Automatic driving electronic system reliability prediction method and system based on deep learning

Abstract

The invention provides an automatic driving electronic system reliability prediction method and system based on deep learning, and the method comprises the steps: building an electromagnetic environment and component interaction model, and tracking an electromagnetic interference source through time-frequency analysis; predicting a failure mode evolution path of a system component by using a dynamic LSTM model, and introducing a time-varying disturbance model to perform dynamic electromagnetic interference regularization; combining a failure mode prediction result, and adopting a self-adaptive time window and a time sequence regression model to predict the combined influence of electromagnetic interference on system failure; adjusting system control signals and resource configuration through an adaptive fault-tolerant control algorithm and a dynamic resource allocation strategy based on failure mode prediction and electromagnetic interference prediction values; and dynamically triggering a failure prevention mechanism according to the health assessment function, and optimizing prevention measures through a feedback correction mechanism. According to the method, the defects of the prior art in the aspects of prediction precision, interference identification, fault-tolerant mechanism and the like are effectively overcome, and a new technical path is provided for improving the reliability of the automatic driving system.

Description

Technical Field

[0001] The present invention belongs to the field of deep learning, and in particular to a method and system for predicting the reliability of an autonomous driving electronic system based on deep learning. Background Art

[0002] With the rapid development of autonomous driving technology, the complexity of the electronic systems of autonomous driving vehicles continues to increase, covering a large number of hardware components such as sensors, control units, power management modules, actuators, etc. In order to ensure the safe and efficient operation of the vehicle under various road conditions and environmental conditions, the electronic components of the autonomous driving system need to have extremely high reliability and stability. This requires the system to continuously monitor the operating status of each component and to take preventive and repair measures in a timely manner in abnormal situations such as failures or electromagnetic interference. However, existing autonomous driving electronic systems still face many challenges in reliability prediction and fault tolerance.

[0003] At present, the reliability prediction technology of autonomous driving systems mainly relies on traditional static methods and simple historical data analysis. Traditional static methods usually make predictions based on existing failure modes and empirical models. This method lacks the ability to dynamically adapt to real-time changing environments and often ignores the impact of external factors such as electromagnetic interference on the system. Electromagnetic interference (EMI) is a common and serious problem in autonomous driving systems, especially in complex traffic environments. Autonomous driving vehicles need to communicate with other vehicles, traffic facilities, and wireless networks in real time, and there are many electronic components inside the vehicle, so they are easily affected by external electromagnetic signals. Electromagnetic interference not only affects the signal quality of the sensor, but may also cause failure of the control unit or data transmission errors. At present, the identification and response strategies of electromagnetic interference mostly rely on empirical rules or simple filtering techniques, but these methods often cannot effectively cope with complex and dynamic electromagnetic environments, which easily leads to insufficient fault tolerance of the system.

[0004] In addition, existing fault tolerance mechanisms are usually based on redundant design and preset fault tolerance strategies. Although these methods can provide protection in some common fault situations, when the system has complex failure modes or is subject to strong electromagnetic interference, traditional redundancy and fault tolerance mechanisms often fail to play their due role. Especially in the context of the continuous advancement of deep learning and artificial intelligence technology, traditional fault tolerance strategies appear to be lagging behind and are difficult to meet the requirements of autonomous driving systems for high reliability and high fault tolerance. Therefore, how to predict system failure modes in complex and dynamic operating environments and optimize the system's fault tolerance mechanism in real time has become an important issue in improving the reliability of autonomous driving electronic systems.

[0005] The existing technology also faces the following major problems:

[0006] Insufficient dynamics of failure prediction: Existing methods lack the ability to monitor and dynamically predict changes in system operating status in real time and cannot adapt to changes in different working environments.

[0007] Limited ability to identify and respond to electromagnetic interference: Traditional electromagnetic interference identification methods fail to effectively combine internal system data analysis with the actual impact of the electromagnetic environment, and lack accurate prediction and fault-tolerant mechanisms for system failure modes caused by electromagnetic interference.

[0008] Limited adaptability of fault-tolerant mechanisms: Traditional fault-tolerant strategies are based on static models and redundant designs, but these methods are difficult to respond promptly and effectively when the system encounters unknown faults or complex electromagnetic interference. Summary of the invention

[0009] The purpose of this invention is to propose a reliability prediction method and system for autonomous driving electronic systems based on deep learning. By introducing a hybrid intelligent model, electromagnetic interference identification and adaptive fault-tolerant mechanism, and dynamic optimization strategy, a new reliability prediction and fault-tolerant optimization method for autonomous driving electronic systems is provided.

[0010] In order to achieve the above object, a first aspect of the present invention provides a method for predicting the reliability of an autonomous driving electronic system based on deep learning, comprising the following steps:

[0011] S1: Establish an electromagnetic environment and component interaction model, generate system health status input through multi-dimensional sensor data fusion, and use time-frequency analysis to track electromagnetic interference sources;

[0012] S2: Identify the failure mode of the system based on the high-order statistical features of the health state; use the dynamic LSTM model to predict the evolution path of the failure mode of the system components, and introduce a time-varying disturbance model for dynamic electromagnetic interference regularization;

[0013] S3: Combined with the failure mode prediction results, adaptive time window and time series regression model are used to predict the joint impact of electromagnetic interference on system failure;

[0014] S4: Based on the failure mode prediction and electromagnetic interference prediction value, the system control signal and resource configuration are adjusted through adaptive fault-tolerant control algorithm and dynamic resource allocation strategy;

[0015] S5: Design a health assessment function based on real-time feedback, dynamically trigger the failure prevention mechanism based on the health assessment function, and optimize the preventive measures through the feedback correction mechanism.

[0016] Furthermore, the electromagnetic environment and component interaction model quantifies the impact of each interference source on the system module through an electromagnetic interference factor; the electromagnetic interference factor is expressed as:

[0017]

[0018] Among them, I EM (t) is the electromagnetic interference factor, α k is the intensity coefficient of the kth electromagnetic source, reflecting the electromagnetic field intensity of the electromagnetic source; r k (t) is the distance between the kth electromagnetic source and the target component at time t. As the system moves, the distance changes dynamically; λ k is the propagation attenuation factor of electromagnetic waves; N is the number of electromagnetic sources in the system;

[0019] The system health status is obtained by weighting the fusion data of multi-dimensional sensor data and the electromagnetic interference factor.

