Bridge deterioration prediction method and system based on quantum computing and deep reinforcement learning

Abstract

This invention discloses a bridge degradation prediction method based on quantum computing and deep reinforcement learning, belonging to the field of structural health monitoring technology. The steps are as follows: 1. Collect and preprocess bridge stress, vibration, temperature, humidity, and crack data; 2. Establish a multi-scale nonlinear degradation model, fusing local and global degradation rates to obtain the degradation state; 3. Construct a deep reinforcement learning model, using a reward function that balances prediction accuracy and maintenance cost, and combine multi-objective optimization to generate the optimal maintenance decision; 4. Utilize the parallelism of quantum computing to accelerate numerical integration and degradation prediction model parameter optimization; 5. Generate early warning signals based on real-time monitoring data and adaptively adjust the degradation prediction model parameters. This invention improves the computational efficiency and prediction accuracy of degradation prediction through the synergy of quantum computing and deep reinforcement learning, achieving a multi-objective balance between maintenance cost and structural safety, and is suitable for real-time health monitoring and intelligent maintenance decision-making for bridge structures.

Description

Technical Field

[0001] This invention belongs to the field of bridge health monitoring and degradation prediction, specifically involving a bridge degradation prediction method and system based on quantum computing and deep reinforcement learning. Background Technology

[0002] With the accelerating pace of urbanization, bridges, as a crucial component of infrastructure, play a vital role in transportation and socio-economic activities. However, with the increasing age of bridges and the impact of environmental factors, bridge degradation has become increasingly apparent, posing growing challenges to bridge maintenance and management. Effective bridge health monitoring and degradation prediction are key to ensuring bridge safety, extending their service life, and reducing maintenance costs.

[0003] Currently, bridge health monitoring and degradation prediction technologies mainly include the following methods:

[0004] (1) Sensor-based bridge health monitoring method

[0005] Traditional bridge health monitoring systems typically collect real-time data from bridges by deploying sensors (such as strain gauges, accelerometers, and displacement sensors). This data is usually used to monitor physical quantities such as stress, displacement, and vibration, as well as the effects of environmental changes (such as temperature and humidity). Long-term tracking and analysis of sensor data can provide a preliminary assessment of the bridge's health status. While existing sensor monitoring methods can provide real-time data support to some extent, they have the following drawbacks:

[0006] Insufficient data processing capabilities: With the increase in the amount of bridge monitoring data, traditional methods face significant challenges in data processing and analysis, making it difficult to obtain accurate degradation trends in real time.

[0007] Lack of dynamic adaptability: Traditional methods rely on pre-defined analytical models and lack the ability to dynamically adjust to real-time data, and cannot self-optimize when data or the environment changes.

[0008] (2) Degradation prediction method based on finite element analysis (FEM)

[0009] The finite element method (FEM) is widely used in the numerical simulation of bridge degradation. By establishing a mathematical model of the bridge structure, it simulates the effects of various loads and environmental factors on the bridge structure. These models can predict the degradation process of bridges under different conditions. However, the finite element method also has the following problems:

[0010] High computational complexity: Finite element models typically require a large amount of computational resources, especially when dealing with complex structures or the influence of multiple factors, resulting in long computation times and difficulty in meeting real-time requirements.

[0011] Limited ability to handle complex nonlinear behavior: The degradation process of bridges is usually highly nonlinear, and traditional finite element analysis methods are difficult to accurately simulate and predict when faced with highly complex degradation behavior.

[0012] (3) Degradation prediction method based on machine learning

[0013] In recent years, machine learning algorithms (such as support vector machines and neural networks) have been widely applied to bridge health monitoring and degradation prediction. Machine learning methods learn degradation patterns from historical data and build regression models for prediction. While machine learning methods have improved prediction accuracy to some extent, some problems still exist:

[0014] Reliance on large amounts of labeled data: Machine learning models typically require a large amount of historical data for training, and the data quality requirements are high. For degradation prediction, accurately labeling degradation state data is very difficult.

[0015] Lack of adaptability: Most existing machine learning methods are trained on static data, which makes it difficult to cope with changes in real-time data and lacks the ability to dynamically adjust.

[0016] (4) Prediction methods based on physical models and data fusion

[0017] To address the shortcomings of the aforementioned methods, prediction methods based on physical models and data fusion have emerged in recent years. These methods combine physical models with sensor data, using data assimilation techniques to predict degradation. While this approach can improve prediction accuracy to some extent, it also has some limitations:

[0018] Limitations of model assumptions: Physical models usually rely on certain assumptions and it is difficult to fully consider all factors that may affect degradation (such as different environmental conditions, load changes, etc.).

[0019] High computational overhead: Data assimilation techniques typically require a large amount of computational resources, especially when dealing with large-scale, multi-dimensional data, where computational complexity is high.

[0020] Although the aforementioned existing technologies have made some progress in predicting bridge degradation, they still have the following major drawbacks and problems:

[0021] Poor real-time performance: Most existing prediction methods are trained based on static or historical data, lacking dynamic adjustment and optimization of real-time monitoring data, and thus cannot provide real-time and accurate degradation predictions.

[0022] High computational complexity: Traditional finite element analysis methods and physical models require a large amount of computational resources, especially when dealing with multi-dimensional and large-scale data. The computation time is long and cannot meet the needs of real-time early warning and decision-making.

[0023] Lack of adaptability: Existing methods cannot dynamically adjust based on real-time data, especially under the influence of factors such as environmental changes and traffic load fluctuations, and cannot effectively cope with the complexity and nonlinearity of bridge degradation processes.

[0024] Insufficient multi-objective optimization: Most existing technologies focus only on optimizing a single objective (such as prediction accuracy or cost), neglecting multi-objective optimization (such as degradation prediction accuracy, maintenance cost, bridge safety, etc.), and thus failing to provide comprehensive decision support. Summary of the Invention

[0025] To address the problems of existing technologies, this invention provides a bridge degradation prediction method and system based on quantum computing and deep reinforcement learning. The aim is to significantly improve the computational speed of the degradation prediction model by utilizing real-time, multi-dimensional monitoring data combined with quantum computing and deep reinforcement learning algorithms. Furthermore, it enables adaptive learning and dynamic optimization based on real-time monitoring data, thereby providing more accurate and real-time prediction results. A multi-objective optimization mechanism is also introduced, which balances multiple objectives such as maintenance costs and bridge safety while optimizing prediction accuracy, ensuring the comprehensiveness and scientific nature of bridge health management. This effectively solves the problems of poor real-time performance, high computational complexity, and lack of adaptive capabilities in existing technologies, providing more accurate and efficient degradation prediction and maintenance decision support.

