix corresponds to a time segment, and each column corresponds to a different perceptual feature or state label.

[0030] S63: Input the time series input matrix into the LSTM model, and the LSTM model extracts the dynamic change features of the leased equipment during long-term operation;

[0031] S64: The comprehensive feature vector formed by combining the anomaly detection results, the dynamic change characteristics of historical operating data and the baseline model is input into the LSTM model. The LSTM automatically captures the abnormal trends and potential changes in the long-term operation of the equipment through its gating mechanism and memory function, thereby generating the hidden state vector.

[0032] S65: Based on the output of the LSTM model, the dynamic change characteristics of the leased equipment are compared with the health parameter range in the baseline model to calculate the failure risk of the equipment during the prediction period.

[0033] Further, the electronic fence boundary is composed of a dynamically generated polygonal geometry constraint by a cloud server, the dynamically generated polygonal geometry constraint is a semantic fence constructed based on high-precision map data, and can be adjusted in real time according to an actual work area.

[0034] Further, the semantic fence constructed based on high-precision map data comprises:

[0035] S81: constructing a rough boundary area based on position data and historical work data of the rental equipment;

[0036] S82: in the rough boundary area, extracting multi-dimensional space constraint factors in the area based on elevation data and image data of a work area where the rental equipment is located, the multi-dimensional space constraint factors comprising an elevation gradient, a slope, and shelter information;

[0037] S83: performing pixel-level classification and identification on the multi-dimensional space constraint factors, identifying an impassable area of the rental equipment and obstacle information, and marking as a restricted area as a space boundary limitation condition of the semantic fence;

[0038] S84: analyzing boundary information of the obstacle data, and generating a restriction polygon allowing the rental equipment to move in combination with historical running behaviors of the rental equipment;

[0039] S85: the semantic fence supports a nested structure, integrates a plurality of the restricted areas into a boundary of the restriction polygon, forms the nested structure, and obtains the semantic fence.

[0040] Further, after S5, comprising:

[0041] S6, according to trajectory deviation data in the early warning notification, calculating a deviation degree of a trajectory point from the electronic fence boundary by using Euclidean distance, and displaying trajectory deviation distribution and failure probability distribution on a visual interface by using a heat map rendering technology, to realize visual monitoring of the rental equipment in a whole life cycle;

[0042] S7, generating a trajectory deviation distribution map and a failure probability distribution map of the rental equipment by using heat map rendering, to assist a management platform in visual analysis.

[0043] In a second aspect, the application provides a rental equipment whole life cycle monitoring system based on Internet of Things, the system comprising:

[0044] An acquisition module is configured to monitor a running process of the rental equipment by using a multi-source sensor, and acquire real-time state data, the real-time state data comprising position data, vibration data, pressure data, and temperature data;

[0045] An extraction module is configured to extract features of the real-time state data, obtain a plurality of feature vectors of the rental equipment, and construct a multi-dimensional feature vector set;

[0046] An anomaly judgment module is configured to input the multi-dimensional feature vector set into a random forest model and determine whether the rental equipment is in an abnormal state by combining a spatio-temporal clustering algorithm;

[0047] A fault diagnosis module is configured to, if yes, combine an abnormal detection result, historical operation data and a preset baseline model to perform fault prediction on the rental equipment and obtain a fault probability value of the rental equipment;

[0048] A position prediction module is configured to predict a future position of the rental equipment based on the position data, set an electronic fence boundary for the rental equipment, generate a pre-warning notification of trajectory deviation if the future position exceeds the electronic fence boundary, and push the pre-warning information to a management platform.

[0049] The technical scheme provided by the embodiment of the present application can include the following beneficial effects:

[0050] 1. The structural fusion and deep feature extraction of multi-source heterogeneous operation data are realized, and the expression ability and information utilization rate of the equipment state data are effectively improved.

[0051] 2. The working condition adaptability and sensitivity of abnormal recognition are enhanced by the dynamic threshold mechanism based on the statistical characteristics of the sliding window, and the misjudgment problem caused by the fixed threshold is avoided.

[0052] 3. The time continuity and spatial consistency recognition of abnormal detection are realized by introducing the spatio-temporal features into classification and clustering, and the understanding ability of the model to the equipment operation mode is enhanced.

[0053] 4. The long short-term memory network is used to realize the dependent modeling of the state sequence, and the time sequence correlation and prediction accuracy of the fault probability prediction are improved.

[0054] 5. The Kalman filter and dynamic semantic fence are linked to realize the trajectory deviation pre-judgment based on the spatial topological relationship, and the high-risk behaviors such as equipment border crossing and mistaken entry into the obstacle area are effectively avoided.

[0055] 6. The Euclidean distance is used to quantify the trajectory deviation degree, and the heat map is used for dynamic rendering, so as to improve the identification intuitiveness and response efficiency of the operation and maintenance personnel to the pre-warning area and risk distribution. BRIEF DESCRIPTION OF DRAWINGS

[0056] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0057] Figure 1 An embodiment schematic diagram of the method for monitoring the whole life cycle of the rental equipment based on the Internet of Things in the embodiments of the present application;

[0058] Figure 2 A process diagram of the rental equipment from data collection to early warning notification in the embodiments of the present application;

[0059] Figure 3 A comprehensive schematic diagram of the abnormality identification of the rental equipment in the embodiments of the present application;

[0060] Figure 4 A performance comparison diagram of the fault prediction model in the embodiments of the present application;

[0061] Figure 5 An embodiment schematic diagram of the system for monitoring the whole life cycle of the rental equipment based on the Internet of Things in the embodiments of the present application. DETAILED DESCRIPTION

[0062] The embodiments of the present application provide a method and system for monitoring the whole life cycle of rental equipment based on the Internet of Things. The terms "first", "second", "third", "fourth" and the like (if any) in the specification and claims of the present application and the above drawings are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the term "comprising" or "having" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0063] For the convenience of understanding, the specific process of the embodiments of the present application will be described below. Please refer to Figure 1 An embodiment of the method for monitoring the whole life cycle of the rental equipment based on the Internet of Things in the embodiments of the present application includes:

[0064] Step S1, monitoring the running process of the rental equipment through the multi-source sensor, acquiring real-time state data, the real-time state data including position data, vibration data, pressure data and temperature data.