[0020] Further, the high-order statistical features include the skewness of the health status data, the kurtosis of the health status data, the standard deviation of the health status and the mean of the health status;

[0021] The high-order statistical feature identification system based on the health state uses a multidimensional high-order statistical feature classifier to identify the failure mode of the system; the multidimensional high-order statistical feature classifier identifies the failure mode by constructing a classifier based on a support vector machine.

[0022] Furthermore, the dynamic LSTM model is a dynamic LSTM structure based on an adaptive convolution gate; the dynamic LSTM structure based on an adaptive convolution gate can introduce a convolution layer into the LSTM model to dynamically weight the input features, and dynamically adjust the convolution kernel through a gating mechanism to cope with different system loads and electromagnetic interference conditions; the dynamic LSTM structure is expressed as:

[0023] h t =LSTM(X t ,W 1 )+ACG(X t ,W 2 )

[0024] Among them, h t is the dynamic LSTM model output at time t; X t As input features, it combines health status data and electromagnetic interference factors; W 1 is the weight parameter of the dynamic LSTM model; W 2 is the weight parameter of the convolution layer; ACG is the adaptive convolution gate, which is used to adjust the size and weight of the convolution kernel.

[0025] Furthermore, the time-varying disturbance model is introduced to perform dynamic electromagnetic interference regularization, wherein the time-varying disturbance model weights the electromagnetic interference by a dynamic weighting factor, and the weighting factor changes with time, thereby simulating the nonlinear time-varying characteristics of the electromagnetic interference effect.

[0026] Furthermore, the adaptive time window automatically adjusts the size of the time window according to the historical data of electromagnetic interference so as to capture the changing law of electromagnetic interference; the time series regression model takes the failure mode prediction and the historical data of electromagnetic interference as input to jointly predict the linkage effect of electromagnetic interference and failure mode; the time series regression model is expressed as:

[0027]

[0028] in, is the electromagnetic interference intensity predicted at time t+1; F predicted (t) is the failure mode prediction value, which indicates the failure risk at the current moment; is the mean value of electromagnetic interference at time t-τ; Var[I EM (t)] is the electromagnetic interference variance at time t; β 1 ,β 2 ,β 3 are the coefficients of the regression model, which are used to weight each input feature.

[0029] Furthermore, the prediction results of electromagnetic interference are combined with the failure mode prediction results to construct a joint prediction model, and the final system failure prediction is obtained, which is expressed as:

[0030]

[0031] Among them, F final (t) is the final system failure prediction result; F predicted (t) is the failure mode prediction; is the electromagnetic interference prediction; 3 ,λ 4 is the weight coefficient used to balance the contribution of failure mode prediction and electromagnetic interference prediction.

[0032] Furthermore, the adaptive fault-tolerant control algorithm is expressed as:

[0033]

[0034] Among them, u(t) is the real-time control signal used to adjust the system behavior; γ 1 ,γ 2 ,γ 3 are new adjustment coefficients, which represent the weight coefficients of failure mode prediction, interference prediction and early control signal adjustment respectively; F predicted (t) is the failure mode prediction value at time t, reflecting the failure risk of the system in the current state; is the electromagnetic interference prediction value at time t, indicating the possible impact of the electromagnetic environment on the system; Δu(t-1) is the change in the control signal at the previous moment, which is used to smooth the adjustment of the system and avoid system instability caused by excessive control;

[0035] The resource allocation problem of the dynamic resource allocation strategy is:

[0036]

[0037] Where R(t) represents the resource allocation vector of the system at time t; represents the system's performance function, which measures the performance of the current resource allocation under the conditions of failure risk and electromagnetic interference; F predicted (t) is the failure mode prediction value at the current moment, which is used to determine the additional resources required by the system when failure is about to occur; is the electromagnetic interference prediction value at the current moment, which is used to adjust the system resource configuration to adapt to the electromagnetic environment; Δu(t) represents the change of the control signal, which affects the system resource allocation decision.

[0038] Furthermore, the health assessment function is expressed as:

[0039]

[0040] in, is the health assessment value of the system. The larger the value, the higher the risk of failure of the system. predicted (t) is the failure mode prediction value, which indicates the possibility of potential failure of the system; is the electromagnetic interference prediction value, reflecting the impact of external interference on the system; Δu(t) is the real-time control signal change, indicating the system's adaptability to external disturbances; w(t) is the dynamic adjustment factor, which is adjusted based on the current operating status of the system and changes in the external environment; the value of w(t) is automatically adjusted as the system operates and the external environment changes. The highest w(t) means that the health of the system is most affected by the outside world.

[0041] Furthermore, the failure prevention mechanism is triggered based on the real-time value of the health assessment. If the threshold is exceeded, the redundant system startup or load adjustment measures are triggered; the feedback correction mechanism optimizes the execution of preventive measures by dynamically adjusting the system health assessment value.

[0042] In a second aspect of the present invention, a deep learning-based autonomous driving electronic system reliability prediction system is provided, the system comprising:

[0043] The signal acquisition module is used to establish an electromagnetic environment and component interaction model, generate system health status input through multi-dimensional sensor data fusion, and track electromagnetic interference sources using time-frequency analysis;

[0044] Failure prediction module, used to identify the failure mode of the system based on the high-order statistical characteristics of the health state; using the dynamic LSTM model to predict the evolution path of the failure mode of the system components, and introducing the time-varying disturbance model for dynamic electromagnetic interference regularization;

[0045] A joint analysis module is used to combine the failure mode prediction results, using adaptive time windows and time series regression models to predict the joint impact of electromagnetic interference on system failure;

[0046] A signal configuration module is used to adjust the system control signal and resource configuration through an adaptive fault-tolerant control algorithm and a dynamic resource allocation strategy based on failure mode prediction and electromagnetic interference prediction values;

[0047] The feedback optimization module is used to design a health assessment function in combination with real-time feedback, dynamically trigger a failure prevention mechanism based on the health assessment function, and optimize preventive measures through a feedback correction mechanism.