[0026] To achieve the above objectives, the specific solution of the present invention is as follows:

[0027] A bridge degradation prediction method based on quantum computing and deep reinforcement learning includes the following steps:

[0028] Step 1, Data Acquisition and Preprocessing: Strain gauges, accelerometers, temperature and humidity sensors and crack sensors are installed at key parts of the bridge to collect real-time monitoring data on stress, vibration acceleration, temperature, humidity and cracks. The collected real-time monitoring data is unified into vector form and then subjected to noise reduction, data filling and elimination of different dimensions. The time domain features, frequency domain features and spatial features of the processed data are extracted as feature vectors.

[0029] Step 2, Multi-scale nonlinear degradation modeling: Using the feature vectors described in Step 1 as external input variables, establish a local degradation model and a global degradation model based on nonlinear dynamic equations, solve for the local degradation rate and the global degradation rate of the bridge respectively, fuse the local degradation rate and the global degradation rate through weight coefficients to obtain the total degradation rate, and use the numerical integration method to solve for the degradation state of the bridge.

[0030] Step 3, Deep Reinforcement Learning and Multi-Objective Optimization: Based on the feature vectors described in Step 1 and the degradation state of the bridge described in Step 2, a state space is constructed. The maintenance decision is used as the action space, and the balance between prediction accuracy and maintenance cost is used as the reward function. A degradation prediction model based on the degradation state of the bridge described in Step 2 is established. The degradation prediction model is optimized through deep reinforcement learning. A multi-objective function is constructed with prediction accuracy, maintenance cost, and structural safety. Multi-objective optimization techniques are used to optimize the multi-objective function to generate the optimal maintenance decision.

[0031] Step 4, Quantum Computing and Model Acceleration: Quantum computing is used to solve the equations of the numerical integration method described in Step 2 and to optimize the parameters and multi-objective functions of the deep reinforcement learning described in Step 3, utilizing quantum parallelism to process high-dimensional data;

[0032] Step 5, Real-time Feedback and Intelligent Decision Support: Based on the real-time monitoring data described in Step 1, input it into the degradation prediction model described in Step 3. Combine the degradation state described in Step 2 and the optimal maintenance decision described in Step 3 to generate an early warning signal. Based on the real-time monitoring data described in Step 1, adaptively adjust the parameters of the degradation prediction model described in Step 3 through deep reinforcement learning described in Step 3.

[0033] Furthermore, the formula for unifying the collected data into vector form as described in step 1 is as follows:

[0034] (1),

[0035] In the formula, Indicates time Data record vector; Indicates time The stress value; Indicates time The vibration acceleration value; Indicates time Temperature value; Indicates time The humidity value; Indicates time Crack data;

[0036] The denoising process includes Kalman filtering for removing noise from the sensor signal and wavelet transform for removing high-frequency noise. The formula for Kalman filtering is as follows:

[0037] (2),

[0038] In the formula: Indicates time The estimated state; This represents the estimated state at time t-1; This represents the Kalman gain, which controls the tradeoff between predicted and actual observations. Indicates time The actual measured value; The measurement matrix represents the relationship between measurements and states.

[0039] The data imputation includes linear interpolation for handling simple missing data cases and spline interpolation for filling more complex missing data cases. The formula for linear interpolation is as follows:

[0040] (3),

[0041] In the formula: Indicates time The interpolated value; and The time for the known data points;

[0042] The process of eliminating different dimensions employs standardization or normalization. Standardization is used to convert the data into a standard normal distribution with a mean of 0 and a standard deviation of 1, as shown in the following formula:

[0043] (4),

[0044] In the formula: This is the original data; The mean of the data; The standard deviation of the data;

[0045] Normalization is used to map the range of data values ​​to the interval [0, 1]. The formula is as follows:

[0046] (5),

[0047] In the formula: This is the original data; and These are the minimum and maximum values ​​of the data.

[0048] Furthermore, the nonlinear dynamic equation described in step 2 is as follows:

[0049] (8),

[0050] In the formula: Indicates the bridge degradation state at time [time]. The rate of change of , i.e. the degradation rate, represents the rate at which the bridge degrades at this moment; Let be the degraded state vector of the bridge, representing the health state of the bridge; The nonlinear dynamic equations describing the bridge degradation process reflect the relationship between the degradation state and external factors such as load and temperature and humidity. External input variables represent environmental factors that affect bridge degradation, such as load, temperature and humidity, and traffic flow. The random perturbation term of the system represents the uncertainty in the degradation process, taking into account the unpredictable influence of the natural environment and usage conditions;

[0051] The local degradation model is established based on nonlinear dynamic equations, as shown in the following formula:

[0052] (10)

[0053] In the formula: This indicates the degree of degradation in a localized area of ​​the bridge, representing the extent of cracks or localized damage. Influencing factors include local loads and local stresses; The nonlinear dynamic equations are locally degenerate; This represents the random perturbation term during the local degradation process;

[0054] The overall degradation model is established based on nonlinear dynamic equations, as shown in the following formula:

[0055] (11),

[0056] In the formula: This represents the overall deterioration state of the bridge structure, indicating the overall health status of the bridge. Factors such as external loads, traffic volume, and ambient temperature and humidity during the overall use process; The nonlinear dynamic equations represent the overall degradation. This refers to the random perturbation term in the overall degradation process;

[0057] The formula for combining the local degradation rate and the overall degradation rate for the weighting coefficients is as follows:

[0058] (12)

[0059] In the formula: , The weighting coefficient controls the proportion of local and global degradation in the overall degradation process;

[0060] The degradation state of the bridge is obtained by numerical integration, and the discretized equations are used for iterative solution. The specific numerical solution method is as follows:

[0061] (13)

[0062] In the formula: Indicates in At time +1, the bridge's degraded state vector at the discrete time step is determined by the current time step. The predicted degradation state is obtained by updating the degradation state through numerical integration of the nonlinear dynamic equation. The time step controls the precision of the calculation; The nonlinear dynamic equations are calculated.

[0063] The numerical integration method is either the Euler method or the Runge-Kutta method.

[0064] Furthermore, the formula for the state space described in step 3 is as follows:

[0065] (15),

[0066] In the formula: S t Representing the state space; Indicates time The stress; Indicates time acceleration (vibration); Indicates time Temperature; Indicates time Humidity; Indicates time Crack data;

[0067] The reward function R t Calculate using the following method:

[0068] (16),

[0069] In the formula: For a moment The accuracy of the prediction; To perform the action The resulting maintenance costs; , These are weighting coefficients used to balance the relationship between prediction accuracy and maintenance costs.

[0070] The multi-objective function includes:

[0071] (19)

[0072] In the formula: The degradation state predicted by the model; This is actual monitoring data;

[0073] (20)

[0074] In the formula: For the execution of the first One maintenance action; The cost of this action;

[0075] (twenty one),

[0076] In the formula: This represents the probability of a bridge structure failing; if the structure is in a safe state, the risk is zero.