[0065] Specifically, in the traditional management of rental equipment, real-time monitoring and fault warning of the equipment rely on manual inspection or single sensor data, which often fails to discover the equipment failure in time when it occurs, affecting the use efficiency of the equipment and the maintenance cost. With the development of Internet of Things technology, it is possible to use multi-source sensor data (such as vibration, temperature, pressure, etc.) to monitor the equipment in real time. However, existing technologies often fail to comprehensively analyze the historical operation data, environmental data and real-time state data of the equipment, lack effective algorithm models for accurate prediction of equipment failure, and cannot real-time predict the location of the equipment and manage the intelligent fence. Therefore, the present application proposes a method for how to monitor the rental equipment throughout its life cycle based on multi-source data and timely warn of failure.

[0066] Firstly, the rental equipment in the present application includes but is not limited to construction equipment (such as tower cranes, excavators, concrete mixers, etc.), agricultural equipment (such as tractors, harvesters, etc.), transportation equipment (such as forklifts, stackers, etc.) and energy equipment (such as generator sets, pump stations, etc.).

[0067] Taking heavy-duty excavating equipment in an engineering work scene as an example, in the running process of the excavating equipment, the multi-source sensor modules arranged on the excavating equipment are used to collect real-time state data, vibration data is obtained by an accelerometer or a vibration sensor to monitor the vibration state of the tower; temperature data is collected by a thermistor sensor at a set sampling frequency to obtain continuous time series fluctuation values, representing the thermal change trend of the equipment environment or core components; pressure data is collected by a pressure sensor, representing the dynamic pressure value inside the hydraulic system or pneumatic system, as an index reflecting the load change and fluid control state; position data is obtained by a GPS sensor or a wireless communication module in real time to obtain the accurate position of the excavating equipment, providing dynamic coordinate changes of the excavating equipment in space, which is used for subsequent trajectory modeling and space boundary judgment.

[0068] The aforementioned data types differ in physical dimensions, time scales, and change characteristics. To achieve unified processing and data fusion, this application introduces an edge computing unit as a data processing front-end during the data acquisition stage. The edge computing unit is deployed in the device's field controller or near-end communication node, possessing preliminary computing capabilities and embedded data preprocessing logic. After acquiring various types of raw data, the edge computing module performs corresponding feature preprocessing procedures for the signal types corresponding to different data. Vibration data is a frequency domain signal, temperature data is a time-series fluctuation signal, pressure data is a dynamically changing signal, and location data is a latitude and longitude sequence. For vibration signals, characteristic indicators such as the amplitude of the main peak in the spectrum and frequency domain energy density are extracted; for temperature and pressure signals, statistical features within the time window, such as fluctuation amplitude and rate of change, are extracted; for location signals, information on the current coordinate point, the difference between previous and subsequent values, and the change in trajectory direction is extracted. The processing results are cached in a structured form in the local storage unit at the edge. After the data stabilizes, it is sent to the cloud via a low-power wide-area network, improving transmission efficiency and real-time performance.

[0069] Even after edge processing, the original vibration data still suffers from high-frequency noise interference. To improve the stability and effectiveness of subsequent feature extraction, this invention employs wavelet denoising to perform multi-scale analysis of the vibration signal, separating the noise frequency band from the main frequency band. The noise-suppressed vibration spectrum is then obtained by reconstructing the signal. Subsequently, to eliminate inconsistencies in the dimensions, value ranges, and dynamic characteristics of different signal data, maximum and minimum value normalization is performed on all collected data to ensure they are expressed within a unified numerical range, forming a standardized sensor dataset. To guarantee stable uploading of edge-collected data and coordinated data transmission across multiple nodes, the data is sent to the cloud server via an NB-IoT communication module.

[0070] Step S2: Extract features from real-time status data to obtain multiple feature vectors of the leased equipment and construct a multi-dimensional feature vector set.

[0071] Specifically, such as Figure 2 The diagram illustrates the process from data acquisition to early warning notification for leased equipment. This application employs a convolutional neural network (CNN) to construct a feature extraction model, enabling unified modeling and salient feature extraction from multi-source data. The CNN model consists of multiple convolutional kernels sharing weights, each acting on a different feature channel. The vibration channel uses a small-scale convolutional kernel to extract the principal components in the frequency domain; the temperature channel uses a medium-scale convolution to identify periodic fluctuations; and the pressure channel uses a cross-scale convolutional structure to identify high-frequency abrupt changes and low-frequency range drift. To avoid single-channel features dominating the network's learning direction, a feature normalization mask and a cross-channel attention mechanism are introduced in the multi-channel fusion stage. Vibration, temperature, and pressure feature vectors are uniformly represented in the intermediate fusion layer, constructing a multi-dimensional feature vector set as the basic input for subsequent classification and prediction modules.

[0072] The multi-dimensional feature vector set retains the main dynamic change patterns under different signal dimensions, and expresses the mutual dependence between features through the parameter weights learned by the network, having time sequence correlation, physical logic integrity, and spatial coupling structure consistency.

[0073] In the process of mining equipment operation, the vibration frequency rises, the pressure fluctuates sharply, and the temperature abnormally rises. The feature weights of multiple channels in the fused feature vector rise together, and the model can instantly reflect that the state is a potential anomaly, thereby significantly improving the recognition accuracy and response timeliness, solving the problems of incomplete extraction based on single channel signal, difficult pattern separation, and system response lag in existing systems, and meeting the key requirements of complex data fusion modeling and fault early warning pre-identification in intelligent operation and maintenance scenarios.

[0074] Step S3, input the multi-dimensional feature vector set into the random forest model, and judge whether the rental equipment is in an abnormal state by combining the space-time clustering algorithm.

[0075] Specifically, as shown in Figure 3 Fig. 1 is a comprehensive schematic diagram of equipment anomaly identification. Based on the multi-dimensional feature vector set extracted in step S2, the running state of the rental equipment is subjected to abnormal identification processing. First, the threshold triggering mechanism set by the present application is used for initial screening. When any type of feature value, such as vibration frequency amplitude, temperature fluctuation gradient, or pressure change interval, exceeds the preset abnormal threshold range corresponding thereto, the feature of this type is marked as a suspected abnormal item and enters the state classification judgment process.