[0048] The beneficial technical effects of the present invention are at least as follows:

[0049] (1) The present invention proposes a new failure mode prediction method by combining deep learning with physical modeling. Deep learning models (such as LSTM, CNN) can dynamically predict the failure modes of various components of the system by analyzing historical fault data and real-time monitoring data, while physical modeling ensures the accuracy and reliability of model prediction. The model can make more accurate failure predictions based on real-time monitored environmental changes and system status, overcoming the shortcomings of static failure mode prediction methods in the prior art.

[0050] (2) The present invention innovatively combines electromagnetic interference identification with an adaptive fault-tolerant mechanism, and uses deep learning algorithms (such as CNN and GNN) to identify electromagnetic interference characteristics in real time, analyze the impact of interference sources on the system, and predict the failure modes that may be caused. Based on this prediction, the system can automatically adjust operating parameters (such as frequency, voltage, etc.) to reduce the impact of electromagnetic interference on system stability. This adaptive fault-tolerant mechanism based on interference prediction can effectively improve the reliability of the system in complex electromagnetic environments, and solves the problem that traditional electromagnetic interference response strategies cannot be adjusted in real time.

[0051] (3) The present invention proposes a dynamic optimization and system-level failure prevention strategy by combining real-time data collection and failure prediction results. The system will automatically adjust the working status of each module according to the state changes and fault predictions monitored in real time, perform fault-tolerant switching or enable redundancy functions, thereby effectively ensuring the continuous operation of the system when a fault occurs. This dynamic optimization mechanism overcomes the static and preset problems of traditional fault-tolerant strategies and improves the adaptability and stability of the system in complex environments.

[0052] (4) By introducing a hybrid intelligent model, electromagnetic interference identification and adaptive fault-tolerant mechanism, and dynamic optimization strategy, a new reliability prediction and fault-tolerant optimization method for autonomous driving electronic systems is provided. This method can not only accurately predict system failure modes, but also respond to the impact of external electromagnetic interference in real time, thereby improving the system's dynamic adaptability and fault-tolerant performance. Through these innovations, the present invention effectively solves the deficiencies of the prior art in terms of prediction accuracy, interference identification, and fault-tolerant mechanism, and provides a new technical path for improving the reliability of autonomous driving systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The present invention is further described using the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.

[0054] Figure 1 This is a schematic diagram of the steps of the deep learning-based autonomous driving electronic system reliability prediction method of the present invention. DETAILED DESCRIPTION

[0055] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0056] In one or more embodiments, Figure 1 As shown, a method for predicting reliability of an autonomous driving electronic system based on deep learning is disclosed, and the method comprises the following steps:

[0057] S1: Establish an electromagnetic environment and component interaction model, generate system health status input through multi-dimensional sensor data fusion, and use time-frequency analysis to track electromagnetic interference sources.

[0058] In this step, we first need to build a basic model that can reflect the status of the autonomous driving electronic system in real time through accurate modeling and data collection. The main task of this model is to capture the electromagnetic environment impact of the autonomous driving system during dynamic operation, as well as the working status of different electronic modules.

[0059] Specifically, the present invention constructs a system modeling-electromagnetic environment and component interaction model, including:

[0060] Electromagnetic interference (EMI) is a factor that cannot be ignored in autonomous driving systems, which may cause unexpected behavior of the system. Therefore, we first need to establish an EMI impact model to describe the interaction between electromagnetic sources and electronic components, and quantify the impact of each interference source on the system module through the model.

[0061] The model uses the following equations to describe the impact of each electromagnetic source on each module of the autonomous driving system:

[0062]

[0063] Among them, α k is the intensity coefficient of the kth electromagnetic source, reflecting the electromagnetic field intensity of the electromagnetic source. k (t) is the distance between the kth electromagnetic source and the target component at time t. As the system moves, the distance changes dynamically. k is the propagation attenuation factor of electromagnetic waves, which depends on factors such as the frequency of the electromagnetic source and the propagation medium. N is the number of electromagnetic sources in the system.

[0064] Through this model, the system can calculate the electromagnetic interference factor IF in real time at each moment. t , which will be used for subsequent failure mode prediction and electromagnetic interference identification.

[0065] Preferably, the autonomous driving system consists of multiple sensors that continuously collect environmental and system status data. To make full use of this data, we use multi-dimensional sensor data fusion technology to integrate data from different sensors into a unified health status representation:

[0066] Assume that the system consists of M modules, each with K m sensors, and the data collected by the sensor at each time t is X m (t) = [x m1 (t),x m2 (t),…,x mKm (t)]. By denoising, normalizing and preprocessing the sensor data of each module, we obtain the standardized data vector of each module Then, the data of all modules are fused into a comprehensive system state vector Df (t):

[0067]

[0068] Among them, ω mk is the weighting coefficient of the kth sensor in the mth module, reflecting the importance of the sensor in the comprehensive system health assessment; x mk (t) is the original data collected by the kth sensor of the mth module at time t; D f (t) is the fused global system state representation, which serves as the input for subsequent steps.

[0069] Through this weighted fusion method, we ensure that the data from different sensors are reasonably integrated and avoid the influence of noise and errors from sensors on the system evaluation results.

[0070] Preferably, in an autonomous driving system with a complex electromagnetic environment, the location and identification of electromagnetic interference sources is crucial. In order to accurately identify electromagnetic sources and track their changes, we propose an electromagnetic interference source tracking method based on time-frequency analysis.

[0071] First, the sensor data X m (t) Perform time-frequency analysis and use short-time Fourier transform (STFT) to extract the frequency features in the signal:

[0072]

[0073] Among them, X f (t,ω) is the frequency domain representation of the signal X(t) after short-time Fourier transform; ω is the frequency variable, reflecting the frequency components in the signal. Through this method, we can identify specific frequency bands in the signal, which usually correspond to the frequency characteristics of the electromagnetic interference source. Combined with the electromagnetic interference model (Section 1.1), we can dynamically adjust the electromagnetic interference factor (I EM (t)) and conduct precise tracking.

[0074] Preferably, based on the previous electromagnetic interference analysis and data fusion, we will generate preliminary input of the system health status in this step, and the input data will provide support for subsequent failure prediction and fault tolerance mechanisms. Since we do not do a specific health assessment at this time, but focus on providing raw data of the health status and electromagnetic interference factors, the core of this step is to generate input data for subsequent steps.