[0077] Furthermore, the quantum computing in step 4 includes using quantum annealing to solve the energy evolution problem of the parameter optimization problem of the local degradation model and the global degradation model in step 2; and using a quantum approximation optimization algorithm to perform quantum circuit iterative optimization of the multi-objective function in step 3 to obtain the global optimal solution.

[0078] Furthermore, the real-time monitoring data described in step 5 undergoes noise removal and standardization preprocessing at the edge computing node; the generation of early warning signals includes analyzing the prediction results of the degradation prediction model described in step 3 and the historical changes of the real-time monitoring data, identifying abnormal changes in the degradation process, and generating risk alarms of different levels when the crack propagation rate exceeds the expected range or the stress value exceeds the safety threshold.

[0079] A bridge degradation prediction system for implementing the method includes:

[0080] The data acquisition module connects to strain gauges, accelerometers, temperature and humidity sensors, and crack sensors at key parts of the bridge to collect data on stress, vibration acceleration, temperature, humidity, and cracks.

[0081] The data preprocessing module is connected to the data acquisition module. It receives the acquired real-time monitoring data and unifies it into vector form. It performs noise reduction, data filling, and elimination of different dimensions, and extracts time-domain features, frequency-domain features, and spatial features as feature vectors.

[0082] The degradation modeling module, connected to the data preprocessing module, receives feature vectors as external input variables, establishes local degradation models and global degradation models based on nonlinear dynamic equations, and solves for the degradation state of the bridge.

[0083] The quantum computing acceleration module is connected to the degradation modeling module and the deep reinforcement learning optimization module, respectively, and is used to perform quantum computing processing on the numerical integration solution process of the degradation modeling module and the parameter optimization process of the deep reinforcement learning optimization module.

[0084] The deep reinforcement learning optimization module is connected to the data preprocessing module, the degradation modeling module, and the quantum computing acceleration module. It receives feature vectors and degradation states to construct a state space, uses maintenance decisions as the action space, establishes and optimizes a degradation prediction model based on the degradation state, and generates the optimal maintenance decision.

[0085] The real-time feedback module is connected to the data acquisition module and the deep reinforcement learning optimization module. It inputs the data collected in real time by the acquisition module into the degradation prediction model, combines the degradation state with the optimal maintenance decision to generate an early warning signal, and adaptively adjusts the parameters of the degradation prediction model according to the real-time data.

[0086] The intelligent decision support module is connected to the real-time feedback module, receives early warning signals, and outputs maintenance decision instructions.

[0087] Advantages of the present invention

[0088] 1. This invention, a bridge degradation prediction method based on quantum computing and deep reinforcement learning, significantly improves the computational accuracy and real-time performance of degradation prediction models by combining quantum computing and deep reinforcement learning. The introduction of quantum computing leverages the advantages of quantum superposition and parallel computing to rapidly process multi-dimensional and high-dimensional data, greatly improving the model's computational efficiency and prediction accuracy. The introduction of deep reinforcement learning enables the model to dynamically adjust and optimize based on real-time collected data, thereby ensuring the timeliness and accuracy of prediction results. This overcomes the limitations of traditional methods, which suffer from high computational resource consumption and low prediction accuracy due to data updates and environmental changes.

[0089] 2. This invention adopts a multi-objective optimization mechanism. By optimizing multiple objectives (such as prediction accuracy, maintenance cost, and safety), it can achieve a better balance in terms of the accuracy of degradation prediction, maintenance cost, and bridge service life, ensuring the comprehensiveness and scientific nature of bridge health management. This effectively reduces maintenance costs, improves bridge operation efficiency and safety, and provides comprehensive decision support for bridge management.

[0090] 3. This invention employs deep reinforcement learning for adaptive model learning, adjusting the parameters of the degradation prediction model through real-time feedback to optimize prediction results. It can also adaptively update the prediction strategy based on different bridge types, usage environments, and actual monitoring data, ensuring the accuracy and timeliness of each maintenance decision. This solves the problem that traditional degradation prediction methods rely on static data and fixed models, making it difficult to cope with the impact of dynamic factors such as environmental changes and traffic load fluctuations.

[0091] 4. This invention accelerates the computation process of the degradation prediction model through quantum computing. Especially when dealing with high-dimensional, large-scale data, it utilizes quantum annealing and the quantum approximation optimization algorithm (QAOA) to find the global optimum in complex optimization problems, minimizing the risk of getting trapped in local optima. This advantage is particularly suitable for multi-objective optimization problems, enabling rapid search for optimal solutions in multi-dimensional solution spaces. The acceleration capability of quantum computing significantly reduces computation time, ensuring the system can output prediction results in a shorter time, providing real-time decision support for bridge management.

[0092] 5. This invention effectively combines local and overall bridge degradation through multi-scale nonlinear degradation modeling, forming a comprehensive degradation prediction model. Compared to traditional finite element analysis or empirical models, the degradation modeling method of this invention can more accurately reflect the degradation process caused by the interaction of multiple factors, and can be continuously updated with the addition of new data, providing more accurate degradation prediction results. Through accurate degradation modeling, bridge managers can take timely measures in the early stages of bridge degradation to prevent serious damage and thus extend the service life of the bridge.

[0093] 6. This invention can adapt to more complex environmental changes and diverse usage conditions of bridges. Due to the dynamic adjustment capabilities of deep reinforcement learning and quantum computing, it can automatically adjust the degradation prediction model and make corresponding optimization decisions when facing different environmental changes (such as temperature and humidity changes, load fluctuations, etc.). The method of this invention can provide personalized and accurate degradation prediction and decision support for different bridges and environmental conditions. It solves the problem that existing technologies typically only model bridge health management in fixed environments, failing to cope with complex and dynamic realities.

[0094] 7. This invention, based on quantum computing and deep reinforcement learning, integrates an intelligent decision support system and an early warning mechanism for bridge degradation prediction. Through real-time data monitoring and adaptive learning, it can issue timely warnings based on degradation prediction results and provide real-time decision support for bridge managers. This invention can also generate intelligent decisions based on the bridge's health condition and automatically issue warnings when the degradation trend approaches a critical point, ensuring that managers can take timely measures to reduce bridge safety risks. This solves the problem that traditional bridge management systems often react slowly in emergency situations, leading to untimely handling of safety hazards. Attached Figure Description

[0095] Figure 1 This is a flowchart of the bridge degradation prediction method based on quantum computing and deep reinforcement learning of the present invention.

[0096] Figure 2 This is a schematic diagram illustrating the working principle of the bridge degradation prediction system based on quantum computing and deep reinforcement learning of this invention. Detailed Implementation

[0097] The present invention will be further explained and described below with reference to the accompanying drawings and specific embodiments. It should be noted that the specific embodiments are not intended to limit the scope of the present invention.

