Adaptive optimization method and device of model, electronic equipment and readable medium

Abstract

The invention relates to a model self-adaptive optimization method and device, electronic equipment and a readable medium, and the method comprises the steps: obtaining a prediction model corresponding to target equipment, and enabling the prediction model to be obtained through training according to the historical fault data of the target equipment; in the process of performing fault prediction on the target equipment by using the prediction model, monitoring a prediction error of the prediction model and an operation condition of the target equipment; judging whether to trigger a model adjustment mechanism according to the prediction error and the operation condition; under the condition that a model adjustment mechanism is triggered, a corresponding target adjustment strategy is selected according to the size of the prediction error and the change of the operation condition, and the target adjustment strategy is a parameter adjustment strategy or a structure adjustment strategy; and adjusting the prediction model according to the target adjustment strategy to complete adaptive optimization of the prediction model. It is ensured that the model can continuously adapt to equipment state changes, and accurate prediction of equipment faults is achieved.

Description

Technical Field

[0001] This application relates to the field of fault prediction technology, and in particular to an adaptive optimization method, apparatus, electronic device, and readable medium for a model. Background Technology

[0002] In modern industrial production, the continuous and stable operation of equipment is directly related to the safety of the production line, product quality, and economic benefits. Traditional equipment maintenance methods mainly include two forms: periodic maintenance and reactive repair. While periodic maintenance can prevent sudden failures to some extent, it often leads to over-maintenance and unnecessary downtime because it fails to dynamically adjust according to the actual health status of the equipment, thereby increasing operation and maintenance costs and wasting resources. Reactive repair, on the other hand, is carried out only after serious equipment failures occur, which not only causes production interruptions and economic losses but may also trigger a chain reaction of equipment damage or even safety accidents.

[0003] Predicting equipment failures in advance is particularly important. Traditional prediction methods are usually based on a single signal (such as vibration or temperature) for fault diagnosis and prediction. They typically use a fixed model structure and static parameter configuration. As a result, the prediction model cannot continuously adapt to changes in equipment status, which leads to a gradual decrease in the prediction accuracy of the fault prediction model during long-term operation.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This application provides an adaptive optimization method, apparatus, electronic device, and readable medium for a model to solve the aforementioned technical problem that "the prediction model cannot continuously adapt to changes in the state of the equipment".

[0006] According to one aspect of the embodiments of this application, this application provides an adaptive optimization method for a model, comprising: acquiring a prediction model corresponding to a target device, wherein the prediction model is trained based on historical fault data of the target device; during the process of using the prediction model to predict faults in the target device, monitoring the prediction error of the prediction model and the operating conditions of the target device; determining whether to trigger a model adjustment mechanism based on the prediction error and the operating conditions; if the model adjustment mechanism is triggered, selecting a corresponding target adjustment strategy based on the magnitude of the prediction error and the change in the operating conditions, wherein the target adjustment strategy is a parameter adjustment strategy or a structural adjustment strategy; and adjusting the prediction model according to the target adjustment strategy to complete the adaptive optimization of the prediction model.

[0007] Optionally, before obtaining the prediction model corresponding to the target device, the method further includes generating the prediction model in the following manner: obtaining the operating parameters of the target device from multiple dimensions and extracting the parameter features of the operating parameters; determining the model architecture matching the target device and judging the first relationship type between the parameter features and the failure probability of the target device, wherein the first relationship type is a linear relationship, a nonlinear relationship, or a time-series dependency relationship; obtaining an algorithm mapping table and determining the first model algorithm corresponding to the first relationship type through the algorithm mapping table, wherein the algorithm mapping table includes multiple sets of correspondences between relationship types and model algorithms; constructing an initial model based on the first model algorithm and the model architecture; and training the initial model using the historical failure data of the target device as training data to obtain the prediction model.

[0008] Optionally, monitoring the prediction error of the prediction model and the operating condition of the target equipment includes: calculating the prediction error based on the output of the prediction model and the actual equipment status; calculating the real-time data distribution of the input data of the prediction model, and comparing the real-time data distribution with the initial data distribution to determine whether the operating condition has changed, wherein the initial data distribution is the parameter distribution in the historical fault data.

[0009] Optionally, the model adjustment mechanism is determined based on the prediction error and the operating conditions, including: if the prediction error is less than or equal to the error threshold and the operating conditions have not changed, the model adjustment mechanism is determined not to be triggered; if the prediction error is greater than the error threshold and / or the operating conditions have changed, the model adjustment mechanism is determined to be triggered.

[0010] Optionally, a corresponding target adjustment strategy can be selected based on the magnitude of the prediction error and the change in operating conditions, including: if the prediction error is greater than the error threshold and the operating conditions have not changed, then the parameter adjustment strategy is used as the target adjustment strategy; if the prediction error is greater than the error threshold and the operating conditions have changed, then the structural adjustment strategy is used as the target adjustment strategy.

[0011] Optionally, if the target adjustment strategy is a structure adjustment strategy, the prediction model is adjusted according to the target adjustment strategy, including at least one of the following methods: training multiple sub-models with different structures in parallel; calculating the prediction performance index of each sub-model under the current working condition; assigning dynamic weights to each sub-model according to each prediction performance index; and weighting and fusing the outputs of each sub-model according to the weights to generate the adjusted prediction model.

[0012] Optionally, after completing the adaptive optimization of the prediction model, the method further includes: using the prediction model to predict the fault of the target device, obtaining the predicted fault result output by the prediction model, wherein the predicted fault result includes the predicted fault probability, the predicted fault type, the faulty component, and the predicted fault time; determining the fault probability interval in which the predicted fault probability falls, and determining the warning level of the target device based on the fault probability interval; generating warning information corresponding to the warning level, and sending the warning information to the terminal, wherein the warning information includes the warning level and the predicted fault result.