[0075] We will pass D f (t) and I EM(t) is comprehensively analyzed to generate the original health assessment input of the system status. Specifically, we weight the status data of each module of the system and the electromagnetic interference factor according to a certain weight to obtain a comprehensive health status index H t :

[0076]

[0077] Among them, H t is the original evaluation result of the system health status, which serves as the input for subsequent failure mode prediction; m is the weight coefficient of the mth module; is the standardized data of the mth module; α is the weighting coefficient of the electromagnetic interference factor. This comprehensive health status index H t It will serve as input to provide necessary data support for subsequent failure mode prediction, electromagnetic interference identification and implementation of fault-tolerant mechanisms.

[0078] Through this step, we have successfully established a real-time modeling framework based on electromagnetic interference and system status, and through real-time data collection and fusion, it has provided a solid foundation for subsequent failure prediction, electromagnetic interference identification, and fault-tolerant optimization.

[0079] S2: Identify the failure mode of the system based on the high-order statistical characteristics of the health state; use the dynamic LSTM model to predict the evolution path of the failure mode of the system components, and introduce a time-varying disturbance model for dynamic electromagnetic interference regularization.

[0080] Among them, in this step, our goal is to use the health status input (H t ), and electromagnetic interference (I EM ) and build a model that can accurately predict the failure mode of the system. We need not only to identify possible failure modes, but also to predict the evolution path of these failure modes in the future. This goal requires us to model the dynamic changes of failure modes and be able to cope with the complex nonlinear behavior of the system and the influence of the electromagnetic environment.

[0081] Specifically, dynamic failure mode identification is first performed (based on high-order statistical features):

[0082] The key to failure mode identification is to accurately extract high-order statistical information representing system failure characteristics from multi-dimensional health status data and identify failure modes based on these characteristics. In traditional methods, simple feature extraction and classification models are usually used, but for complex autonomous driving systems, simple features often cannot capture the full picture of system failures. Therefore, we propose an innovative "dynamic high-order statistical feature extraction" method.

[0083] We start from the health status data H t We can extract high-order statistics (such as skewness, kurtosis, etc.) from the data and combine them with time series analysis to obtain more complex nonlinear features. Through high-order statistics, we can capture more subtle volatility in the data, which often reflects the potential failure of the system.

[0084] We define “dynamic high-order statistical features” as Where t represents the time step, feature It is defined by the following formula:

[0085]

[0086] Among them, skew(H t ) is the skewness of health status data, reflecting the asymmetry of data distribution; Kurt (H t ) is the kurtosis of health status data, reflecting the sharpness of data distribution; std(H t ) is the standard deviation, describing the fluctuation range of health status;

[0087] mean(H t ) is the mean, indicating the average level of health status.

[0088] After extracting these high-order statistical features, we can use them as input features to identify the failure mode of the system through the multidimensional high-order statistical feature classifier (MHSFC). The MHSFC model identifies the failure mode by building a classifier based on a support vector machine (SVM):

[0089]

[0090] Among them, F predicted is the predicted failure mode; Dynamic high-order statistical features extracted from health status data.

[0091] Preferably, based on the failure mode identification, the next step is to perform time series prediction of the failure mode. The traditional LSTM model can capture the long-term dependencies of the time series, but in complex autonomous driving electronic systems, the failure mode is often affected by multiple nonlinear factors (such as temperature, electromagnetic interference, etc.), so we need to introduce a dynamic LSTM model to handle this complexity.

[0092] We propose a dynamic LSTM structure based on adaptive convolution gate (ACG), which can introduce convolution layers in LSTM to dynamically weight input features and dynamically adjust the convolution kernel through the gating mechanism to cope with different system loads and electromagnetic interference conditions. The formula is defined as:

[0093] h t =LSTM(X t ,W 1 )+ACG(X t ,W 2 ) (7)

[0094] Among them, h t is the LSTM output at time t; X t As input features, it combines health status data and electromagnetic interference factors; W 1 is the weight parameter of LSTM; W 2 is the weight parameter of the convolution layer; ACG is the adaptive convolution gate, which is used to adjust the size and weight of the convolution kernel. The function of the adaptive convolution gate is to enhance the prediction ability of failure modes in different electromagnetic interference environments by dynamically selecting the appropriate convolution kernel size and weight.

[0095] Through this innovative dynamic LSTM structure, we are able to accurately predict the evolution path of the failure mode in future time steps taking into account the influence of electromagnetic interference.

[0096] Preferably, a dynamic electromagnetic interference regularization is constructed (based on a time-varying disturbance model):

[0097] In order to further enhance the stability of the model, we introduce an innovative dynamic electromagnetic interference regularization mechanism. EM ) is an important factor affecting the reliability of autonomous driving systems, but its effect is often time-varying, so we cannot regard it as a static factor.

[0098] To this end, we define a time-varying disturbance model (TVDM) to describe the dynamic impact of electromagnetic interference on failure mode prediction. The core of the TVDM model is to weight the electromagnetic interference by a dynamic weighting factor, which changes with time, thereby simulating the nonlinear time-varying characteristics of the electromagnetic interference effect. The loss function of TVDM is defined as:

[0099]

[0100] Among them, α t I is the time-varying weighting factor of electromagnetic interference, which dynamically adjusts the sensitivity to electromagnetic interference; EM (t) is the electromagnetic interference factor at time t; is the electromagnetic interference factor predicted by the model. This regularization term will adjust the prediction results of the failure mode according to the real-time electromagnetic environment, so that the model can still maintain high accuracy in a changing electromagnetic environment.

[0101] Preferably, after completing the failure mode recognition and timing prediction model, we use a composite loss function to optimize the model. The composite loss function combines failure mode recognition, failure mode prediction and electromagnetic interference regularization to ensure the comprehensive performance of the model. The final training loss function is:

[0102] L final =L SVM +λ 1 L LSTM +λ 2 L EMI (9)

[0103] Among them, L SVM is the SVM loss for failure mode recognition; L LSTM LSTM loss for failure mode prediction; L EMI is the electromagnetic interference regularization loss; 1 ,λ 2 is the weight coefficient of each task.

[0104] S3: Combined with the failure mode prediction results, the adaptive time window and timing regression model are used to predict the joint impact of electromagnetic interference on system failure.