[0076] Secondly, the multi-dimensional feature vector set is input into the random forest model for abnormal state judgment. The random forest model constructs multiple weak classifiers with a tree structure integrated classification algorithm. The model will classify the state of the rental equipment according to the multi-dimensional feature vector, and obtain multiple state types, including “normal”, “mild abnormality”, and “severe abnormality”, etc. To improve the working condition adaptability and spatial generalization ability of the discrimination result, the feature data collection time is also introduced as a prior judgment of the random model for input at the input end of the model. The detailed content will be described later.

[0077] To enhance the accurate recognition ability of the abnormal discrimination result, the present application also introduces a space-time clustering algorithm, such as DBSCAN or density neighborhood clustering, to analyze the spatial distribution density of the short-time continuous triggered abnormal labels, identify whether there is a group, regional, or phased abnormal behavior, and form an abnormal risk report facing the geographical area. The specific implementation process will be described later.

[0078] Taking the large-scale engineering equipment rental scene as an example, if the pressure of a certain type of equipment in a certain area continuously appears frequent mutation accompanied by continuous temperature rise, the system first automatically identifies or uses a random forest model to identify that the equipment is in an abnormal load state, and judges whether it is a regional risk through a spatial clustering mechanism, so as to generate an operation and maintenance intervention suggestion in advance. This method effectively solves the problems of poor model generalization, inability to identify geographical environment related anomalies, high false positive rate and other problems in traditional anomaly judgment, has practicality and deployment efficiency for engineering application, and constitutes a key intelligent identification link in the equipment monitoring system of the application.

[0079] Step S4, if yes, combining the abnormal detection result, the historical operation data and the preset state baseline model, the fault probability value of the rental equipment is obtained.

[0080] Specifically, in order to realize dynamic modeling and fault risk prediction of the rental equipment operation state, the system constructs a sequence feature input set based on the identified abnormal detection result, long-term historical operation data and a preset state baseline model, and uses a long short-term memory network to model the time sequence dependence of the sequence. The abnormal detection result constitutes an explicit state variable in the input feature, which is used to mark the existing operation anomaly label in the short term; the historical operation data includes the original or extracted features such as vibration, temperature and pressure of the equipment at multiple time points; the preset baseline model records the health parameter interval of the equipment in different operation stages, which is used as a comparison scale to assist the network in judging the trend change in the time sequence.

[0081] In the data preparation process, the system slices the structured equipment operation data with a fixed time window to construct a multi-dimensional time sequence input matrix. Each row in the matrix corresponds to a time segment, and each column corresponds to a different sensing feature or state label. The input matrix is input into a long short-term memory (LSTM) structure, which learns the time evolution pattern and long-term dependence relationship between input features through a gating mechanism, automatically capturing the accumulation process or latent changes of abnormal trends in the equipment operation process. Since LSTM has the ability to remember long-term feature changes, the hidden state vector output by LSTM can effectively reflect the state transition law of the equipment in the continuous operation period.

[0082] In the training phase, the system adjusts the network parameters based on the historical labeled data, so that the LSTM can learn the contribution relationship of different state features to the future health state of the equipment; in the prediction phase, the current operation data and the LSTM model are used to generate the state evolution trajectory of the equipment in the prediction period, and the corresponding fault probability value is output. The probability value can be normalized to the [0, 1] interval through a Sigmoid mapping function, which is used to reflect the possibility of the equipment to fail or have a major deviation in the future under the current state.

[0083] In the whole process, the risk label output by the anomaly detection module is not only part of the network input, but also prior knowledge input for the prediction target of the LSTM model, effectively improving the accuracy of the model in fault precursor identification and enhancing the response ability of the system to weak trend signals. This technical means makes the algorithm features closely coupled with sensor perception, anomaly discrimination and early warning generation process, has structural integrity and functional synergy, and can effectively solve the problem of single-point anomaly in existing methods that is difficult to infer trend faults, and is suitable for reliability prediction and maintenance scheduling of rental equipment in multiple scenarios, with obvious technical advantages in improving equipment safety and operation efficiency. This part of the content as a key component of the overall technical scheme reflects the fusion and innovation of algorithm structure and technical means in the direction of fault prediction, and has significant engineering practical value and creativity.

[0084] Step S5, based on the position data, predicting the future position of the rental equipment, setting an electronic fence boundary for the rental equipment, if the future position exceeds the electronic fence boundary, generating a warning notification of trajectory deviation, and pushing the warning information to the management platform.

[0085] Specifically, to realize the linkage perception and risk warning of the running state and spatial behavior of the rental equipment, the system uses the Kalman filter algorithm to continuously and dynamically predict the future spatial trajectory of the equipment based on the device failure probability value generated in step S4 and the current collected position information. The Kalman filter takes the current latitude and longitude position, velocity vector and acceleration estimate as state input, constructs a prediction model through a state transition matrix, and iteratively optimizes the historical trajectory and sensor noise covariance information to output the predicted estimate value of the trajectory points in the future time period. The predicted trajectory result is expressed in time sequence form, forming a set of trajectory points composed of multiple latitude and longitude coordinates.

[0086] The trajectory point set and the electronic fence boundary data generated by the cloud server in real time are judged for spatial topological relationship. The electronic fence boundary is a polygonal geometric constraint constructed by a dynamic boundary generation algorithm according to the actual geographical range of the work area and the task requirements. This boundary can be dynamically adjusted according to changes in device work area, task type, etc. The system performs spatial containment detection on the predicted trajectory points one by one with the fence boundary. If one or more trajectory points fall outside the boundary, it is judged that the equipment has a trajectory deviation trend, which constitutes a trajectory anomaly warning condition.

[0087] Meanwhile, the system compares the current failure probability value of the device with a preset threshold value, and if the probability value exceeds the set upper limit of tolerance, it is considered that the device state has entered a high-risk interval of potential failure, constituting a failure abnormality early warning condition. When any type of early warning condition is triggered, the system automatically generates an early warning notification containing multiple structured fields such as the current failure state evaluation value, the predicted trajectory point offset, the offset direction, and the corresponding fence boundary point, and uploads the early warning data packet to the device management platform using the low-power Internet of Things protocol through the NB-IoT communication module. Based on the notification, the platform can visually display the current and future operation risks of the device and drive the dispatch, maintenance, and other system linkages to execute corresponding strategies.