[0098] like Figure 1 As shown in the figure, this specific embodiment provides a bridge degradation prediction method based on quantum computing and deep reinforcement learning, including the following steps:

[0099] Step 1, Data Acquisition and Preprocessing: Strain gauges, accelerometers, temperature and humidity sensors, and crack sensors are deployed at key locations on the bridge. The sampling frequency of each sensor is set according to its monitoring object and data variation characteristics. Real-time monitoring data of stress, vibration acceleration, temperature, humidity, and cracks are collected. The collected real-time monitoring data is unified into vector form and subjected to noise reduction, data filling, and elimination of different dimensions. The time domain features, frequency domain features, and spatial features of the processed data are extracted as feature vectors; the details are as follows:

[0100] 1. Data Acquisition: Multiple sensor arrays are deployed at key locations on the bridge to collect real-time health data. These sensor arrays include:

[0101] Strain gauges: Used to monitor the stress and strain state of bridges, collecting real-time monitoring data on the stress borne by various parts of the bridge, such as: Indicates time The stress value is determined. The strain gauge is preferably a resistance strain gauge with a range of ±2000~±10000με, a sensitivity coefficient of 2.0~2.2, and a sampling frequency of not less than 100 Hz, which can meet the accuracy and stability requirements for long-term health monitoring of bridge structures.

[0102] Accelerometer: Used to collect bridge vibration data and analyze the bridge's dynamic response. It collects real-time monitoring data of the bridge's vibration acceleration, such as: Indicates time The acceleration value is obtained. The accelerometer is preferably a triaxial MEMS accelerometer or a piezoelectric accelerometer with a range of ±2 g to ±10 g, a frequency response range of 0.1 Hz to 200 Hz, a sensitivity of not less than 100 mV / g, and a sampling frequency of not less than 200 Hz, to meet the application requirements of bridge structure vibration monitoring and dynamic characteristic analysis.

[0103] Temperature and humidity sensors: Used to monitor the impact of ambient temperature and humidity on bridges. They collect real-time monitoring data on the ambient temperature and humidity around the bridge, for example... Indicates time The ambient temperature value below, Indicates time The ambient relative humidity value is measured; the temperature and humidity sensor is preferably a digital temperature and humidity sensor, with a temperature measurement range of -40 ℃ to 85 ℃ and a measurement accuracy better than ±0.5 ℃, a humidity measurement range of 0% to 100% RH and a measurement accuracy better than ±3% RH, and a sampling period of no more than 1 min, which can meet the application requirements of long-term environmental monitoring and degradation analysis of bridges.

[0104] Crack sensors: Used to monitor the propagation of cracks in bridges, providing data on crack width or length. They collect real-time monitoring data of bridge cracks, such as: Indicates time The crack data includes crack width and / or crack length; the crack sensor is preferably a resistive crack gauge, a displacement crack sensor, or a fiber optic grating (FBG) crack sensor, with a crack measurement range of 0 to 10 mm, a measurement accuracy better than 0.01 mm, and a sampling period of no more than 10 s, which can meet the application requirements for long-term crack propagation monitoring of bridge structures.

[0105] 2. Data Sampling Frequency and Recording Format: The sampling frequency for strain gauges, accelerometers, temperature and humidity sensors, and crack sensors is set according to the monitored object and data variation characteristics. For vibration monitoring and crack propagation monitoring requiring high timeliness, a higher sampling frequency (e.g., 10Hz) can capture rapidly changing signals. For data such as temperature, humidity, and stress, a lower sampling frequency (e.g., 1Hz) is sufficient.

[0106] The collected real-time monitoring data is standardized into vector form for subsequent analysis and processing, as shown in the following formula:

[0107] (1),

[0108] In the formula, Indicates time Data record vector; Indicates time The stress value; Indicates time The vibration acceleration value; Indicates time Temperature value; Indicates time The humidity value; Indicates time Crack data.

[0109] 3. Noise reduction processing:

[0110] The raw data from the collected real-time monitoring data undergoes noise reduction processing to ensure data accuracy and reliability. Common noise reduction methods include:

[0111] (1) Kalman Filtering: Used to remove noise from sensor signals, especially suitable for processing time series data. The Kalman filter makes predictions and corrections based on the system's dynamic model and measurement data, as shown in the following formula:

[0112] (2),

[0113] In the formula: Indicates time The estimated state; This represents the Kalman gain, which controls the tradeoff between predicted and actual observations. Indicates time The actual measured value; This represents the measurement matrix, which shows the relationship between measurements and states.

[0114] (2) Wavelet Transform: Suitable for removing high-frequency noise while preserving the low-frequency characteristics of the signal. This method is often used in signal processing for denoising vibration and stress data.

[0115] 4. Data Filling:

[0116] For data lost due to sensor malfunction or other reasons, the following methods can be used to fill it in:

[0117] (1) Linear interpolation: Used to handle simple missing data cases, filling in missing points through linear derivation. The formula is as follows:

[0118] (3),

[0119] In the formula: Indicates time The interpolated value; and The time for the known data points.

[0120] (2) Spline Interpolation: It is used to fill in complex situations where data is missing. It can smooth the data better and ensure a smooth transition of the interpolated data.

[0121] 5. Data standardization and normalization:

[0122] All collected data signals are standardized or normalized for subsequent modeling and processing. Standardization and normalization can eliminate the influence of different units of measurement on model training.

[0123] (1) Standardization: Convert the data into a standard normal distribution with a mean of 0 and a standard deviation of 1. The formula is as follows:

[0124] (4),

[0125] In the formula: This is the original data; The mean of the data; denoted as the standard deviation of the data.

[0126] (2) Normalization: Mapping the range of data values ​​to the interval [0, 1], as shown in the following formula:

[0127] (5),

[0128] In the formula: This is the original data; and These are the minimum and maximum values ​​of the data.

[0129] 6. Data Feature Extraction

[0130] Feature extraction is performed on the processed data to facilitate subsequent modeling and analysis. Common feature extraction methods include:

[0131] (1) Time-domain feature extraction: such as mean, maximum value, variance, skewness, kurtosis, etc., to help analyze the distribution characteristics of the data. For example:

[0132] (6),

[0133] (7),

[0134] In the formula: The mean of the data; For the first Data points.

[0135] (2) Frequency domain feature extraction: The frequency features of the vibration signal are extracted using Fourier transform or wavelet transform. The spectrum of the vibration signal is obtained by fast Fourier transform (FFT) to help analyze the dynamic response of the bridge.

[0136] (3) Spatial feature extraction: By analyzing the crack propagation pattern through crack sensor data, features such as crack length or width are extracted and used as input features for modeling.

[0137] Step 2, Multi-scale Nonlinear Degradation Modeling: Using the feature vectors described in Step 1 as external input variables, establish local degradation models and global degradation models based on nonlinear dynamic equations. Solve for the local degradation rate and global degradation rate of the bridge respectively. Fuse the local degradation rate and global degradation rate through weighting coefficients to obtain the total degradation rate, and use numerical integration methods to solve for the degradation state of the bridge, thereby establishing a degradation prediction model; the specific content is as follows:

[0138] 1. Nonlinear dynamics modeling:

[0139] During the degradation process of bridges, multiple factors such as external loads, environmental temperature and humidity, and material fatigue combine to create a highly nonlinear degradation process. To accurately model this degradation process, it is assumed that the bridge's degradation state changes over time. Changes occur, and the degradation process is influenced by a complex interplay of internal and external factors.