[0013] According to another aspect of the embodiments of this application, this application provides an adaptive optimization apparatus for a model, comprising: an acquisition module for acquiring a prediction model corresponding to a target device, wherein the prediction model is trained based on historical fault data of the target device; a monitoring module for monitoring the prediction error of the prediction model and the operating condition of the target device during the process of using the prediction model to predict faults in the target device; a judgment module for judging whether to trigger a model adjustment mechanism based on the prediction error and the operating condition; a selection module for selecting a corresponding target adjustment strategy based on the magnitude of the prediction error and the change in the operating condition when the model adjustment mechanism is triggered, wherein the target adjustment strategy is a parameter adjustment strategy or a structural adjustment strategy; and an adjustment module for adjusting the prediction model according to the target adjustment strategy to complete the adaptive optimization of the prediction model.

[0014] According to another aspect of the embodiments of this application, this application provides an electronic device, including a memory, a processor, a communication interface and a communication bus. The memory stores a computer program that can run on the processor. The memory and the processor communicate with each other through the communication bus and the communication interface. When the processor executes the computer program, it implements the steps of the above method.

[0015] According to another aspect of the embodiments of this application, this application also provides a computer-readable medium having processor-executable non-volatile program code that causes the processor to perform the above-described method.

[0016] Compared with related technologies, the technical solutions provided in this application have the following advantages: This application provides an adaptive optimization method for a model, comprising: acquiring a prediction model corresponding to a target device, wherein the prediction model is trained based on historical fault data of the target device; during the process of using the prediction model to predict faults in the target device, monitoring the prediction error of the prediction model and the operating conditions of the target device; determining whether to trigger a model adjustment mechanism based on the prediction error and operating conditions; if the model adjustment mechanism is triggered, selecting a corresponding target adjustment strategy based on the magnitude of the prediction error and changes in operating conditions, wherein the target adjustment strategy is a parameter adjustment strategy or a structural adjustment strategy; and adjusting the prediction model according to the target adjustment strategy to complete the adaptive optimization of the prediction model. By monitoring the prediction error and operating conditions in real time, automatically selecting a parameter adjustment strategy or a structural adjustment strategy to adaptively optimize the prediction model, the method ensures that the model can continuously adapt to changes in device state and achieve accurate prediction of device faults. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the hardware environment for an optional adaptive optimization method for a model provided according to an embodiment of this application; Figure 2 This is a flowchart of an optional adaptive optimization method for a model provided according to an embodiment of this application; Figure 3 This is a schematic diagram of an optional system architecture for device data acquisition and transmission according to an embodiment of this application; Figure 4 This is a schematic diagram of an optional equipment fault early warning report provided according to an embodiment of this application; Figure 5 This is a schematic diagram of an optional bearing fault diagnosis and prediction system provided according to an embodiment of this application; Figure 6 This is a block diagram of an optional adaptive optimization device for a model according to an embodiment of this application; Figure 7 This is a schematic diagram of an optional electronic device structure provided in an embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] In the following description, the use of suffixes such as "module," "part," or "unit" to denote elements is solely for the purpose of illustration and has no specific meaning in itself. Therefore, "module" and "part" may be used interchangeably.

[0022] In modern industrial production, the continuous and stable operation of equipment is directly related to the safety of the production line, product quality, and economic benefits. Traditional equipment maintenance methods mainly include two forms: periodic maintenance and reactive repair. While periodic maintenance can prevent sudden failures to some extent, it often leads to over-maintenance and unnecessary downtime because it fails to dynamically adjust according to the actual health status of the equipment, thereby increasing operation and maintenance costs and wasting resources. Reactive repair, on the other hand, is carried out only after serious equipment failures occur, which not only causes production interruptions and economic losses but may also trigger a chain reaction of equipment damage or even safety accidents.

[0023] Predicting equipment failures in advance is particularly important. Traditional prediction methods are usually based on a single signal (such as vibration or temperature) for fault diagnosis and prediction. They typically use a fixed model structure and static parameter configuration. As a result, the prediction model cannot continuously adapt to changes in equipment status, which leads to a gradual decrease in the prediction accuracy of the fault prediction model during long-term operation.

[0024] To address the problems mentioned in the background art, according to one aspect of the embodiments of this application, an embodiment of an adaptive optimization method for a model is provided.

[0025] Optionally, in the embodiments of this application, the adaptive optimization method of the above model can be applied to, for example... Figure 1 The hardware environment shown consists of terminal 101 and server 103. Figure 1 As shown, server 103 is connected to terminal 101 via a network and can be used to provide services to the terminal or clients installed on the terminal. Database 105 can be set up on the server or independently of the server to provide data storage services for server 103. The network mentioned above includes, but is not limited to, wide area network, metropolitan area network or local area network. Terminal 101 includes, but is not limited to, PC, mobile phone, tablet computer, etc.

[0026] An adaptive optimization method for a model in this embodiment can be executed by server 103, or it can be jointly executed by server 103 and terminal 101, such as... Figure 2 As shown, it includes: Step 201: Obtain the prediction model corresponding to the target device, wherein the prediction model is trained based on the historical fault data of the target device; Step 202: During the process of using the prediction model to predict the fault of the target equipment, monitor the prediction error of the prediction model and the operating condition of the target equipment. Step 203: Determine whether to trigger the model adjustment mechanism based on the prediction error and operating conditions; Step 204: When the model adjustment mechanism is triggered, select the corresponding target adjustment strategy according to the magnitude of the prediction error and the change in the operating conditions. The target adjustment strategy is either a parameter adjustment strategy or a structural adjustment strategy. Step 205: Adjust the prediction model according to the target adjustment strategy to complete the adaptive optimization of the prediction model.

[0027] Obtain the prediction model corresponding to the target device. This prediction model is trained based on the historical fault data of the target device and is used to predict the future fault state of the device.

[0028] During the process of using a predictive model for fault prediction, the prediction error of the model is monitored in real time, and the operating data of the target device in multiple dimensions is acquired. At this time, the operating data is the input data of the predictive model.