[0105] In step 2, we have obtained the predicted value F of the failure mode. predicted and electromagnetic interference raw data I EM In this step, our goal is to combine the two to identify the strength of EMI and accurately predict its impact on system failure. To achieve this goal, we propose an innovative EMI Timing Analysis and Impact Prediction Method (EMI-TAPM), which combines timing analysis technology with regression models to effectively capture the time dependence and nonlinear characteristics of EMI.

[0106] Specifically, electromagnetic interference is time-varying, and its impact will fluctuate significantly in different time periods, so we need to use time series modeling methods to capture the dynamic characteristics of electromagnetic interference. To this end, we introduced the Adaptive Time Window (ATW), the core idea of ​​which is to automatically adjust the size of the time window based on the historical data of electromagnetic interference in order to capture the changing laws of electromagnetic interference. The adjustment formula of the time window is as follows:

[0107]

[0108] Among them, ΔtEMI (t) represents the dynamic time window at time t, including the mean and variance of electromagnetic interference; is the mean value of electromagnetic interference at time τ in the past, reflecting the impact of historical electromagnetic interference; Var[I EM (t)] is the variance of electromagnetic interference at the current moment, describing the fluctuation of electromagnetic interference at the current moment. By adaptively adjusting the time window, the model can better capture the time-varying characteristics of electromagnetic interference and provide dynamic window data for subsequent prediction processes.

[0109] Preferably, in order to effectively predict the future trend of electromagnetic interference, combined with the failure mode prediction (F predicted ) for the final system failure assessment, we designed an innovative Time-Series Regression Model (TSRM) that performs joint prediction based on past EMI data and failure mode predictions.

[0110] Specifically, we use a multivariate regression model to predict the failure mode (F predicted ) and historical data of electromagnetic interference (I EM ) as input to build a model to predict the impact of future electromagnetic interference. The formula of the regression model is as follows:

[0111]

[0112] in, is the electromagnetic interference intensity predicted at time t+1; F predicted (t) is the failure mode prediction value obtained in step 2, which represents the failure risk at the current moment; is the mean value of electromagnetic interference at time t-τ;

[0113] Var[I EM (t)] is the electromagnetic interference variance at time t; β 1 ,β 2 ,β 3 are the coefficients of the regression model, which are used to weight each input feature.

[0114] Through joint regression analysis, we can combine EMI and failure mode prediction to obtain the intensity of future EMI and provide a reliable basis for subsequent failure assessment.

[0115] Preferably, in order to achieve the best effect of the electromagnetic interference prediction model, we use a composite loss function to simultaneously optimize the accuracy and stability of the timing regression model. The composite loss function takes into account the error of electromagnetic interference prediction and the influence of failure mode, and the form is:

[0116] L final =LEMI +λ 1 L regression +λ 2 L stability (12)

[0117] Among them, L EMI is the prediction error of electromagnetic interference; L regression is the regression error of the time series regression model; L stability is the stability loss of the model during training, used to prevent overfitting; λ 1 ,λ 2 is the weight coefficient in the loss function, which is used to balance various tasks.

[0118] By optimizing the loss function, we can better balance the EMI prediction and failure mode prediction, thereby improving the robustness and stability of the system.

[0119] Preferably, at the end of this step, we combine the electromagnetic interference prediction results with the failure mode prediction results to obtain the final system failure prediction. We introduce a joint prediction model (F final (t)):

[0120]

[0121] Among them, F final (t) is the final system failure prediction result; F predicted (t) is the failure mode prediction obtained in step 2; is the electromagnetic interference prediction obtained in step 3; 3 ,λ 4 is the weight coefficient used to balance the contribution of failure mode prediction and electromagnetic interference prediction.

[0122] Through this joint model, the system can comprehensively consider the impact of electromagnetic interference and the risk of failure modes, provide more accurate failure predictions, and provide a reliable basis for subsequent fault-tolerant decisions.

[0123] Through this method, the system can predict the intensity of electromagnetic interference in real time and analyze its impact on the system, thereby providing accurate decision-making basis for the fault-tolerant mechanism in future electromagnetic interference environments, and improving the reliability and safety of the autonomous driving system in complex environments.

[0124] S4: Based on the failure mode prediction and electromagnetic interference prediction value, the system control signal and resource configuration are adjusted through adaptive fault-tolerant control algorithm and dynamic resource allocation strategy.

[0125] One of the core tasks of this step is to feed back the prediction data of electromagnetic interference and failure mode into the system control decision in real time to dynamically adjust the control parameters and prevent the system from failing prematurely. To achieve this goal, we propose an adaptive fault-tolerant control algorithm based on prediction data, which incorporates the prediction values ​​of failure mode and electromagnetic interference into the decision-making process in a weighted manner.

[0126] Specifically, based on the prediction information obtained in the previous step, we designed the following control algorithm to adjust the control signal in real time:

[0127]

[0128] Among them, u(t) is a real-time control signal used to adjust system behavior (for example, adjusting the power, load, working mode, etc. of the equipment). 1 ,γ 2 ,γ 3 are new adjustment coefficients, which represent the weight coefficients of failure mode prediction, interference prediction and previous control signal adjustment. predicted (t) is the failure mode prediction value at time t, reflecting the failure risk of the system in the current state; is the electromagnetic interference prediction value at time t, indicating the possible impact of the electromagnetic environment on the system. Δu(t-1) is the change in the control signal at the previous moment, which is used to smooth the adjustment of the system and avoid system instability caused by excessive control.

[0129] The key to this control algorithm is to dynamically update the control signal of the system using the predicted value and historical adjustment information. For example, under high failure mode risk (F predicted (t) is large), the system will reduce the probability of failure by reducing power or adjusting load; while in a high electromagnetic interference environment ( If the signal processing capability is large, the system may strengthen the signal processing capability or ensure stable operation through redundant design.

[0130] Preferably, in addition to fault-tolerant control, another core task of this step is to optimize the allocation of system resources in real time to ensure that the remaining resources can be optimally utilized when a fault occurs. To achieve this goal, we introduced a real-time optimization algorithm based on system performance, which uses the system's electromagnetic interference and failure mode prediction to dynamically adjust resource allocation to ensure the system's long-term stability and short-term performance in the best way.