[0140] (1) Nonlinear dynamic equations

[0141] The modeling of bridge degradation processes is based on nonlinear dynamic equations, assuming a degradation state. Over time Changes, and are affected by external environmental factors such as load, temperature, and humidity. And the effects of inherent material degradation factors. The nonlinear dynamic equations are as follows:

[0142] (8),

[0143] In the formula: Indicates the bridge degradation state at time [time]. The rate of change (i.e., the degradation rate) represents the rate at which the bridge degrades at this moment; Let be the degraded state vector of the bridge, representing the health state of the bridge; The nonlinear dynamic equations describing the degradation process of bridges reflect the relationship between the degradation state and external factors such as load, temperature and humidity. External input variables represent environmental factors that affect bridge degradation, such as load, temperature and humidity, and traffic flow. The random perturbation term of the system represents the uncertainty in the degradation process, taking into account unpredictable influencing factors such as the natural environment and usage conditions.

[0144] This nonlinear equation describes the dynamic behavior of the bridge during the degradation process, and its solution involves solving ordinary differential equations (ODEs) in nonlinear dynamics.

[0145] (2) Nonlinear function form

[0146] It is the main nonlinear function that determines the evolution of the bridge's degradation state during the degradation process, and is usually expressed in polynomial or power-law form:

[0147] (9),

[0148] In the formula: , , , The regression coefficients represent the system's sensitivity. , It is a constant that controls the degree of nonlinearity in the degradation process.

[0149] This function can fully capture the coupling effect of load-stress and environmental factors during bridge degradation.

[0150] 2. Multi-scale degradation modeling

[0151] (1) Local degradation model

[0152] Local degradation models are used to describe in detail the local damage to bridge structures, such as crack propagation and fatigue damage. Considering that the degradation of local areas of a bridge is affected by factors such as local loads and stress concentration, the formula for a local degradation model based on nonlinear dynamic equations is as follows:

[0153] (10)

[0154] In the formula: This indicates the degree of degradation in a localized area of ​​the bridge, representing the extent of cracks or localized damage. Influencing factors include local loads and local stresses; The nonlinear dynamic equations are locally degenerate; This represents the random perturbation term during the local degradation process.

[0155] (2) Overall degradation model

[0156] The overall degradation model describes factors such as material aging and fatigue failure throughout the entire lifespan of a bridge. This model considers the long-term use of the bridge and incorporates the gradual aging of materials. The formula for the overall degradation model based on nonlinear dynamic equations is as follows:

[0157] (11),

[0158] In the formula: This represents the overall deterioration state of the bridge structure, indicating the overall health status of the bridge. Factors such as external loads, traffic volume, and ambient temperature and humidity during the overall use process; The nonlinear dynamic equations represent the overall degradation. This represents the random perturbation term in the overall degradation process.

[0159] (3) Joint modeling

[0160] Multi-scale degradation modeling is achieved by combining local and global degradation models. The degradation prediction model is established through joint modeling using the following equations:

[0161] (12)

[0162] In the formula: , The weighting coefficient controls the proportion of local and global degradation in the overall degradation process.

[0163] The degradation prediction model obtained through this joint modeling approach can simultaneously consider local damage (such as crack propagation) and overall aging of the bridge, ensuring the comprehensiveness and accuracy of the model.

[0164] 3. Solving and Validating the Degradation Prediction Model

[0165] (1) Numerical solution method

[0166] To solve the nonlinear dynamic equations, numerical integration methods, such as the Euler method or the Runge-Kutta method, are used to solve for the degradation state at each time step to obtain the degradation state of the bridge. Discretized equations are then used for iterative solutions. The specific numerical solution method is as follows:

[0167] (13)

[0168] In the formula: Indicates in At time +1, the bridge's degraded state vector at the discrete time step is determined by the current time step. The predicted degradation state is obtained by updating the degradation state through numerical integration of the nonlinear dynamic equation. The time step controls the precision of the calculation; The nonlinear dynamic equations are calculated.

[0169] (2) Validation of degradation prediction model

[0170] The degradation prediction model was validated by comparing it with actual monitoring data, using evaluation metrics such as mean squared error (MSE) to verify the prediction accuracy of the degradation prediction model.

[0171] (14)

[0172] In the formula: The degradation state predicted by the model; This is actual monitoring data; This represents the number of data points.

[0173] Model validation ensures that the degradation prediction model accurately reflects the degradation trend of bridges.

[0174] The purpose of multi-scale nonlinear degradation modeling is to accurately simulate various influencing factors in the bridge degradation process through multi-dimensional nonlinear dynamic equations. By establishing local and overall degradation models of the bridge, this step can simultaneously capture the effects of short-term loads and long-term material aging and environmental changes, ensuring the accuracy and stability of the degradation prediction model.

[0175] Step 3, Deep Reinforcement Learning and Multi-Objective Optimization: Based on the feature vectors described in Step 1 and the degradation state of the bridge described in Step 2, a state space is constructed. Maintenance decisions are used as the action space, and the balance between prediction accuracy and maintenance cost is used as the reward function. A degradation prediction model based on the degradation state of the bridge described in Step 2 is established. This model is then optimized using deep reinforcement learning. A multi-objective function is constructed, considering prediction accuracy, maintenance cost, and structural safety. Multi-objective optimization techniques are employed to optimize this function and generate the optimal maintenance decision. The specific details are as follows:

[0176] 1. Deep reinforcement learning optimizes degradation prediction models:

[0177] Reinforcement learning is a learning method based on behavioral feedback that self-optimizes through interaction with the environment. In this embodiment, reinforcement learning is used to optimize a bridge degradation prediction model to achieve an adaptive degradation prediction process. The learning objective of the degradation prediction model is to maximize cumulative rewards, improve prediction accuracy, and reduce maintenance costs by continuously adjusting the strategy.

[0178] The basic process of reinforcement learning includes:

[0179] State space S: Represents the current health status of the bridge and the influencing factors of the external environment, such as stress, displacement, temperature and humidity data;

[0180] Action space A: Represents the maintenance decisions that can be made, such as whether to perform repairs, replace materials, or adjust the maintenance cycle;

[0181] Reward function R: A criterion used to evaluate the predictive performance of the model. The reward function should be evaluated based on factors such as the accuracy of the model's predictions, maintenance costs, and the bridge's lifespan.