[0029] Based on the magnitude of the prediction error and changes in equipment operating conditions, determine whether to trigger the model adjustment mechanism. The adjustment mechanism is triggered when the prediction error exceeds the error threshold or when operating conditions change significantly; otherwise, model optimization is not required.

[0030] When the model adjustment mechanism is triggered, an appropriate target adjustment strategy is selected based on the magnitude of the prediction error and changes in operating conditions. The target adjustment strategy includes parameter adjustment and structural adjustment strategies. The prediction model is automatically adjusted according to the selected target adjustment strategy to achieve adaptive optimization, making it adaptable to the dynamic operating state of the target equipment.

[0031] This application dynamically collects and analyzes multi-dimensional operating parameters of the equipment, adaptively adjusts the structure and parameters of the prediction model, accurately predicts potential equipment failures, provides factual basis for preventive maintenance of the equipment, and enables real-time monitoring and dynamic optimization of the prediction model.

[0032] As an optional embodiment, before obtaining the prediction model corresponding to the target device, the method further includes generating the prediction model in the following manner: obtaining the operating parameters of the target device from multiple dimensions and extracting the parameter features of the operating parameters; determining the model architecture matching the target device and judging the first relationship type between the parameter features and the failure probability of the target device, wherein the first relationship type is a linear relationship, a nonlinear relationship, or a time-series dependency relationship; obtaining an algorithm mapping table and determining the first model algorithm corresponding to the first relationship type through the algorithm mapping table, wherein the algorithm mapping table includes multiple sets of correspondences between relationship types and model algorithms; constructing an initial model based on the first model algorithm and the model architecture; and training the initial model using the historical failure data of the target device as training data to obtain the prediction model.

[0033] The system collects multiple operational parameters of the target equipment in real time, including but not limited to vibration signals, temperature data, pressure values, current and voltage parameters, operating speed, lubrication status, and operation logs. The collected data is transmitted to a central data processing system via sensors installed at key locations on the equipment.

[0034] Various types of sensors, such as vibration sensors, temperature sensors, pressure sensors, and current sensors, are used to collect equipment operating parameters in real time at high frequency and from all directions. For example, in wind turbine equipment, vibration sensors can monitor the vibration amplitude and frequency of key components such as blades and gearboxes in real time; temperature sensors continuously track the temperature of heat-prone parts such as motors and bearings.

[0035] Because the raw data collected is prone to noise interference and data gaps, it is necessary to preprocess the raw data. For example, median filtering can be used to remove abnormal spike noise from vibration data; linear interpolation can be used to fill in missing temperature data caused by transient sensor malfunctions; and moving average can be used to smooth temperature and pressure data curves.

[0036] To unify the data units of different parameters, a normalization method is used to map all cleaned data to the interval [0, 1] for subsequent model processing.

[0037] Figure 3 This application provides a schematic diagram of a system architecture for equipment data acquisition and transmission. This architecture can be used to realize the real-time acquisition, transmission, and storage of operating data from industrial equipment such as CNC machine tools, PLCs (Programmable Logic Controllers), smart meters, and sensor modules. The architecture can be divided into four parts: the underlying device layer, the data acquisition layer, the network transmission layer, and the data storage and security layer. The following are detailed descriptions of each part: 1. Bottom layer (industrial equipment cluster), including: CNC machine tools, core processing equipment in industrial production, which upload operating parameters in real time through a data acquisition gateway and wired connection; PLC, the core component of industrial control system, responsible for the logic control and status monitoring of equipment, whose data is accessed by the data acquisition gateway; smart meters, used to monitor energy consumption data of industrial electricity, providing data support for energy management and equipment energy efficiency analysis; sensor modules, which can deploy various sensors such as vibration, temperature, and pressure according to scenario requirements to collect environmental or operating status data of key parts of the equipment (such as bearing vibration, equipment surface temperature, etc.).

[0038] 2. Data acquisition layer, including multiple acquisition gateways. The acquisition gateways act as an intermediate bridge between the devices and the network, and connect to the underlying industrial devices through wired connections (such as Ethernet) to complete the unified acquisition, protocol parsing and preliminary processing of data from multiple types of devices, and achieve compatibility and adaptation of "multiple devices and multiple protocols".

[0039] 3. Network transmission layer, including: 5G CPE (Customer Premises Equipment), which serves as the access terminal for the 5G network and uploads the device data transmitted by the collection gateway via 5G wireless communication to meet the real-time data transmission requirements of industrial scenarios; 5G base station, which is responsible for 5G signal coverage and data forwarding, and transmits the data uploaded by 5G CPE to the back-end data center via wireless link.

[0040] 4. Data storage and security layer, including: firewall, deployed between 5G network and data server, which plays a role in network security isolation, prevents unauthorized access and network attacks, and ensures the security of industrial data; data acquisition server, as the final storage and management node of data, receives and stores all device operation data from 5G network.

[0041] Feature extraction is performed on the collected operating parameters to obtain key features that reflect the operating status and fault modes of the target equipment. For example, the mean, variance, peak value, root mean square value, skewness, kurtosis, and other statistics of each operating parameter are calculated to describe the overall distribution of the operating parameters; time-domain features are extracted from time-series data (such as vibration, current, and pressure fluctuations); and for vibration and noise signals, discrete Fourier transform is used to convert the signal from the time domain to the frequency domain to extract frequency-domain features.

[0042] To reduce dimensionality while retaining key information, principal component analysis (PCA) is used to linearly reduce the dimensionality of vectors containing multiple parameter features. Combining the structural characteristics of the target equipment, historical fault data, and expert experience, features highly correlated with fault probability are identified. For example, for bearing equipment, envelope spectrum energy is prioritized; for motor equipment, three-phase current harmonic components and temperature rise trends are prioritized; and for pressure systems, pressure fluctuation amplitude and pressure stability indices can be selected.

[0043] The appropriate model architecture is determined based on the device type and operating characteristics of the target device. In this process, different priority model architectures are configured for each device in advance based on experience and device characteristics. When actually building the model, the model architecture with the highest priority is selected by default.