[0131] We define the system resource allocation problem as:

[0132]

[0133] Among them, R(t) is the resource allocation vector representing the system at time t, which usually includes energy, computing power and redundant resources. To represent the system's effectiveness function, the performance of the current resource allocation under the conditions of failure risk and electromagnetic interference is measured. It can be a multi-objective function, taking into account the system's stability, response speed, energy efficiency and other dimensions. predicted (t) is the predicted value of the failure mode at the current moment, which is used to determine the additional resources required by the system when a failure is about to occur. is the electromagnetic interference prediction value at the current moment, which is used to adjust the system resource configuration to adapt to the electromagnetic environment. Δu(t) represents the change of the control signal in the previous step, which affects the system resource allocation decision.

[0134] In practical applications, the goal of this optimization problem is to maximize the overall performance of the system in the face of failures and interference by adjusting R(t) (for example, increasing computing resources, adding redundant systems, etc.). For example, the system can be based on F predicted Adjust redundancy, or according to Enhance signal processing capabilities and optimize resource allocation to keep the system running efficiently.

[0135] Preferably, as the system state changes, the fault tolerance mechanism needs to be dynamically adjusted to respond to different operating environments and failure modes. To achieve this, we designed a dynamic fault tolerance correction mechanism that corrects the previous fault tolerance strategy through real-time feedback and fine-tunes the system.

[0136] The fault tolerance adjustment formula is as follows:

[0137]

[0138] Where, ΔF adjusted (t) is the failure mode prediction value after dynamic adjustment, indicating the corrected fault-tolerant decision. 1 ,δ 2 ,δ 3 ,δ 4 is a new adjustment coefficient used to adjust the impact of failure prediction, interference prediction, control signal change and feedback information. predicted (t) is the failure mode prediction value at the current moment, which is used to adjust the fault tolerance strategy. is the electromagnetic interference prediction value at the current moment, which affects the fault-tolerant decision of the system. Δu(t) is the change of the system control signal, which indicates the adjustment range of the adaptive control. feedback (t) is the system feedback signal, which indicates the correction value of the system failure mode during actual operation.

[0139] Through this dynamic adjustment formula, the system can adjust its fault tolerance mechanism in real time according to the prediction of electromagnetic interference and failure mode, avoiding over-correction or slow response of the system. feedback (t) is an indispensable part of the dynamic system. It corrects the fault-tolerant decision by the difference with the predicted value, so that the system always maintains the best performance.

[0140] Preferably, in order to further improve the stability and adaptability of the system, we designed a multi-objective optimization algorithm, which is not only based on the prediction of failure modes and electromagnetic interference, but also includes multiple optimization objectives of the system's long-term stability, short-term effectiveness and energy efficiency. By introducing a multi-objective feedback loop, the system can automatically select the optimal strategy in different working scenarios.

[0141] Multi-objective optimization can be expressed in the following form:

[0142]

[0143] Wherein, R(t) is the system resource allocation vector. It is a multi-objective optimization function, which means optimizing multiple objectives such as system performance, stability and energy efficiency at the same time.

[0144] Through this optimization model, the system can adjust resource allocation and control strategies according to environmental changes (including electromagnetic interference and failure prediction) at every moment, ultimately achieving long-term stability and short-term efficient operation of the system.

[0145] S5: Design a health assessment function based on real-time feedback, dynamically trigger the failure prevention mechanism based on the health assessment function, and optimize the preventive measures through the feedback correction mechanism.

[0146] Specifically, the core goal of this step is to conduct system-level health assessment and failure prevention based on the failure modes and electromagnetic interference information predicted in the previous steps, combined with real-time feedback. We will identify potential failure risks by evaluating the overall health of the system and take necessary preventive measures based on the evaluation results.

[0147] The input of this step is:

[0148] Failure mode prediction value F predicted (t);

[0149] Electromagnetic interference prediction value

[0150] Real-time control signal change Δu(t).

[0151] These inputs will provide the necessary parameters for the health assessment model to evaluate the health status of the system in real time and take timely intervention measures when there is a risk.

[0152] Preferably, the health assessment function is designed as follows:

[0153] System health assessment function It not only reflects the influence of failure modes and electromagnetic interference, but also considers the impact of control signal changes on system health. The design takes into account the following key factors:

[0154] Failure mode prediction value F of the system predicted (t);

[0155] Electromagnetic interference effects on the system

[0156] The change of the real-time control signal Δu(t(, which reflects the system's ability to respond to external interference.

[0157] Innovatively, we introduced a dynamic weight adjustment factor w*t), which automatically adjusts the weight of each factor according to the current operating status of the system and changes in the external environment, so as to more accurately evaluate the health status of the system. The core formula of the health assessment function is:

[0158]

[0159] in, is the health assessment value of the system. The larger the value, the higher the risk of failure of the system. predicted (t) is the failure mode prediction value, which indicates the possibility of potential failure of the system. is the electromagnetic interference prediction value, reflecting the impact of external interference on the system. Δu*t(is the real-time control signal change, indicating the system's adaptability to external disturbances. w(t) is a dynamic adjustment factor, which is adjusted based on the current operating status of the system and changes in the external environment. The value of w(t) is automatically adjusted as the system operates and the external environment changes. A higher w(t) means that the health of the system is greatly affected by the outside world, while a lower w(t) means that the system is relatively stable.

[0160] This innovative design enables the health assessment model to respond more flexibly to different external environmental changes and fluctuations in system status, thereby more accurately assessing health risks.

[0161] Preferably, once the system health assessment value Exceeds a preset threshold The system needs to trigger the failure prevention mechanism. In this step, we propose a trigger mechanism based on dynamic health assessment and introduce a special prevention adjustment factor k(t), which dynamically adjusts the strength of the preventive measures based on the rate of change of the current health assessment value of the system.

[0162] The formula for the failure prevention trigger mechanism is as follows:

[0163]

[0164] Where T(t) is the failure prevention trigger signal. Trigger the preventive action; otherwise, do not trigger it. The preset threshold of the health assessment value determines whether to trigger the prevention mechanism. The triggering of failure prevention is based on the real-time value of the health assessment. If the system health status is poor (i.e. If the load exceeds the threshold, preventive measures are triggered, such as reducing the load and starting the redundant system.