[0182] Assuming the current time The bridge status is Based on real-time monitoring data and degradation prediction results, the learning process of the degradation prediction model is as follows:

[0183] (1) State space :time The bridge's health status, combined with real-time monitoring data from all strain gauges, accelerometers, temperature and humidity sensors, and crack sensors (such as stress, displacement, cracks, etc.), is represented by the following state vector:

[0184] (15)

[0185] In the formula: Indicates time The stress; Indicates time acceleration (vibration); Indicates time Temperature; Indicates time Humidity; Indicates time Crack data.

[0186] (2) Action space : Executable operations, such as: (Perform maintenance); (Extended warranty period); (Replace materials).

[0187] (3) Reward function The reward is calculated based on the effect of each action. The reward function should balance prediction accuracy and maintenance cost, and its specific form is as follows:

[0188] (16)

[0189] In the formula: For a moment The accuracy of the prediction; To perform the action The resulting maintenance costs; , These are weighting coefficients used to balance the relationship between prediction accuracy and maintenance costs.

[0190] Through continuous interaction with the environment, reinforcement learning algorithms can constantly adjust their prediction strategies and decisions to maximize cumulative rewards.

[0191] 2. Reinforcement Learning Algorithm Selection

[0192] This embodiment uses depth The Deep Q-Network (DQN) algorithm is used to optimize the degradation prediction model. DQN uses deep neural networks to approximate... Value function Furthermore, the stability of the algorithm is improved through experience replay and target network mechanisms. The value update formula is as follows:

[0193] (17)

[0194] In the formula: Indicates time state Take action below of value; Indicates time The reward; This represents the discount factor, which determines the impact on future rewards; Represents the learning rate, controlling The step size for value updates; Indicates time +1 All possible actions Corresponding maximum value.

[0195] Through repeated learning and optimization, the DQN algorithm can automatically adjust the model and provide more accurate degradation prediction results.

[0196] 3. Multi-objective optimization mechanism:

[0197] In bridge degradation prediction and maintenance decision-making, in addition to accurate degradation prediction, it is necessary to balance multiple optimization objectives, such as prediction accuracy, maintenance cost, bridge service life, and structural safety. Therefore, a multi-objective optimization method is adopted to optimize multiple objective functions in order to achieve a globally optimal solution.

[0198] (1) Multi-objective optimization objective function

[0199] Define multiple objective functions to describe the optimization objective:

[0200] (18)

[0201] In the formula: This represents the accuracy of the degradation prediction, i.e., the minimization of the prediction error; This refers to maintenance cost, specifically minimizing the cost when optimizing maintenance strategies. This refers to structural safety, which means ensuring the structural safety of a bridge through optimized maintenance plans.

[0202] (19)

[0203] In the formula: The degradation state predicted by the model; This is actual monitoring data.

[0204] (20)

[0205] In the formula: For the execution of the first One maintenance action; The cost of this action.

[0206] (twenty one),

[0207] In the formula: This represents the probability of a bridge structure failing; if the structure is in a safe state, the risk is zero.

[0208] (2) Weighted solution of the objective function

[0209] To balance multiple optimization objectives, a weighted sum method is used to weight the objective function, and the weight coefficients are adjusted according to actual needs. The optimization objective is:

[0210] (twenty two),

[0211] In the formula: , , ..., The weight coefficients for each objective function represent the priority of each objective.

[0212] By adjusting the weighting coefficients, the relationship between forecast accuracy, cost, and structural safety can be balanced according to the needs of bridge managers.

[0213] (3) Selection of multi-objective optimization algorithm

[0214] In this embodiment, Particle Swarm Optimization (PSO) or Genetic Algorithm (GA) is used for multi-objective optimization. Taking the Genetic Algorithm as an example, its basic operations include selection, crossover, and mutation, which can effectively solve multi-objective optimization problems.

[0215] The purpose of this step is to optimize the bridge degradation prediction model using deep reinforcement learning (DRL) and combine it with multi-objective optimization techniques to balance multiple optimization objectives. Deep reinforcement learning can automatically adjust the prediction model based on real-time monitoring data and environmental changes, improving the accuracy of degradation predictions and the level of intelligent decision-making. Multi-objective optimization ensures that the model balances multiple objectives, making bridge maintenance decisions more accurate, economical, and safe.

[0216] Step 4, Quantum Computing and Model Acceleration: Quantum computing is used to solve the equations of the numerical integration method described in Step 2, as well as to optimize the parameters and multi-objective functions of the deep reinforcement learning described in Step 3, utilizing quantum parallelism to process high-dimensional data; the specific content is as follows:

[0217] The introduction of quantum computing aims to accelerate the computation of degradation prediction models, especially when dealing with large-scale, high-dimensional data, where quantum computing offers tremendous computational speedup. Through quantum computing, higher efficiency and more accurate results can be achieved in complex nonlinear dynamics modeling, degradation prediction model optimization, and multi-objective optimization computation. The application of quantum computing can significantly reduce the time required by traditional computational methods, thereby improving the real-time performance and accuracy of the system.

[0218] 1. Quantum computing accelerates degradation prediction model

[0219] Quantum computing utilizes qubits (qubits) and phenomena such as quantum superposition, quantum entanglement, and quantum interference to give it an advantage over classical computers in solving certain problems. Specifically, quantum computing can handle multiple computational paths simultaneously, significantly accelerating the optimization process, and is particularly suitable for degradation prediction models with large parameter spaces and high complexity.

[0220] 2. Optimization of the quantum degradation prediction model

[0221] In degradation prediction models, the main application of quantum computing is to accelerate the optimization process through quantum optimization algorithms (such as quantum annealing and quantum approximation optimization algorithms). The quantum approximation optimization algorithm (QAOA) is used to perform quantum circuit iterative optimization on the multi-objective function described in step 3 to obtain the global optimum. Traditional optimization algorithms, such as particle swarm optimization (PSO) or genetic algorithms (GA), have long computation times when dealing with high-dimensional data, while quantum computing can explore multiple solution spaces in parallel through quantum superposition and quantum parallelism, thus finding the optimum much faster.

[0222] Quantum optimization method: Quantum annealing is a special quantum computing method suitable for optimization problems, especially combinatorial optimization problems. In degradation prediction models, quantum annealing is used to solve parameter optimization problems. Quantum annealing is employed to solve the parameter optimization problems of the local degradation model and the global degradation model described in step 2 using energy evolution. The quantum annealing algorithm gradually leads the quantum system to the minimum energy state through the energy evolution of the quantum system, ultimately finding the optimal solution.

[0223] Suppose an optimization objective function The optimization process of quantum annealing is expressed as:

[0224] (twenty three),

[0225] In the formula: Let be the Hamiltonian of the system, representing the energy of the quantum system; The initial Hamiltonian is typically set to an energy expression relevant to the optimization problem. Let be the objective Hamiltonian, and let represent the optimized objective function. The time parameter ranges from 0 to 1, representing the gradual evolution of the system from the initial state to the target state.

[0226] The quantum annealing algorithm utilizes the properties of quantum mechanics, in When the system's energy is minimized, the global optimal solution to the optimization problem is found.