[0088] The above technical solution forms a complete risk early warning judgment mechanism by fusing the failure trend modeling in the time domain and the trajectory dynamic prediction in the spatial domain. The Kalman filter model is logically associated with the device state probability output in the prediction stage, and the electronic fence boundary and the trajectory offset discrimination result support each other, with significant linkage. The algorithm features, physical state detection, and map boundary constraints form a functionally consistent structure, which can effectively solve the problems of failure recognition lag and trajectory offset early warning insensitivity in the prior art, significantly improve the forward-looking judgment ability of the monitoring system for the running state of the device in a complex working environment, and are suitable for intelligent management scenarios of mobile rental equipment.

[0089] In a specific embodiment, the feature extraction of the real-time state data in step S2 specifically includes the following steps:

[0090] The real-time state data is extracted based on a convolutional neural network model, and the convolutional neural network model includes a deep feature fusion layer composed of multiple convolution and pooling structures, which is used to extract the frequency pattern of vibration data, the mutation pattern of pressure data, and the periodic pattern of temperature data.

[0091] Specifically, the real-time state data is extracted based on a convolutional neural network model, and the convolutional neural network model includes a deep feature fusion layer composed of multiple convolution and pooling layers, which can automatically learn and extract key features in the data. The convolution layer extracts local features of the data through local perception regions, such as frequency patterns in vibration data, mutation patterns in pressure data, and periodic patterns in temperature data. Specifically, the convolution operation slides the filter over the input data, gradually extracting features of different scales in the time or spatial dimension. The parameters of these convolution layers are optimized through training data, so that the model can gradually learn the features of different signal patterns and retain them in the network for subsequent analysis.

[0092] For vibration data, the convolutional neural network captures periodic vibrations or frequency anomalies that may occur during the operation of the device, which are usually related to device failure or overload by learning their frequency patterns; for pressure data, the convolutional neural network focuses on capturing abrupt patterns, which refer to sharp changes in pressure signals within a short period of time, usually occurring at the moment when the device fails or abnormally. By extracting these abrupt patterns through the pooling layer, the network can also compress irrelevant data and improve its sensitivity to key abnormal patterns; for temperature data, the convolutional neural network can extract periodic patterns, i.e. the variation of temperature of the device when it is working normally. Temperature usually changes with the load of the device and has a certain periodicity.

[0093] The pooling layer in the convolutional neural network is used to further reduce the dimension of the data, helping to retain the most discriminative feature information. In the pooling process, CNN reduces unnecessary details and redundant information by downsampling the features of the local region, while improving the robustness of the network to deformation, noise or changes. The feature vector set extracted by the convolutional neural network is used for subsequent device state judgment and fault prediction. In this technical solution, the convolutional neural network can effectively extract important features related to device failure through layer-by-layer convolution and pooling operations, thereby providing accurate data support for subsequent anomaly detection and fault prediction.

[0094] In a specific embodiment, the step of inputting the multi-dimensional feature vector set into the random forest model in step S3 specifically comprises the following steps:

[0095] S31: continuously sampling each type of feature data in the multi-dimensional feature vector set based on a sliding time window;

[0096] S32: calculating the mean and standard deviation of each type of feature data within the preset window to construct a dynamic statistical feature set;

[0097] S33: determining that the feature data is an abnormal warning point when the deviation of the newly collected feature data from the corresponding window mean exceeds the predetermined abnormal triggering threshold;

[0098] S34: inputting the abnormal warning result as a priori judgment of the random forest model.

[0099] Specifically, to solve the problem of static abnormal threshold setting and poor adaptability to multiple working conditions in traditional equipment state recognition, a dynamic threshold self-adaptive judgment mechanism is introduced before the feature vector in step S3 is input into the random forest model. The monitoring feature data is continuously sampled through a sliding time window, and the dynamic mean and standard deviation indexes are constructed according to the historical statistical behavior of the same feature in the window interval, forming a set of dynamic statistical features with time sensitivity. The feature set is continuously updated in actual operation, maintaining the real-time representation ability of the equipment operation behavior at each time point, so that the system can adapt to the state change law of different equipment types, different load environments and different running periods.

[0100] On this basis, the deviation of the latest collected feature data from the mean value of the same feature in the corresponding sliding window is compared, and the deviation coefficient is constructed based on the standard deviation to determine whether it exceeds the abnormal trigger interval. If the deviation of the current data point from the statistical mean value in the window exceeds the preset dynamic abnormal coefficient threshold, the data point is marked as an abnormal warning point, and an abnormal label vector is constructed. The label information is injected as prior feature into the subsequent classification model, and the original physical feature, time stamp information and geographical location feature form a high-dimensional input vector.

[0101] The above input vector is used as the input feature of the random forest model. In the model training and prediction process, the dynamic abnormal prediction label is used as one of the explicit classification dimensions, which significantly improves the model's discrimination sensitivity to high-risk points and avoids blind judgment or misjudgment of non-stationary fluctuation data. At the same time, the time and space clustering algorithm is combined with the model output backend to aggregate the warning points with the same spatial or temporal distribution characteristics, realizing the recognition transformation from single-point abnormality to regional or trend fault, and improving the recognition efficiency of the system for weak latent abnormal behavior in complex operating environment.

[0102] This technical feature is functionally linked with the dynamic perception subsystem, feature fusion network and fault prediction module, and logically provides dynamic prior auxiliary judgment based on statistical behavior for model decision-making, forming a substantial improvement from static threshold mechanism to adaptive recognition mechanism. In the field of engineering equipment leasing, especially in the scene of mixed deployment of multiple models and heterogeneous operation in multiple geographical regions, the use of this structure can effectively identify abnormal behavior changes caused by differences in terrain, working conditions or task load, and output risk signals in a timely manner to provide data support for equipment maintenance and scheduling. As an embodiment of the deep integration of algorithm and monitoring technology, this feature substantially improves the intelligent level of equipment state perception.

[0103] In a specific embodiment, the step of judging whether the leased equipment is in an abnormal state in step S3 specifically includes the following steps:

[0104] S41: input the multi-dimensional feature vector set into the trained random forest model, which is trained by a training data set containing historical operation data of the rental equipment at each use stage and corresponding state labels;

[0105] S42: the random forest model performs preliminary classification processing according to the input feature data, and outputs the prediction probability of each state category;

[0106] S43: based on the prediction probability, the data at the time point where the anomaly occurs is labeled;

[0107] S44: based on the spatio-temporal clustering algorithm, the data points marked as abnormal are aggregated and analyzed in the spatial and temporal dimensions to identify whether there is a group, regional or stage abnormal behavior, and obtain a clustering result;

[0108] S45: based on the clustering result, an abnormal pattern of a spatially continuous region or a time period is formed;

[0109] S46: compare the abnormal pattern with the common behavior pattern of the rental equipment in the historical data, calculate the similarity, and if the similarity is greater than a preset first threshold, mark it as an abnormal state.