[0044] To determine the first type of relationship between parameter characteristics and equipment failure probability, the methods include: measuring the linear correlation between two variables using the Pearson correlation coefficient; if the absolute value is greater than a certain preset value (such as 0.8), it is considered a strong linear correlation; or using residual analysis to check whether the prediction error of the linear model is randomly distributed; if the residuals do not have obvious systematic shifts, it indicates that the linear assumption is valid.

[0045] The types of relationships involved in this application include linear relationships, non-linear relationships, or temporal dependencies. In practice, they may also include multivariate coupling relationships and conditional dependencies.

[0046] The algorithm mapping table is constructed by statistically analyzing and summarizing a large amount of historical equipment data, typical failure modes, and the performance of existing models. Specifically, based on the experimental results of a large amount of historical equipment data, the average performance of each model algorithm under different relation types is statistically analyzed, the optimal algorithm is selected, and the mapping is stored.

[0047] For example, if the relationship type is linear, the corresponding algorithm is linear regression or LASSO (Least Absolute Shrinkage and Selection Operator); if the relationship type is non-linear, the corresponding algorithm is deep neural network or XGBoost (eXtreme Gradient Boosting); if the relationship type is temporal dependency, the corresponding algorithm is LSTM (Long Short-Term Memory) / GRU (Gated Recurrent Unit).

[0048] Historical fault data of the target equipment is used as training data to train the initial model, resulting in a prediction model corresponding to the target equipment. For example, historical operating parameters and corresponding fault labels are parsed from historical fault data to construct training data containing normal and fault samples. The training data includes multi-dimensional operating parameters such as vibration, temperature, current, and pressure, as well as their corresponding fault categories or health states.

[0049] The training data is input into the initial model built based on the first model algorithm and model architecture. The model parameters are iteratively optimized through backpropagation, gradient descent, optimal splitting strategy or other algorithm-specific training mechanisms. After the training process converges or reaches the set stopping condition, the parameter optimization of the initial model is completed, and a predictive model that can characterize the operating characteristics and fault patterns of the target equipment is obtained.

[0050] Historical operating parameters include multiple influencing factors of equipment failure. Each influencing factor and the failure label are input into the initial model (such as a BP neural network) so that it learns how to predict failure conditions based on the influencing factors.

[0051] This application can automatically select appropriate model algorithms and model architectures based on the parameter characteristics and fault modes of the target equipment, construct an initial model that matches the operating characteristics of the equipment, and then use the historical fault data of the equipment as training data to train and obtain a prediction model.

[0052] As an optional embodiment, monitoring the prediction error of the prediction model and the operating condition of the target equipment includes: calculating the prediction error based on the output of the prediction model and the actual equipment status; calculating the real-time data distribution of the input data of the prediction model, and comparing the real-time data distribution with the initial data distribution to determine whether the operating condition has changed, wherein the initial data distribution is the parameter distribution in the historical fault data.

[0053] It receives the output of the prediction model on the current operating status of the target device, obtains the actual operating status of the target device, calculates the prediction error based on the difference between the output of the prediction model and the actual operating status, and uses it to evaluate the accuracy of the current model prediction.

[0054] Taking binary classification prediction of equipment failure (normal or faulty) as an example, the cross-entropy loss function is used to measure the difference between the predicted value and the true value. The calculation formula is: , Where L represents the prediction error, y represents the true value (the value corresponding to the actual operating state of the equipment, 0 for normal state and 1 for fault state), and p represents the probability of the equipment operating normally output by the prediction model (which can be understood as the health of the equipment).

[0055] For example, if the probability of the device operating normally output by the prediction model is p, while the actual device is in a fault state (the true value is 1), then the cross-entropy loss is L=-(1*log(p)+(1-1)*log(1-p))=-log(p).

[0056] For more complex multi-class fault prediction or equipment remaining life regression prediction, the corresponding multi-class cross-entropy loss function or mean squared error loss function is used to accurately calculate the prediction error.

[0057] Extract real-time feature distribution parameters from the input data of the prediction model, including statistical features such as mean, variance, skewness, and kurtosis, and construct the real-time data distribution.

[0058] The real-time data distribution is compared with the initial data distribution of the corresponding parameters in the historical fault data. When the real-time distribution deviates significantly in statistical characteristics, it is determined that the operating condition of the target equipment has changed; otherwise, the operating condition is considered to remain stable.

[0059] Changes in operating conditions can be caused by the emergence of new types of faults or changes in the operating environment.

[0060] The initial data distribution formed by historical fault data represents the parameter range of the equipment under typical operating conditions. When the real-time input data deviates from the baseline distribution (such as mean shift or variance expansion), it indicates that the equipment's operating environment, load, or external conditions have changed, thus identifying the potential impact of these changes on the predictive model.

[0061] This embodiment enables simultaneous monitoring of prediction model error and equipment operating conditions. It can quantify the model prediction accuracy and capture the dynamic changes in the distribution of equipment operating data, thereby providing a reliable basis for whether to trigger the model adjustment mechanism and improving the accuracy and timeliness of model adaptive adjustment.

[0062] As an optional embodiment, determining whether to trigger the model adjustment mechanism based on the prediction error and operating conditions includes: if the prediction error is less than or equal to the error threshold and the operating conditions have not changed, then it is determined that the model adjustment mechanism has not been triggered; if the prediction error is greater than the error threshold and / or the operating conditions have changed, then it is determined that the model adjustment mechanism has been triggered.

[0063] The error threshold involved in this application is a dynamic value. An initial value is first set according to the actual situation, and then it may change with the change of operating conditions during the operation of the equipment.

[0064] When the prediction error is less than or equal to the error threshold, it indicates that the prediction performance of the prediction model meets the requirements under the current operating conditions; when the prediction error is greater than the error threshold, it indicates that the prediction performance of the prediction model has declined.