[0227] 3. The Quantum Approximate Optimization Algorithm (QAOA) is another quantum optimization method, particularly suitable for combinatorial optimization problems. In this embodiment, QAOA is used for high-dimensional parameter optimization of a degenerate prediction model. QAOA designs quantum circuits through quantum gate operations and updates parameters incrementally to find the optimal solution to the objective function.

[0228] The optimization process of QAOA is achieved through the following steps:

[0229] (1) Initialize the state of the qubit .

[0230] (2) A series of quantum gate operations are applied in the quantum circuit to iteratively adjust the state of the qubit to make it close to the optimal solution.

[0231] (3) Obtain the final optimization results through measurement operations.

[0232] The goal of quantum circuits is to minimize the objective function. The value of is given by the following formula:

[0233] (twenty four),

[0234] In the formula: To optimize the Hamiltonian of the problem; It is a quantum state.

[0235] QAOA optimization can achieve an efficient balance among multiple optimization objectives and quickly converge to the optimal solution.

[0236] 4. Applications of quantum computing in high-dimensional data processing

[0237] (1) High-dimensional data processing

[0238] In degradation prediction models, data often has high dimensionality (such as data collected from multiple sensors). Traditional computational methods are slow and prone to getting trapped in local optima when processing large-scale data. Quantum computing, through its quantum parallelism, can explore a larger solution space in parallel, thereby improving data processing efficiency and avoiding getting trapped in local optima.

[0239] For example, when modeling degradation processes, quantum computing can use quantum algorithms to simultaneously process parameter combinations of multiple degradation models, greatly accelerating the parameter search process in multi-scale degradation modeling.

[0240] (2) Combination of quantum computing and deep learning

[0241] In degradation prediction models, quantum computing can be combined with deep learning (such as deep neural networks) to accelerate the model training process. Quantum computing can optimize the weight updates and gradient descent processes in deep learning models, thereby improving the convergence speed and accuracy of deep learning models.

[0242] When optimizing deep learning models, the introduction of quantum computing can accelerate gradient calculation in the backpropagation algorithm, making model training more efficient, especially when dealing with large-scale data, and can provide significant performance improvement.

[0243] 5. Quantum computing accelerates multi-objective optimization

[0244] (1) Multi-objective optimization problem

[0245] In bridge degradation prediction and maintenance decision-making, there are multiple optimization objectives, such as prediction accuracy, maintenance cost, and service life. Quantum computing can effectively handle multiple objective functions simultaneously and find Pareto optimal solutions. Quantum optimization algorithms can efficiently explore the solution space of multiple objective functions and achieve optimal trade-offs among the objectives.

[0246] (2) Quantum multi-objective optimization algorithm

[0247] Quantum computing, through its parallel computing capabilities, can process multiple objective functions simultaneously in an optimization process. For example, quantum computing can simultaneously evaluate the solution space of multiple objective functions through quantum parallelism, thereby accelerating the multi-objective optimization process. The core idea of ​​quantum multi-objective optimization algorithms is to use quantum algorithms to process each objective function in parallel, progressively optimizing the trade-offs between multiple objectives.

[0248] Step 5, Real-time Feedback and Intelligent Decision Support: Based on the real-time monitoring data described in Step 1, the real-time monitoring data undergoes noise removal and standardization preprocessing at the edge computing node; it is then input into the degradation prediction model described in Step 3. Combining the degradation state described in Step 2 and the optimal maintenance decision described in Step 3, an early warning signal is generated. Generating the early warning signal involves analyzing the prediction results of the degradation prediction model described in Step 3 and the historical changes of the real-time monitoring data, identifying abnormal changes in the degradation process, and generating different levels of risk alarms when the crack propagation rate exceeds the expected range or the stress value exceeds the safety threshold. Furthermore, based on the real-time monitoring data described in Step 1, the parameters of the degradation prediction model described in Step 3 are adaptively adjusted through deep reinforcement learning as described in Step 3. The specific details are as follows:

[0249] 1. Real-time feedback mechanism

[0250] (1) Real-time data acquisition and transmission: In this embodiment, the bridge health monitoring system acquires real-time monitoring data of multiple modes, including stress, strain, vibration, temperature, and humidity. The real-time monitoring data is uploaded to a cloud computing platform or edge computing node in real time via Internet of Things (IoT) devices to ensure the timeliness and accuracy of the data. The specific steps are as follows:

[0251] Data acquisition frequency: Set an appropriate sampling frequency according to different sensor types.

[0252] Among them, the sampling frequency of the stress and strain sensors is set to 1 Hz to 10 Hz, which is used to collect the response data of the bridge structure under quasi-static or low-frequency loads.

[0253] The sampling frequency of the vibration acceleration sensor is set to no less than 50 Hz to meet the real-time monitoring requirements for the dynamic response and vibration characteristic analysis of the bridge structure.

[0254] The temperature and humidity sensor is set to sample once every 30 minutes to 1 hour to collect information on changes in the bridge's service environment.

[0255] Real-time data transmission: The collected real-time monitoring data is uploaded to the central processing platform in real time via wireless sensor networks (WSN) or Internet of Things communication protocols (such as MQTT or LoRa) to ensure that the data can be transmitted and stored quickly.

[0256] Real-time monitoring allows for the acquisition of the latest bridge health status data at any time, providing timely updated input for degradation prediction models.

[0257] 2. Real-time data processing and updating

[0258] As new real-time monitoring data is collected, the degradation prediction model needs to be continuously updated to maintain prediction accuracy. By introducing edge computing, preliminary data processing and preprocessing (such as noise reduction, filtering, and standardization) can be performed near the data source (e.g., sensor nodes, routers), reducing data transmission bandwidth requirements and accelerating real-time feedback.

[0259] The specific process is as follows:

[0260] Data preprocessing: Preliminary processing such as noise removal, data interpolation, and standardization is performed at edge nodes to reduce system load.

[0261] Real-time feedback: The preprocessed data is directly fed into the degradation prediction model for real-time analysis and prediction. The model parameters are updated through deep reinforcement learning or traditional regression analysis to generate new prediction results.

[0262] The real-time feedback mechanism ensures that the system can adjust the prediction model in real time based on the current data, providing accurate bridge degradation trends.

[0263] 3. Intelligent Decision Support and Early Warning System

[0264] (1) The Intelligent Decision Support System (DSS) combines prediction results and multi-objective optimization mechanisms to provide bridge management personnel with scientific maintenance decisions and emergency response plans. It primarily provides decision support through the following methods:

[0265] Maintenance strategy optimization: Based on degradation prediction results, the optimal maintenance plan is automatically generated. Factors considered in the maintenance plan include the bridge's current condition, maintenance cost, prediction accuracy, traffic flow, and environmental conditions. These objectives are balanced using multi-objective optimization algorithms (such as genetic algorithms or particle swarm optimization) to provide the optimal decision.