[0110] Specifically, for the accurate identification of abnormal behavior in equipment operation state monitoring, an enhanced input vector containing time labels and geographic coordinates is constructed based on the multi-dimensional feature vector set output by the convolutional neural network, the time dimension and location features are fused with the original working condition features, and the random forest model is used as the joint discriminant input. The random forest model divides the feature boundary in different dimensions through multiple decision tree sub-models, and judges whether the equipment is in an abnormal state in an ensemble learning manner.

[0111] To further enhance the model's ability to aggregate and identify abnormal behavior in multiple devices and multiple scenarios, a spatio-temporal clustering algorithm is introduced to process the abnormal label sequence based on the classification result. A density-based clustering method is used to perform spatial aggregation analysis on the data points marked as abnormal, and abnormal data with spatio-temporal concentration is classified into the same abnormal event set. Isolated points and outliers are eliminated to avoid false positives caused by occasional fluctuations. This processing mechanism can set an asynchronous processing module in the back end of the model output, automatically aggregate multiple abnormal judgment points within a time window and a spatial neighborhood range, form a spatially continuous region or a time period aggregation abnormal pattern, and mark it as a high-risk event.

[0112] In practical applications, if the temperature of the device in a certain geographical area suddenly rises and the pressure fluctuates within a certain period of time, and similar situations do not occur in other areas, the spatio-temporal clustering mechanism can accurately merge such behaviors into a specific regional abnormal event, effectively avoiding misidentification caused by device differences or the diversity of deployment scenarios. This mechanism is technically associated with a closed loop of the front-end feature extraction module, the random forest classifier, and the subsequent fault prediction logic, significantly improving the system's ability to identify and respond to local or phased abnormal states in complex working conditions. The above technical means, as the core abnormal identification component of the overall monitoring system, not only enhances the spatial generalization performance of the classification model, but also realizes the visualization of regional device health risks on the management platform through compatibility with the structured geographic information system, forming the basis for the intelligent perception and management linkage of the system.

[0113] In a specific embodiment, the baseline model records the health parameter range of the rental device in different operating stages, which is used to distinguish the trend changes in the time series.

[0114] Specifically, the baseline model is a mathematical model established based on the long-term historical operation data of the device, aiming to record and depict various operating parameters of the device in normal working conditions. These operating parameters include but are not limited to vibration, temperature, pressure and other key indicators. The baseline model defines a health range or health interval for each operating stage of the device (such as the startup stage, stable operation stage, overload stage, etc.). This health interval is usually based on historical data statistics of the device, reflecting the expected range of changes in these key parameters when the device is operating under normal conditions. By comparing with the real-time operating data of the device, the baseline model can help the system determine whether the state of the device has deviated, thereby identifying potential fault risks. For example, when the vibration frequency or temperature of the device exceeds the health interval defined by the baseline model, the system can issue a warning, indicating that the device may be in an abnormal state or about to fail.

[0115] In a specific embodiment, the fault prediction of the rental device in step S4 specifically includes the following steps:

[0116] S61: Take the abnormal detection result as an explicit state variable in the input feature of the LSTM model, and identify the short-term operating abnormality of the device;

[0117] S62: Slice the historical operating data of the rental device with a fixed time window to construct a multi-dimensional time series input matrix, where each row corresponds to a time segment and each column corresponds to a different sensing feature or state label;

[0118] S63: Input the time series input matrix into the LSTM model, and the LSTM model extracts the dynamic change features in the long-term operation of the rental device;

[0119] S64: The comprehensive feature vector formed by combining the abnormality detection result, the dynamic change characteristics of the historical operation data, and the baseline model is input into the LSTM model. The LSTM automatically captures the abnormal trend and potential change in the long-term operation of the device through its gating mechanism and memory function, thereby generating a hidden state vector.

[0120] S65: Based on the output of the LSTM model, the dynamic change characteristics of the rental device are compared with the health parameter interval in the baseline model, and the failure risk of the device in the prediction period is calculated.

[0121] Specifically, the abnormality detection result is one of the input features of the LSTM model, which is used to identify the abnormal operation of the device in the short term and reflects the health status of the device in a short period of time, such as temperature exceeding the standard, excessive vibration, etc. This result will be used as the explicit state variable of the LSTM model, providing an immediate and direct failure marker for the model when performing time series prediction.

[0122] The historical operation data (such as the vibration, temperature, and pressure of the device) is sliced with a fixed time window, and a multi-dimensional time series input matrix is constructed. The device state data in each time window will form the row data of the matrix. Each column corresponds to a different perception feature or state label, and all this information will collectively constitute a multi-dimensional input matrix. The LSTM model starts modeling the long-term operation state of the device by inputting the time series input matrix constructed above, captures the long-term dependence of time series data, and automatically identifies the dynamic change characteristics of the device in the long-term operation process, such as gradually increasing temperature and gradually increasing vibration. The LSTM model not only receives historical data, but also combines the abnormality detection result and the feature vector generated by the baseline model with the dynamic change characteristics of the device to form a comprehensive feature vector and input it into the LSTM. The LSTM automatically captures the long-term operation trend and potential change of the device through its gating mechanism and memory function, thereby generating a hidden state vector, which represents the "internal performance" of the device under the current state. When the dynamic characteristics of the device deviate from the health parameter interval defined by the baseline model, the LSTM model can calculate a failure probability value indicating the risk of device failure. The failure probability value is calculated through the output layer of the LSTM and calibrated with the baseline model to further improve the accuracy of the prediction. The output of the failure probability value can be normalized to the [0, 1] interval through the Sigmoid mapping function, representing the possibility of device failure. Based on this probability, the system can generate an early warning and provide an estimate of the time and severity of the failure.