[0065] By comparing the real-time data distribution of the input data with the initial data distribution obtained from historical fault data, it is determined whether the operating conditions of the equipment have changed: when the real-time data distribution does not deviate from the initial data distribution, it is determined that the operating conditions have not changed; when the real-time data distribution shows a significant deviation, it is determined that the operating conditions have changed.

[0066] If the prediction error is less than or equal to the error threshold and the operating conditions remain unchanged, the model adjustment mechanism is not triggered. If the prediction error is greater than the error threshold and / or the operating conditions change, the model adjustment mechanism is triggered. The prediction error reflects the model's fit to the current data, while changes in operating conditions reflect whether the input distribution has shifted, meaning that the historical data structure during model training is no longer consistent with the current operating environment.

[0067] It should be noted that if the prediction error is less than or equal to the error threshold, but the operating conditions change, the prediction error is still low, indicating that the prediction model can still fit the current data well. However, as the operating conditions shift further, the performance may drop sharply, so a light adjustment to the prediction model is required.

[0068] Lightweight adjustments include at least one of the following methods: for features that have changed in the input data, adjust the model’s connections or processing methods for these features while keeping the unchanged parts of the model structure unchanged to avoid overall retraining; if the model adopts a sub-model combination structure, only the sub-models affected by the change in working conditions can be reassigned weights without affecting the output of other sub-models of the overall model; only delete or reconfigure nodes or connections that are less affected by the changed features.

[0069] This application triggers a model adjustment mechanism in a timely manner when the model's predictive performance is insufficient or the equipment's operating conditions change, thereby avoiding the model from continuously outputting inaccurate prediction results due to error accumulation or input distribution drift, and improving the model's stability and prediction reliability at different operating stages.

[0070] As an optional embodiment, a corresponding target adjustment strategy is selected based on the magnitude of the prediction error and the change in operating conditions, including: if the prediction error is greater than the error threshold and the operating conditions have not changed, then the parameter adjustment strategy is used as the target adjustment strategy; if the prediction error is greater than the error threshold and the operating conditions have changed, then the structural adjustment strategy is used as the target adjustment strategy.

[0071] When the prediction error exceeds the error threshold and the operating conditions remain unchanged, a parameter adjustment strategy is executed to adaptively optimize the parameters of the prediction model, including adjusting parameters such as model weights and learning rate through gradient updates or hyperparameter optimization to reduce the prediction error.

[0072] For example, the model parameters are adjusted using the Stochastic Gradient Descent (SGD) algorithm. This algorithm calculates the gradient for each training sample and updates the model parameters in the opposite direction of the gradient. Assuming the model parameters are θ and the learning rate is α, in the t-th iteration, the gradient calculated for sample i is... Then the parameter update formula is: The model parameters are optimized iteratively until the preset convergence condition or the maximum number of iterations is reached.

[0073] To improve the stability and convergence speed of the SGD algorithm, further optimization algorithms such as momentum method, Adagrad, and Adadelta can be employed. Furthermore, global optimization algorithms such as Bayesian optimization and genetic algorithms can be used to search for better model hyperparameters, such as the number of layers in the neural network and the number of neurons in the hidden layers, to improve the overall performance of the model.

[0074] When the prediction error exceeds the error threshold and the operating conditions change, a structural adjustment strategy is executed to adaptively adjust the network structure of the model, including reconstructing the network hierarchy, introducing new network units, or fusing sub-models with different structures, so that the model structure can adapt to new input data features and fault modes.

[0075] The model adjustment mechanism provided in this application will automatically update the model's input features and parameter matrix based on newly collected data, or switch to a prediction model architecture that is more suitable for the current equipment status, so as to ensure that the model can always accurately adapt to the dynamic changes of the equipment and maintain high fault prediction accuracy.

[0076] This application enables the automatic selection of the most suitable adjustment strategy based on the combined state of prediction error and operating conditions, thereby achieving adaptive optimization of the model under different operating conditions, thus improving the model's adaptability and stability to changes in equipment operating conditions while ensuring prediction accuracy.

[0077] As an optional implementation, if the target adjustment strategy is a structure adjustment strategy, the prediction model is adjusted according to the target adjustment strategy, including at least one of the following methods: training multiple sub-models with different structures in parallel; calculating the prediction performance index of each sub-model under the current working condition; assigning dynamic weights to each sub-model according to each prediction performance index; and weighting and fusing the outputs of each sub-model according to the weights to generate the adjusted prediction model.

[0078] Construct multiple sub-models with different structures. These sub-models differ at least in the number of network layers, layer types, or connection methods. Train each sub-model on the data characteristics of the input data under the current operating conditions to adapt to the current operating conditions.

[0079] Calculate the predictive performance metrics for each sub-model, including prediction error, accuracy, and other applicable quantitative indicators. Based on these metrics, assign dynamic weights to each sub-model, with higher-performing sub-models receiving larger weights and lower-performing sub-models receiving smaller weights.

[0080] The outputs of each sub-model are weighted and fused according to the assigned dynamic weights to generate an adjusted prediction model.

[0081] For example, when product type switching in an industrial production line causes a significant change in equipment load, by monitoring the distribution changes of input data, if the variance, mean, or other statistical characteristics of the data exceed a preset threshold range, an automatic model structure adjustment mechanism is triggered. A model fusion strategy can be used to dynamically combine multiple sub-models with different structures, such as fusing local features extracted by a convolutional neural network with time-series features captured by a recurrent neural network, to better adapt to new failure modes.

[0082] Training multiple sub-models in parallel can provide different structural perspectives, thus making them more adaptable to complex or changing operating conditions. By dynamically allocating weights based on the actual prediction performance of each sub-model under the current operating conditions, and then weighted and fused the outputs of the sub-models, the prediction results can be guaranteed to prioritize the sub-models with better performance. This can combine the advantages of each sub-model and improve the overall prediction accuracy and stability.