[0165] Preferably, to ensure the stability and long-term healthy operation of the system, we introduce a feedback correction mechanism to optimize the execution of preventive measures by dynamically adjusting the system health assessment value. In this mechanism, we compare the current health assessment value with the past assessment value and adapt to the changes in the system health status through the correction factor δ(t). The feedback correction formula is:

[0166]

[0167] in, is the corrected health assessment value, which reflects the evaluation of the current health status of the system after considering the past health change trend. δ(t) is the feedback correction factor, which indicates the correction strength of the health assessment change rate. It is the time derivative of the health assessment value, indicating the rate of change of the health assessment value.

[0168] By introducing a feedback correction mechanism, the system can dynamically adjust the intensity of preventive measures based on the changing trend of the health assessment value. This approach avoids triggering prevention based only on the current health assessment value, but takes into account the changing trend of the system status, thereby achieving more accurate failure prevention.

[0169] In one or more implementations, another embodiment of the present invention provides an autonomous driving electronic system reliability prediction system based on deep learning, the system comprising:

[0170] The signal acquisition module is used to establish an electromagnetic environment and component interaction model, generate system health status input through multi-dimensional sensor data fusion, and track electromagnetic interference sources using time-frequency analysis;

[0171] Failure prediction module, used to identify the failure mode of the system based on the high-order statistical characteristics of the health state; using the dynamic LSTM model to predict the evolution path of the failure mode of the system components, and introducing the time-varying disturbance model for dynamic electromagnetic interference regularization;

[0172] A joint analysis module is used to combine the failure mode prediction results, using adaptive time windows and time series regression models to predict the joint impact of electromagnetic interference on system failure;

[0173] A signal configuration module is used to adjust the system control signal and resource configuration through an adaptive fault-tolerant control algorithm and a dynamic resource allocation strategy based on failure mode prediction and electromagnetic interference prediction values;

[0174] The feedback optimization module is used to design a health assessment function in combination with real-time feedback, dynamically trigger a failure prevention mechanism based on the health assessment function, and optimize preventive measures through a feedback correction mechanism.

[0175] In summary, the present invention provides a new reliability prediction and fault tolerance optimization method for autonomous driving electronic systems by introducing a hybrid intelligent model, electromagnetic interference identification and adaptive fault tolerance mechanism, and dynamic optimization strategy. This method can not only accurately predict system failure modes, but also respond to the impact of external electromagnetic interference in real time, and improve the dynamic adaptability and fault tolerance performance of the system. Through these innovations, the present invention effectively solves the deficiencies of the prior art in terms of prediction accuracy, interference identification and fault tolerance mechanism, and provides a new technical path for improving the reliability of autonomous driving systems.

[0176] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0177] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0178] Although embodiments of the present invention have been shown and described, those skilled in the art will appreciate that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

[0179] The various illustrative logic blocks, modules, and circuits described in conjunction with the embodiments disclosed herein may be implemented or performed with a general purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. A general purpose processor may be a microprocessor, but in the alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in cooperation with a DSP core, or any other such configuration.

[0180] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. The software module may reside in a RAM memory, a flash memory, a ROM memory, an EPROM memory, an EEPROM memory, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to a processor so that the processor can read and write information from / to the storage medium. In an alternative, a storage medium may be integrated into a processor. The processor and the storage medium may reside in an ASIC. The ASIC may reside in a user terminal. In an alternative, the processor and the storage medium may reside in a user terminal as discrete components.

[0181] In one or more exemplary embodiments, the functions described may be implemented in hardware, software, firmware, or any combination thereof. If implemented as a computer program product in software, each function may be stored on or transmitted by a computer-readable medium as one or more instructions or codes. Computer-readable media include both computer storage media and communication media, including any medium that facilitates the transfer of a computer program from one place to another. Storage media may be any available medium that can be accessed by a computer. As an example and not limitation, such a computer-readable medium may include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, disk storage or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of an instruction or data structure and can be accessed by a computer. Any connection is also properly referred to as a computer-readable medium. For example, if the software is transmitted from a website, a server, or other remote source using a coaxial cable, a fiber optic cable, a twisted pair, a digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwaves, the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwaves are included in the definition of the medium. Disk and disc as used herein include compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk and Blu-ray disc, wherein disk often reproduces data magnetically, while disc reproduces data optically with lasers. Combinations of the above should also be included within the scope of computer-readable media.

[0182] The previous description of the disclosure is provided to enable any person skilled in the art to make or use the disclosure. Various modifications to the disclosure will be apparent to those skilled in the art, and the general principles defined herein may be applied to other variations without departing from the spirit or scope of the disclosure. Thus, the disclosure is not intended to be limited to the examples and designs described herein, but should be granted the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A reliability prediction method for autonomous driving electronic systems based on deep learning, characterized in that: The following steps are involved: S1: Establish an electromagnetic environment and component interaction model, generate system health status input through multi-dimensional sensor data fusion, and use time-frequency analysis to track electromagnetic interference sources; S2: Identify the failure mode of the system based on the high-order statistical features of the health state; use the dynamic LSTM model to predict the evolution path of the failure mode of the system components, and introduce a time-varying disturbance model for dynamic electromagnetic interference regularization; S3: Combined with the failure mode prediction results, adaptive time window and time series regression model are used to predict the joint impact of electromagnetic interference on system failure; S4: Based on the failure mode prediction and electromagnetic interference prediction value, the system control signal and resource configuration are adjusted through adaptive fault-tolerant control algorithm and dynamic resource allocation strategy; S5: Design a health assessment function based on real-time feedback, dynamically trigger the failure prevention mechanism based on the health assessment function, and optimize the preventive measures through the feedback correction mechanism.

2. The deep learning-based reliability prediction method for autonomous driving electronic systems according to claim 1, characterized in that: The electromagnetic environment and component interaction model quantifies the impact of each interference source on the system module through the electromagnetic interference factor; the electromagnetic interference factor is expressed as: Among them, I EM (t) is the electromagnetic interference factor, α k is the intensity coefficient of the kth electromagnetic source, reflecting the electromagnetic field intensity of the electromagnetic source; r k (t) is the distance between the kth electromagnetic source and the target component at time t. As the system moves, the distance changes dynamically; λ k is the propagation attenuation factor of electromagnetic waves; N is the number of electromagnetic sources in the system; The system health status is obtained by weighted calculation based on fusion data of multi-dimensional sensor data and electromagnetic interference factors.