[0123] As Figure 4As shown, it is a comparison chart of fault prediction model performance, which shows the performance comparison of different fault prediction models, including baseline model, LSTM model and prediction results of combination of baseline and LSTM model. The baseline model (dotted line) is used to show the fault prediction ability of the traditional method, and the prediction result fluctuates relatively large, and in some cases fails to accurately capture the time of fault occurrence. The LSTM model (solid line) shows the effect of fault prediction using the LSTM model. The LSTM model shows a smooth and accurate prediction, especially in capturing the fault occurrence point. The combination of baseline and LSTM model prediction (thick solid line) combines the advantages of LSTM and baseline model to provide more accurate fault prediction results. The combined model can more accurately identify the fault time and reduce errors. Therefore, the combined model is superior to the traditional baseline model and single LSTM model in accuracy and robustness, especially in identifying the time of fault occurrence, which can reduce false positives and improve the accuracy of prediction, provide reliable technical support for the operation and maintenance of rental equipment, reduce the occurrence of equipment failure, and improve the accuracy and efficiency of operation and maintenance.

[0124] In a specific embodiment, the electronic fence boundary is composed of a dynamically generated polygonal geometric constraint by a cloud server. The dynamically generated polygonal geometric constraint is a semantic fence constructed based on high-precision map data and can be adjusted in real time according to the actual work area.

[0125] Specifically, to achieve dynamic adaptation of boundary control and trajectory deviation detection of equipment operating space, the cloud server constructs a set of real-time electronic fence boundaries based on the current equipment task parameters, historical trajectory data and regional configuration instructions, through a polygon boundary generation algorithm. The boundary is composed of a polygonal geometric constraint composed of a plurality of two-dimensional latitude and longitude coordinate points in sequence. The algorithm extracts the activity area density, boundary coverage range and high-frequency access area distribution from the device historical work trajectory, and combines the passable area standard defined by the work type, to form a set of dynamic boundary constraints matched with the actual work environment. Each boundary polygon in the set has a unique identifier and is bound to the corresponding device task.

[0126] When constructing the fence boundary, the cloud analyzes the work task data issued by the device management platform, including task number, belonging plot, work period and work mode, and the work restriction conditions corresponding to the specified geographic area are parameterized in the fence construction logic. The generated polygon data structure not only contains boundary point coordinates, but also includes attribute fields such as region type, restriction level, buffer radius and other semantic information. The server delivers the above geometric constraint data to the device end in real time through the fence configuration interface, and establishes an association mapping between the task boundary and the device identifier in the database, and supports periodic refreshing and condition triggered reconstruction of the boundary.

[0127] In the process of device movement, the current positioning information collected can be projected in real time onto the two-dimensional plane defined by the above-mentioned fence boundary, and whether it is in the legal operation area is determined by calculating the spatial inclusion relationship between the current coordinates and the boundary polygon. At the same time, the trajectory point sequence output by the prediction module is also projected and checked point by point whether it violates the spatial restriction constituted by the fence boundary. If the trajectory point falls outside the boundary, the trajectory deviation type early warning logic is triggered. The dynamic construction capability of the above-mentioned electronic fence can adapt to the geographical differences and task change requirements of different devices in different operation scenarios. By combining the cloud parameter analysis capability, historical trajectory modeling capability and task semantic configuration capability, the generation of the fence boundary not only has geometric accuracy, but also embodies the operation context semantics, thereby significantly enhancing the spatial compliance judgment capability of the device running behavior.

[0128] In a specific embodiment, the semantic fence constructed based on high-precision map data specifically includes the following steps:

[0129] S81: constructing a rough boundary region based on the position data and historical operation data of the rental device;

[0130] S82: within the rough boundary region, extracting multi-dimensional space constraint factors including elevation gradient, slope and shelter information based on the elevation data and image data of the working area where the rental device is located;

[0131] S83: performing pixel-level classification and identification on the multi-dimensional space constraint factors, identifying the non-passable area of the rental device and obstacle information, and marking as a restricted area as a spatial boundary restriction condition of the semantic fence;

[0132] S84: analyzing the boundary information of the obstacle data, and generating a restriction polygon allowing the rental device to move in combination with the historical running behavior of the rental device;

[0133] S85: integrating multiple restricted areas into the boundary of the restriction polygon to form a nested structure and obtain a semantic fence.

[0134] Specifically, to improve the spatial constraint accuracy of device trajectory deviation judgment and enhance the semantic expression ability of early warning, the traditional electronic fence mechanism based on static geometric boundary is extended to a semantic fence mechanism constructed based on high-precision map data.

[0135] The construction process: first, the geographical space range corresponding to the rental equipment operation area calls the digital elevation model (DEM) data containing the terrain information and the DOM data, combines image recognition and geographical space analysis algorithm to identify the obstacle units with physical unattainable properties in the region, including buildings, mountains, walls and traffic separation belts, etc., to form a multi-class obstacle boundary data set in the spatial dimension. After the above obstacle boundary data is vectorized, a series of closed obstacle surface structures are generated, each obstacle surface has a unique number and type attribute, and is combined with the device historical operation trajectory polygon to form a composite boundary for limiting the future path accessibility of the device.

[0136] The composite boundary is defined as the main structure of the semantic fence, where the work allowed area is represented as the main polygon, and the obstacle surface is represented as the nested structure in the form of embedded polygon, forming a constraint boundary system with double expression ability of spatial limitation and semantic attribute. The system supports dynamic configuration and real-time refresh of the semantic fence at the task level. When the device is adjusted or the environment changes, the fence boundary can be automatically updated according to the latest map data and task template, ensuring that the fence structure is dynamically consistent with the actual working environment.

[0137] After the device state prediction module completes the future trajectory point calculation, the system performs spatial topological relationship analysis on the predicted trajectory point set and the above semantic fence structure, including point-surface inclusion relationship and point-boundary nearest distance calculation. When any trajectory point falls into the impassable area defined by the obstacle surface or has a significant deviation from the main boundary, the system determines that a trajectory anomaly has occurred. Such abnormal judgment results are written into the warning data structure, and the coordinate information of the offset point and the triggered obstacle surface identifier are attached, realizing semantic-level description and traceability record of the trajectory deviation event.

[0138] The semantic fence technology feature extends the static spatial limitation to a dynamic area model containing terrain and building semantic labels through fusion processing with map data, and effectively cooperates with device state data in trajectory prediction result processing, not only improving the practicability and accuracy of deviation identification, but also enhancing the semantic interpretation ability of the warning logic, thereby forming an organic cooperation of algorithm features and spatial modeling technology, playing a core role in the overall monitoring system, having reusability, engineering adaptability and significant technological innovation.