[0083] In addition to model fusion strategies, structural adjustment strategies can also include network reconstruction strategies or model pruning strategies. Network reconstruction strategies include: monitoring changes in the dimensions of input features and their relationship with the target variable; determining whether the existing network structure matches the new data features; if not, dynamically adding or deleting network layers based on feature complexity; adjusting inter-layer connections or introducing new network units (such as convolutional layers, LSTM layers, etc.) to complete network structure reconstruction. Model pruning strategies include: calculating the contribution index of each network node or connection; determining whether low-contribution nodes or connections are below a set threshold; if below the threshold, deleting nodes or connections; updating the model structure and retraining to form a pruned, simplified network model.

[0084] As an optional embodiment, after completing the adaptive optimization of the prediction model, the method further includes: using the prediction model to predict the fault of the target device, obtaining the predicted fault result output by the prediction model, wherein the predicted fault result includes the predicted fault probability, the predicted fault type, the faulty component, and the predicted fault time; determining the fault probability interval in which the predicted fault probability falls, and determining the warning level of the target device based on the fault probability interval; generating warning information corresponding to the warning level, and sending the warning information to the terminal, wherein the warning information includes the warning level and the predicted fault result.

[0085] Using the trained adaptive fault prediction model, the real-time operating data of the target device is predicted to obtain the predicted fault probability or health status of the target device in the future. The sum of the health status and the fault probability value is 1.

[0086] Based on the importance of the target equipment, historical fault data, and safety standards, early warning thresholds for various levels are pre-set. The early warning levels provided in this application include Level 1, Level 2, and Level 3 early warnings, and more levels can be determined based on the actual number of intervals set.

[0087] A Level 1 warning indicates that an emergency malfunction may be imminent; a Level 2 warning indicates that a potential malfunction may occur in the near future; and a Level 3 warning indicates that the likelihood of a potential malfunction occurring in the near future is relatively low.

[0088] The predicted failure probability is compared with the warning thresholds at each level to determine the current warning level of the equipment. For example, when the predicted failure probability is greater than or equal to the first-level threshold, it is determined to be a first-level warning; when the second-level threshold is less than or equal to the predicted failure probability but less than the first-level threshold, it is determined to be a second-level warning; and when the third-level threshold is less than or equal to the predicted failure probability but less than the second-level threshold, it is determined to be a third-level warning.

[0089] Based on the determined warning level, corresponding warning information is generated, including the warning level, prediction time window, possible fault types, and information on affected components.

[0090] Warning information, including the warning level and predicted failure outcome, is sent to the terminal so that maintenance personnel can initiate corresponding response strategies based on the warning level. For example, if it is a Level 1 warning, the maintenance system is immediately notified and a shutdown for maintenance is recommended; if it is a Level 2 warning, maintenance personnel are dispatched to conduct key inspections or replace vulnerable parts; if it is a Level 3 warning, the monitoring frequency is increased and the equipment operating status is continuously tracked.

[0091] Figure 4 This is a schematic diagram of the equipment fault early warning report provided in this application. The report presents the equipment health status and maintenance recommendations from two aspects: a diagnostic overview and specific fault details.

[0092] The diagnostic overview section in the figure includes: Machine health, only 33.33%, indicating that the overall health of the equipment is poor and there are many potential failure points; Diagnostic confidence, 100%, indicating that the system is very confident in the reliability of the current diagnostic results; Trend curve, showing the change of equipment health over time, gradually decreasing from 100% health in the early stage, indicating that the failure is gradually developing and the evolution process of the failure can be traced.

[0093] The machine condition diagnosis, assessment, and maintenance recommendations section in the diagram includes: component information, specifically the "drive axle half-shaft" of the "axle housing" component; health status, marked "dangerous," indicating a serious fault requiring immediate attention; drive model "Bearing," meaning the fault originates from the bearing; diagnosis conclusion "severe wear," clarifying the bearing fault type; maintenance recommendations "check for metal particles in the grease and whether the grease is within its expiration date, and also check the bearing for any issues," providing specific troubleshooting directions; processing status, "pending," indicating that the fault has not yet been repaired and needs to be addressed as soon as possible; and an "click to process" entry in the operation section, facilitating maintenance personnel to follow up on the fault handling process.

[0094] Figure 5 This application provides a schematic diagram of a bearing fault diagnosis and prediction system. Its core is to achieve accurate diagnosis and prediction of bearing faults through a complete process of data acquisition, feature engineering, algorithm modeling, and result verification. The architecture includes the following four parts: 1. Data acquisition module, including: bearing fault test bench data, which collects standardized fault data by simulating various bearing faults (such as wear, cracks, insufficient lubrication, etc.) in the experimental environment to provide benchmark samples for algorithm training; bearing data acquisition from vehicle fault components, which collects operating data from faulty bearings in actual vehicles, covering multi-dimensional information such as vibration, temperature, and speed under real working conditions to ensure the "real-world adaptability" of the model.

[0095] 2. Feature Engineering Module, including: Feature Extraction, which extracts key fault-related information from raw data. For example, performing time-domain analysis (extracting peak value, RMS value, kurtosis, etc.), frequency-domain analysis (identifying fault characteristic frequencies through Fourier transform), or time-frequency-domain analysis (capturing transient fault information through wavelet transform) on vibration signals, transforming the raw data into a "feature set that can be understood by the algorithm"; Feature Selection, which filters the most relevant subset of features from a large number of extracted features, eliminating redundant or irrelevant features (e.g., through correlation analysis, variance filtering, etc.), reducing the computational load of the model, and improving diagnostic / prediction accuracy.

[0096] 3. Algorithm modeling module, including: Diagnosis (classification algorithm construction), which builds classification models (such as Support Vector Machine (SVM), Random Forest, Deep Learning CNN, etc.) based on the selected features to achieve "fault type identification"—determining whether the bearing is faulty and what type of fault it belongs to (such as inner race fault, outer race fault, rolling element fault); Prediction (regression algorithm construction), which builds regression models (such as linear regression, LSTM time series model, etc.) based on features to achieve "fault development trend prediction", which can estimate the severity of the fault, remaining life and other quantitative indicators.