3. The deep learning-based reliability prediction method for autonomous driving electronic systems according to claim 1, characterized in that: The high-order statistical characteristics include the skewness of the health status data, the kurtosis of the health status data, the standard deviation of the health status and the mean of the health status; Then the high-order statistical feature identification system based on the health state uses a multi-dimensional high-order statistical feature classifier to identify the failure mode of the system; The multi-dimensional high-order statistical feature classifier identifies the failure mode by constructing a classifier based on a support vector machine.

4. The deep learning-based reliability prediction method for autonomous driving electronic systems according to claim 3, characterized in that: The dynamic LSTM model is a dynamic LSTM structure based on an adaptive convolution gate; the dynamic LSTM structure based on an adaptive convolution gate can introduce a convolution layer into the LSTM model to dynamically weight the input features, and dynamically adjust the convolution kernel through a gating mechanism to cope with different system loads and electromagnetic interference conditions; the dynamic LSTM structure is expressed as: h t =LSTM(X t ,W1)+ACG(X t ,W2) Among them, h t is the dynamic LSTM model output at time t; X t is the input feature, which combines the health status data and electromagnetic interference factor; W1 is the weight parameter of the dynamic LSTM model; W2 is the weight parameter of the convolution layer; ACG is the adaptive convolution gate, which is used to adjust the size and weight of the convolution kernel.

5. The deep learning-based reliability prediction method for autonomous driving electronic systems according to claim 4, characterized in that: The time-varying disturbance model is introduced to perform dynamic electromagnetic interference regularization, wherein the time-varying disturbance model weights the electromagnetic interference through a dynamic weighting factor, and the weighting factor changes with time, thereby simulating the nonlinear time-varying characteristics of the electromagnetic interference effect.

6. The deep learning-based autonomous driving electronic system reliability prediction method according to claim 1, characterized in that: The adaptive time window automatically adjusts the size of the time window according to the historical data of electromagnetic interference so as to capture the changing law of electromagnetic interference; the time series regression model takes the failure mode prediction and the historical data of electromagnetic interference as input to jointly predict the linkage effect of electromagnetic interference and failure mode; the time series regression model is expressed as: in, is the electromagnetic interference intensity predicted at time t+1; F predicted (t) is the failure mode prediction value, which indicates the failure risk at the current moment; is the mean value of electromagnetic interference at time t-τ; Var[I EM (t)] is the electromagnetic interference variance at time t; β1, β2, and β3 are the coefficients of the regression model, which are used to weight each input feature.

7. The deep learning-based reliability prediction method for autonomous driving electronic systems according to claim 6, characterized in that: The prediction results of electromagnetic interference are combined with the failure mode prediction results to build a joint prediction model, and the final system failure prediction is obtained, which is expressed as: Among them, F final (t) is the final system failure prediction result; F predicted (t) is the failure mode prediction; is the electromagnetic interference prediction; λ3, λ4 are weight coefficients used to balance the contribution of failure mode prediction and electromagnetic interference prediction.

8. The deep learning-based reliability prediction method for autonomous driving electronic systems according to claim 7, characterized in that: The adaptive fault-tolerant control algorithm is expressed as: Among them, u(t) is the real-time control signal used to adjust the system behavior; γ1, γ2, γ3 are new adjustment coefficients, which represent the weight coefficients of failure mode prediction, interference prediction and previous control signal adjustment respectively; F predicted (t) is the failure mode prediction value at time t, reflecting the failure risk of the system in the current state; is the electromagnetic interference prediction value at time t, indicating the possible impact of the electromagnetic environment on the system; Δu(t-1) is the change in the control signal at the previous moment, which is used to smooth the adjustment of the system and avoid system instability caused by excessive control; The resource allocation problem of the dynamic resource allocation strategy is: Where R(t) represents the resource allocation vector of the system at time t; represents the system's performance function, which measures the performance of the current resource allocation under the conditions of failure risk and electromagnetic interference; F predicted (t) is the failure mode prediction value at the current moment, which is used to determine the additional resources required by the system when failure is about to occur; is the electromagnetic interference prediction value at the current moment, which is used to adjust the system resource configuration to adapt to the electromagnetic environment; Δu(t) represents the change of the control signal, which affects the system resource allocation decision.

9. The deep learning-based reliability prediction method for autonomous driving electronic systems according to claim 8, characterized in that: The health assessment function is expressed as: in, is the health assessment value of the system. The larger the value, the higher the risk of failure of the system. predicted (t) is the failure mode prediction value, which indicates the possibility of potential failure of the system; is the electromagnetic interference prediction value, reflecting the impact of external interference on the system; Δu(t) is the real-time control signal change, indicating the system's adaptability to external disturbances; w(t) is a dynamic adjustment factor, which is adjusted based on the current operating state of the system and changes in the external environment; the value of w(t) is automatically adjusted as the system operates and the external environment changes. The highest w(t) means that the health of the system is most affected by the outside world. The failure prevention mechanism is triggered based on the real-time value of the health assessment. If the threshold is exceeded, the redundant system startup or load adjustment measures are triggered; the feedback correction mechanism optimizes the execution of preventive measures by dynamically adjusting the system health assessment value.

10. The reliability prediction system of autonomous driving electronic system based on deep learning is characterized by: The system comprises: The signal acquisition module is used to establish an electromagnetic environment and component interaction model, generate system health status input through multi-dimensional sensor data fusion, and track electromagnetic interference sources using time-frequency analysis; Failure prediction module, used to identify the failure mode of the system based on the high-order statistical characteristics of the health state; using the dynamic LSTM model to predict the evolution path of the failure mode of the system components, and introducing the time-varying disturbance model for dynamic electromagnetic interference regularization; Joint analysis module, which combines the failure mode prediction results and uses adaptive time windows and time series regression models to predict the joint impact of electromagnetic interference on system failure; A signal configuration module is used to adjust the system control signal and resource configuration through an adaptive fault-tolerant control algorithm and a dynamic resource allocation strategy based on failure mode prediction and electromagnetic interference prediction values; The feedback optimization module is used to design a health assessment function in combination with real-time feedback, dynamically trigger a failure prevention mechanism based on the health assessment function, and optimize preventive measures through a feedback correction mechanism.

Images