[0139] In a specific embodiment, after step S5, further comprising:

[0140] S6, according to the trajectory deviation data in the warning notification, using the Euclidean distance to calculate the deviation degree of the trajectory point and the electronic fence boundary, and through the heat map rendering technology to display the trajectory deviation distribution and the fault probability distribution on the visualization interface, realizing the whole life cycle visualization monitoring of the rental equipment;

[0141] S7, generating a lease equipment trajectory deviation distribution map and a fault probability distribution map through heat map rendering to assist the management platform in visual analysis.

[0142] Specifically, to improve the explainability of trajectory deviation and equipment state anomaly and the response efficiency of management, a visual monitoring module is further set after step S5 to build a comprehensive display mechanism with Euclidean distance analysis and heat map rendering as the core. After the cloud server identifies the deviated trajectory point based on the predicted trajectory and the semantic fence structure, the system performs nearest distance calculation on the trajectory point and the boundary line of the semantic fence, uses the Euclidean distance to measure the degree of trajectory deviation, and uses the value as a weight factor of spatial deviation intensity to build a heat map of trajectory deviation distribution. At the same time, in combination with the equipment fault probability value output by the LSTM model, the probability level corresponding to each time step is also used as a weight dimension to participate in the generation of heat map data, forming a fault probability distribution map in a two-dimensional space-time domain.

[0143] The two types of heat maps are embedded into the interactive interface of the management platform in a standard visualization layer rendering manner. The trajectory deviation map expresses the deviation intensity through different color gradients, and the fault probability map expresses the fault trend evolution through the density of node color blocks. The two maps can be superimposed or displayed in layers according to user selection, and support time axis dynamic playback function to realize synchronous dynamic monitoring of equipment state and path deviation. Through this technical means, the management platform can quickly identify the severity and time characteristics of the equipment deviation area, assist decision makers in identifying potential operation risks and developing intervention strategies, thereby forming a terminal visualization output mechanism that is closely coupled with the aforementioned prediction algorithm module in function, enhancing the real-time perception ability and response speed of the system as a whole to fault risks and spatial deviation, and having significant practical value and system optimization effect.

[0144] The above describes a kind of lease equipment full life cycle monitoring method based on Internet of Things in the embodiment of the present application, the following describes a kind of lease equipment full life cycle monitoring system based on Internet of Things in the embodiment of the present application, please refer to Figure 5 A kind of lease equipment full life cycle monitoring system based on Internet of Things in the embodiment of the present application one embodiment includes:

[0145] The acquisition module is used to monitor the running process of the lease equipment through the multi-source sensor, and obtain real-time state data, including position data, vibration data, pressure data and temperature data.

[0146] The extraction module is used to extract features from the real-time state data, obtain a plurality of feature vectors of the lease equipment, and construct a multi-dimensional feature vector set.

[0147] Anomaly judgment module, configured to input the multi-dimensional feature vector set into a random forest model, and determine whether the rental equipment is in an abnormal state in combination with a space-time clustering algorithm;

[0148] Fault diagnosis module, configured to, if yes, perform fault prediction on the rental equipment in combination with an abnormal detection result, historical operation data and a preset baseline model, and obtain a fault probability value of the rental equipment;

[0149] Position prediction module, configured to predict a future position of the rental equipment based on position data, set an electronic fence boundary for the rental equipment, generate a warning notification of track deviation if the future position exceeds the electronic fence boundary, and push the warning information to a management platform.

[0150] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described system, system and unit can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.

[0151] The integrated unit, if realized in the form of a software function unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on such understanding, the technical solutions of the present application essentially or say the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods described in various embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0152] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for monitoring the entire lifecycle of leased equipment based on the Internet of Things, characterized in that, The method includes: S1. Monitor the operation of the rental equipment through multi-source sensors to obtain real-time status data, including location data, vibration data, pressure data, and temperature data; S2. Extract features from the real-time status data to obtain multiple feature vectors of the rental equipment, and construct a multi-dimensional feature vector set; S3. Input the multidimensional feature vector set into the random forest model and combine it with a spatiotemporal clustering algorithm to determine whether the rental equipment is in an abnormal state; wherein, before inputting the multidimensional feature vector set into the random forest model, the method further includes: S31: continuously sampling each type of feature data in the multidimensional feature vector set based on a sliding time window; S32: calculating the mean and standard deviation of each type of feature data within a preset window to construct a dynamic statistical feature set; S33: when the deviation of newly collected feature data from the mean within the corresponding window exceeds a predetermined abnormal trigger threshold, the feature data is determined to be an abnormal warning point; S34: inputting the abnormal warning result as a priori judgment of the random forest model; The determination of whether the rental equipment is in an abnormal state includes: S41: inputting the multidimensional feature vector set into a trained random forest model, which is trained on a training dataset containing historical operating data of the rental equipment at each usage stage and corresponding state labels; S42: the random forest model performs preliminary classification processing based on the input feature data and outputs the predicted probability of each state category; S43: labeling the data at the time point where the abnormality occurred based on the predicted probability; S44: performing aggregation analysis on the data points marked as abnormal in the spatial and temporal dimensions based on a spatiotemporal clustering algorithm to identify whether there are group, regional, or phased abnormal behaviors and obtain clustering results; S45: forming abnormal patterns of spatially continuous regions or time periods based on the clustering results; S46: comparing the abnormal patterns with common behavior patterns of the rental equipment in historical data, calculating the similarity, and if the similarity is greater than a preset first threshold, then marking it as an abnormal state; S4. If so, combine the anomaly detection results, historical operating data, and the preset baseline model to predict the failure probability of the leased equipment and obtain the failure probability value of the leased equipment. The failure prediction of the leased equipment includes: S61: Using the anomaly detection results as explicit state variables in the input features of the LSTM model to identify short-term operational anomalies of the equipment; S62: Slicing the historical operating data of the leased equipment into segments using a fixed time window to construct a multi-dimensional time-series input matrix, where each row corresponds to a time segment and each column corresponds to different perceptual features or state labels; S63: Inputting the time-series input matrix into the LSTM model, which extracts the dynamic change features of the leased equipment during long-term operation; S64: Inputting the comprehensive feature vector formed by combining the anomaly detection results, the dynamic change features of historical operating data, and the baseline model into the LSTM model. The LSTM, through its gating mechanism and memory function, automatically captures abnormal trends and potential changes in the long-term operation of the equipment, thereby generating a hidden state vector; S65: Based on the output of the LSTM model, comparing the dynamic change features of the leased equipment with the health parameter range in the baseline model to calculate the failure risk of the equipment during the prediction period. S5. Based on the location data, predict the future location of the rental equipment, set an electronic fence boundary for the rental equipment, and if the future location exceeds the electronic fence boundary, generate a trajectory deviation warning notification and push the warning information to the management platform.