[0097] 4. Result verification and application module, including: result output, the diagnostic model outputs the fault type, the prediction model outputs the fault development trend, forming a complete "fault diagnosis-prediction" conclusion; expert system verification, comparing the conclusion with the experience judgment of domain experts to verify the accuracy of the model; at the same time, the conclusion is fed back to the "vehicle faulty components" link to form a closed loop verification and continuously optimize the model performance.

[0098] This application provides an adaptive optimization method for a model, comprising: acquiring a prediction model corresponding to a target device, wherein the prediction model is trained based on historical fault data of the target device; during the process of using the prediction model to predict faults in the target device, monitoring the prediction error of the prediction model and the operating conditions of the target device; determining whether to trigger a model adjustment mechanism based on the prediction error and operating conditions; if the model adjustment mechanism is triggered, selecting a corresponding target adjustment strategy based on the magnitude of the prediction error and changes in operating conditions, wherein the target adjustment strategy is a parameter adjustment strategy or a structural adjustment strategy; and adjusting the prediction model according to the target adjustment strategy to complete the adaptive optimization of the prediction model. By monitoring the prediction error and operating conditions in real time, automatically selecting a parameter adjustment strategy or a structural adjustment strategy to adaptively optimize the prediction model, the method ensures that the model can continuously adapt to changes in device state and achieve accurate prediction of device faults.

[0099] According to another aspect of the embodiments of this application, this application provides an adaptive optimization apparatus for a model, such as... Figure 6 As shown, it includes: The acquisition module 601 is used to acquire a prediction model corresponding to the target device, wherein the prediction model is trained based on the historical fault data of the target device; The monitoring module 602 is used to monitor the prediction error of the prediction model and the operating condition of the target equipment during the process of using the prediction model to predict the fault of the target equipment. The judgment module 603 is used to determine whether to trigger the model adjustment mechanism based on the prediction error and the operating conditions. Select module 604 is used to select the corresponding target adjustment strategy based on the magnitude of the prediction error and the change in operating conditions when the model adjustment mechanism is triggered. The target adjustment strategy is either a parameter adjustment strategy or a structural adjustment strategy. The adjustment module 605 is used to adjust the prediction model according to the target adjustment strategy in order to complete the adaptive optimization of the prediction model.

[0100] It should be noted that the acquisition module 601 in this embodiment can be used to execute step 201 in this application embodiment, the monitoring module 602 in this embodiment can be used to execute step 202 in this application embodiment, the judgment module 603 in this embodiment can be used to execute step 203 in this application embodiment, the selection module 604 in this embodiment can be used to execute step 204 in this application embodiment, and the adjustment module 605 in this embodiment can be used to execute step 205 in this application embodiment.

[0101] Optionally, the device further includes a generation module, used to generate a prediction model in the following manner before obtaining a prediction model corresponding to the target device: obtaining the operating parameters of the target device from multiple dimensions and extracting the parameter features of the operating parameters; determining the model architecture matching the target device and judging the first relationship type between the parameter features and the failure probability of the target device, wherein the first relationship type is a linear relationship, a nonlinear relationship, or a time-series dependency relationship; obtaining an algorithm mapping table and determining the first model algorithm corresponding to the first relationship type through the algorithm mapping table, wherein the algorithm mapping table includes multiple sets of correspondences between relationship types and model algorithms; constructing an initial model based on the first model algorithm and the model architecture; and training the initial model using historical failure data of the target device as training data to obtain a prediction model.

[0102] Optionally, the monitoring module 602 is specifically used to calculate the prediction error based on the output of the prediction model and the actual equipment status; calculate the real-time data distribution of the input data of the prediction model, and compare the real-time data distribution with the initial data distribution to determine whether the operating conditions have changed, wherein the initial data distribution is the parameter distribution in the historical fault data.

[0103] Optionally, the judgment module 603 is specifically used to determine that the model adjustment mechanism has not been triggered if the prediction error is less than or equal to the error threshold and the operating conditions have not changed; and to determine that the model adjustment mechanism has been triggered if the prediction error is greater than the error threshold and / or the operating conditions have changed.

[0104] Optionally, the selection module 604 is specifically used to select the parameter adjustment strategy as the target adjustment strategy if the prediction error is greater than the error threshold and the operating conditions have not changed; and to select the structural adjustment strategy as the target adjustment strategy if the prediction error is greater than the error threshold and the operating conditions have changed.

[0105] Optionally, if the target adjustment strategy is a structure adjustment strategy, the adjustment module 605 is used to adjust the model in at least one of the following ways: train multiple sub-models with different structures in parallel; calculate the prediction performance index of each sub-model under the current working condition; assign dynamic weights to each sub-model according to each prediction performance index; and weight and fuse the outputs of each sub-model according to the weights to generate the adjusted prediction model.

[0106] Optionally, the device further includes an early warning module, which, after completing the adaptive optimization of the prediction model, uses the prediction model to predict the fault of the target device and obtains the predicted fault result output by the prediction model. The predicted fault result includes the predicted fault probability, the predicted fault type, the faulty component, and the predicted fault time. The module then determines the fault probability interval in which the predicted fault probability falls and determines the early warning level of the target device based on the fault probability interval. Finally, the module generates early warning information corresponding to the early warning level and sends the early warning information to the terminal. The early warning information includes the early warning level and the predicted fault result.

[0107] It should be noted that the examples and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the content disclosed in the above embodiments. It should also be noted that the above modules, as part of a device, can operate in environments such as... Figure 1 The hardware environment shown.

[0108] According to another aspect of the embodiments of this application, this application provides an electronic device, such as... Figure 7 As shown, the device includes a memory 701, a processor 703, a communication interface 705, and a communication bus 707. The memory 701 stores a computer program that can run on the processor 703. The memory 701 and the processor 703 communicate through the communication interface 705 and the communication bus 707. When the processor 703 executes the computer program, it implements the steps of the above method.

[0109] The memory and processor in the aforementioned electronic devices communicate with each other via a communication bus and a communication interface. The communication bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into an address bus, a data bus, a control bus, etc.

[0110] The memory may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0111] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0112] According to another aspect of the embodiments of this application, a computer-readable medium having processor-executable non-volatile program code is also provided.

[0113] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments, and will not be repeated here.

[0114] In specific implementation, the embodiments of this application can be referred to the above embodiments and have corresponding technical effects.

[0115] It is understood that the embodiments described herein can be implemented in hardware, software, firmware, middleware, microcode, or a combination thereof. For hardware implementation, the processing unit can be implemented in one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers, microprocessors, other electronic units for performing the functions described herein, or combinations thereof.

[0116] For software implementation, the techniques described herein can be implemented by units that perform the functions described herein. The software code can be stored in memory and executed by a processor. The memory can be implemented in the processor or external to the processor.

[0117] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0118] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0119] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0120] The units described as separate components may or may not be physically separate. The components shown 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 units can be selected to achieve the purpose of this embodiment according to actual needs.

[0121] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0122] If the aforementioned function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiments of this application, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks. It should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. In the absence of further restrictions, an element defined by the phrase "comprising a..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0123] The above description is merely a specific embodiment of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A method of adaptive optimization of a model, characterized in that, The method comprises the following steps: acquiring a prediction model corresponding to a target device, wherein the prediction model is obtained by training historical fault data of the target device; monitoring a prediction error of the prediction model and an operating condition of the target device during fault prediction of the target device by using the prediction model; judging whether to trigger a model adjustment mechanism according to the prediction error and the operating condition; if the model adjustment mechanism is triggered, selecting a corresponding target adjustment strategy according to the size of the prediction error and the change of the operating condition, wherein the target adjustment strategy is a parameter adjustment strategy or a structure adjustment strategy; adjusting the prediction model according to the target adjustment strategy to complete adaptive optimization of the prediction model.

2. The method of claim 1, wherein, Before acquiring the prediction model corresponding to the target device, the method further comprises generating the prediction model in the following manner: acquiring operating parameters of the target device from multiple dimensions and extracting parameter features of the operating parameters; determining a model architecture matched with the target device and judging a first relationship type between the parameter features and a fault probability of the target device, wherein the first relationship type is a linear relationship, a nonlinear relationship or a time sequence dependent relationship; acquiring an algorithm mapping table and determining a first model algorithm corresponding to the first relationship type through the algorithm mapping table, wherein the algorithm mapping table comprises a plurality of corresponding relationships between relationship types and model algorithms; constructing an initial model according to the first model algorithm and the model architecture; training the initial model by taking historical fault data of the target device as training data to obtain the prediction model.

3. The method of claim 2, wherein, The monitoring of the prediction error of the prediction model and the operating condition of the target device comprises: calculating the prediction error according to an output result of the prediction model and an actual device state; calculating a real-time data distribution of input data of the prediction model and comparing the real-time data distribution with an initial data distribution to judge whether the operating condition has changed, wherein the initial data distribution is a parameter distribution in the historical fault data.

4. The method of claim 3, wherein, The judging of whether to trigger the model adjustment mechanism according to the prediction error and the operating condition comprises: if the prediction error is less than or equal to an error threshold value and the operating condition has not changed, it is determined that the model adjustment mechanism is not triggered; if the prediction error is greater than the error threshold value and / or the operating condition has changed, it is determined that the model adjustment mechanism is triggered.

5. The method of claim 4, wherein, The selecting of the corresponding target adjustment strategy according to the size of the prediction error and the change of the operating condition comprises: if the prediction error is greater than the error threshold value and the operating condition has not changed, the parameter adjustment strategy is selected as the target adjustment strategy; if the prediction error is greater than the error threshold value and the operating condition has changed, the structure adjustment strategy is selected as the target adjustment strategy.

6. The method of claim 3, wherein, If the target adjustment strategy is the structure adjustment strategy, the adjusting of the prediction model according to the target adjustment strategy comprises at least one of the following manners: Parallel training of multiple sub-models with different structures; calculating a prediction performance index of each of the sub-models under the current working condition; assigning a dynamic weight to each of the sub-models according to the prediction performance index; weighting and fusing the outputs of each of the sub-models to generate an adjusted prediction model.

7. The method of claim 1, wherein, After completing the adaptive optimization of the prediction model, the method further comprises: using the prediction model to predict the failure of the target device to obtain a predicted failure result output by the prediction model, wherein the predicted failure result includes a predicted failure probability, a predicted failure type, a failed component, and a predicted failure time; determining a failure probability interval in which the predicted failure probability is located, and determining a warning level of the target device according to the failure probability interval; generating warning information corresponding to the warning level, and sending the warning information to a terminal, wherein the warning information includes the warning level and the predicted failure result.

8. An apparatus for adaptive optimization of a model, characterized by comprises: an acquisition module configured to acquire a prediction model corresponding to a target device, wherein the prediction model is obtained by training historical failure data of the target device; a monitoring module configured to monitor a prediction error of the prediction model and a running condition of the target device during a process of predicting a failure of the target device by using the prediction model; a determination module configured to determine whether to trigger a model adjustment mechanism according to the prediction error and the running condition; a selection module configured to select a corresponding target adjustment strategy according to a size of the prediction error and a change of the running condition when the model adjustment mechanism is triggered, wherein the target adjustment strategy is a parameter adjustment strategy or a structure adjustment strategy; an adjustment module configured to adjust the prediction model according to the target adjustment strategy to complete adaptive optimization of the prediction model.

9. An electronic device comprising a memory, a processor, a communication interface and a communication bus, the memory having stored therein a computer program executable on the processor, the memory, the processor being in communication via the communication bus and the communication interface, characterized in that, The processor executes the computer program to realize the steps of the method of any one of claims 1 to 7.

10. A computer readable medium having non-transitory program code executable by a processor, the program code comprising instructions for: The program code causes the processor to execute the method of any one of claims 1 to 7.

Images