2. The method according to claim 1, characterized in that, Feature extraction of the real-time status data includes: Feature extraction is performed on the real-time state data based on a convolutional neural network model. The convolutional neural network model includes a deep feature fusion layer composed of multiple convolutional and pooling structures, which is used to extract the frequency pattern of the vibration data, the abrupt change pattern of the pressure data, and the periodic pattern of the temperature data.

3. The method according to claim 1, characterized in that, The baseline model records the health parameter ranges of the leased equipment at different operating stages, which are used to identify trend changes in the time series.

4. The method according to claim 1, characterized in that, The electronic fence boundary is composed of polygonal geometric constraints dynamically generated by a cloud server. These dynamically generated polygonal geometric constraints are semantic fences built based on high-precision map data and can be adjusted in real time according to the actual work area.

5. The method according to claim 4, characterized in that, Semantic fences built based on high-precision map data include: S81: Construct a coarse boundary region based on the location data and historical operation data of the leased equipment; S82: Within the coarse boundary area, based on the elevation data and image data of the working area where the rental equipment is located, extract the multi-dimensional spatial constraint factors within the area. The multi-dimensional spatial constraint factors include elevation gradient, slope, and occlusion information. S83: Perform pixel-level classification and recognition on the multidimensional spatial constraint factors, identify the areas that the rental equipment cannot pass through and the information on obstacles, and mark them as restricted areas as spatial boundary constraints of the semantic fence; S84: Parse the boundary information of the obstacle data, combine it with the historical operating behavior of the leased equipment, and generate a restriction polygon that allows the leased equipment to move. S85: The semantic fence supports a nested structure, integrating multiple restricted regions into the boundary of the restricted polygon to form the nested structure, thereby obtaining the semantic fence.

6. The method according to claim 1, characterized in that, Following S5, further additions include: S6. Based on the trajectory deviation data in the warning notification, the deviation between the trajectory point and the electronic fence boundary is calculated using Euclidean distance, and the trajectory deviation distribution and fault probability distribution are displayed on the visualization interface using heat map rendering technology, so as to realize the full life cycle visualization monitoring of the leased equipment. S7. Generate a trajectory deviation distribution map and a fault probability distribution map of the leased equipment through heat map rendering to assist the management platform in visual analysis.

7. An IoT-based full lifecycle monitoring system for leased equipment, used to implement the IoT-based full lifecycle monitoring method for leased equipment as described in any one of claims 1-6, characterized in that, The system includes: The acquisition module is used to monitor the operation of the rental equipment through multi-source sensors and acquire real-time status data, including location data, vibration data, pressure data and temperature data. The extraction module is used to extract features from the real-time status data, obtain multiple feature vectors of the rental equipment, and construct a multi-dimensional feature vector set. An anomaly detection module is used to input the multidimensional feature vector set into a random forest model and combine it with a spatiotemporal clustering algorithm to determine whether the rental equipment is in an abnormal state. Before inputting the multidimensional feature vector set into the random forest model, the module further includes: S31: continuously sampling each type of feature data in the multidimensional feature vector set based on a sliding time window; S32: calculating the mean and standard deviation of each type of feature data within a preset window to construct a dynamic statistical feature set; S33: when the deviation of newly collected feature data from the mean within the corresponding window exceeds a predetermined anomaly trigger threshold, determining that feature data is an anomaly warning point; S34: using the anomaly warning result as a priori judgment input into the random forest model; wherein, determining whether the rental equipment is in an abnormal state includes: S41: inputting the multidimensional feature vector set into the trained random forest model. The trained random forest model is trained on a training dataset containing historical operating data of the rental equipment at each stage of use and corresponding state labels; S42: The random forest model performs preliminary classification processing based on the input feature data and outputs the predicted probability of each state category; S43: The data at the time point where the anomaly occurred is labeled based on the predicted probability; S44: The data points labeled as anomalies are aggregated and analyzed in spatial and temporal dimensions based on a spatiotemporal clustering algorithm to identify whether there are group, regional, or phased abnormal behaviors and obtain clustering results; S45: Based on the clustering results, anomaly patterns of spatially continuous regions or time periods are formed; S46: The anomaly patterns are compared with common behavior patterns of the rental equipment in historical data, and the similarity is calculated. If the similarity is greater than a preset first threshold, it is marked as an abnormal state; The fault diagnosis module is used to determine if a fault is present. If so, it combines the anomaly detection results, historical operating data, and a preset baseline model to predict the fault probability of the leased equipment. The fault prediction process includes: S61: Using the anomaly detection results as explicit state variables in the input features of the LSTM model to identify short-term operational anomalies; S62: Slicing the historical operating data of the leased equipment into segments with a fixed time window to construct a multi-dimensional time-series input matrix. Each row in the matrix corresponds to a time segment, and each column corresponds to different perceptual features or state labels; S63: Inputting the time-series input matrix into the LSTM model, which extracts the dynamic change features of the leased equipment during long-term operation; S64: Inputting a comprehensive feature vector formed by combining the anomaly detection results, the dynamic change features of historical operating data, and the baseline model into the LSTM model. The LSTM, through its gating mechanism and memory function, automatically captures abnormal trends and potential changes in the long-term operation of the equipment, thereby generating a hidden state vector; S65: Based on the output of the LSTM model, comparing the dynamic change features of the leased equipment with the health parameter range in the baseline model to calculate the fault risk of the equipment during the prediction period. The location prediction module is used to predict the future location of the rental equipment based on the location data, set an electronic fence boundary for the rental equipment, and generate a trajectory deviation warning notification if the future location exceeds the electronic fence boundary, and push the warning information to the management platform.

Images