[0266] Optimization objectives include:

[0267] Maximize prediction accuracy: Ensure the accuracy of the system's prediction results.

[0268] Minimize maintenance costs: Reduce repair and maintenance expenses as much as possible while ensuring safety.

[0269] Extending bridge lifespan: Extending the lifespan of bridges through optimized maintenance cycles and solutions.

[0270] Decision Trees and Rule Engines: The intelligent decision support system combines decision tree algorithms and rule engines to automatically generate appropriate decisions based on the actual situation. For example, when the predictive model shows that the bridge is degrading too quickly, the system may automatically trigger emergency repairs or adjust the maintenance cycle; if the degradation rate is slow, it may recommend extending the inspection cycle or reducing the maintenance frequency.

[0271] (2) The early warning system combines real-time monitoring data and degradation prediction results to automatically generate early warning signals and promptly issue risk alerts to management personnel. The early warning system is based on the following two key factors:

[0272] Anomaly detection: By analyzing degradation prediction models and historical monitoring data, abnormal changes in the degradation process are identified. For example, an alarm will be issued if the crack propagation rate exceeds the expected range or the stress value exceeds the safety threshold.

[0273] Prediction and Risk Assessment: Combining degradation prediction models and risk assessment algorithms, the potential degradation of the bridge over a future period is evaluated. If the degradation rate is too rapid, the system will automatically calculate the risk of potential structural damage and issue different levels of warnings based on the risk level.

[0274] Early warning systems can issue warnings before problems occur, helping managers take preventative maintenance measures.

[0275] 4. Visual interface

[0276] To improve decision-making efficiency, the intelligent decision support system provides a visual interface, allowing bridge managers to intuitively view the following information:

[0277] Degradation trend chart: Shows the historical degradation trend of bridges and future forecasts;

[0278] Risk Assessment Report: Structural safety risk assessment based on model output;

[0279] Maintenance recommendations: Repair schedules and solutions provided based on optimization results;

[0280] Warning information: Indicates whether there are any structural problems or anomalies that are about to occur.

[0281] The visual interface design enables managers to monitor the bridge's health status in real time and make quick decisions.

[0282] 5. Adaptive learning and optimization

[0283] (1) Adaptive learning: The adaptive learning mechanism continuously optimizes the degradation prediction model based on real-time monitoring data and feedback information. Whenever new data is input, the system updates the formula according to the Q-value in deep reinforcement learning, automatically adjusts the model parameters, and optimizes the model prediction effect;

[0284] Incremental learning methods are used to continuously accumulate new data and improve prediction accuracy, thereby enhancing the system's long-term adaptability. Adaptive learning ensures that the system maintains high prediction accuracy as data increases, the environment changes, and bridge degradation patterns evolve.

[0285] 6. Model optimization and accuracy improvement

[0286] By continuously adjusting the prediction model through feedback mechanisms and adaptive learning, the accuracy and stability of predictions are improved. When encountering different bridge types, usage environments, or degradation patterns, the learning algorithm can be automatically adjusted based on historical data, resulting in more accurate predictions. For example, bridge degradation models in harsh environments may exhibit different characteristics than those in normal environments; the learning parameters are adaptively adjusted based on environmental changes.

[0287] This step aims to dynamically adjust the degradation prediction model based on the actual health condition of the bridge through real-time data feedback and an intelligent decision support system, providing intelligent decision support for bridge managers. By integrating edge computing and adaptive learning, maintenance strategies can be continuously optimized based on real-time monitoring data, improving the efficiency and accuracy of bridge health management.

[0288] A bridge degradation prediction system implementing the above method includes:

[0289] The data acquisition module is connected to strain gauges, accelerometers, temperature and humidity sensors, and crack sensors in key parts of the bridge. The sampling frequency of each sensor is set according to the monitoring object and data change characteristics to collect stress, vibration acceleration, temperature, humidity, and crack data.

[0290] The data preprocessing module is connected to the data acquisition module. It receives the acquired real-time monitoring data and unifies it into vector form. It performs noise reduction, data filling, and elimination of different dimensions, and extracts time-domain features, frequency-domain features, and spatial features as feature vectors.

[0291] The degradation modeling module, connected to the data preprocessing module, receives feature vectors as external input variables, establishes local degradation models and global degradation models based on nonlinear dynamic equations, and solves for the degradation state of the bridge.

[0292] The quantum computing acceleration module is connected to the degradation modeling module and the deep reinforcement learning optimization module, respectively, and is used to perform quantum computing processing on the numerical integration solution process of the degradation modeling module and the parameter optimization process of the deep reinforcement learning optimization module.

[0293] The deep reinforcement learning optimization module is connected to the data preprocessing module, the degradation modeling module, and the quantum computing acceleration module. It receives feature vectors and degradation states to construct a state space, uses maintenance decisions as the action space, establishes and optimizes a degradation prediction model based on the degradation state, and generates the optimal maintenance decision.

[0294] The real-time feedback module is connected to the data acquisition module and the deep reinforcement learning optimization module. It inputs the data collected in real time by the acquisition module into the degradation prediction model, combines the degradation state with the optimal maintenance decision to generate an early warning signal, and adaptively adjusts the parameters of the degradation prediction model according to the real-time data.

[0295] The intelligent decision support module is connected to the real-time feedback module, receives early warning signals, and outputs maintenance decision instructions.

Claims

1. A bridge degradation prediction method based on quantum computing and deep reinforcement learning, characterized in that, Includes the following steps: Step 1, Data Acquisition and Preprocessing: Strain gauges, accelerometers, temperature and humidity sensors and crack sensors are installed at key parts of the bridge to collect real-time monitoring data on stress, vibration acceleration, temperature, humidity and cracks. The collected real-time monitoring data is unified into vector form and then subjected to noise reduction, data filling and elimination of different dimensions. The time domain features, frequency domain features and spatial features of the processed data are extracted as feature vectors. Step 2, Multi-scale nonlinear degradation modeling: Using the feature vectors described in Step 1 as external input variables, establish a local degradation model and a global degradation model based on nonlinear dynamic equations, solve for the local degradation rate and the global degradation rate of the bridge respectively, fuse the local degradation rate and the global degradation rate through weight coefficients to obtain the total degradation rate, and use the numerical integration method to solve for the degradation state of the bridge. Step 3, Deep Reinforcement Learning and Multi-Objective Optimization: Based on the feature vectors described in Step 1 and the degradation state of the bridge described in Step 2, a state space is constructed. The maintenance decision is used as the action space, and the balance between prediction accuracy and maintenance cost is used as the reward function. A degradation prediction model based on the degradation state of the bridge described in Step 2 is established. The degradation prediction model is optimized through deep reinforcement learning. A multi-objective function is constructed with prediction accuracy, maintenance cost, and structural safety. Multi-objective optimization techniques are used to optimize the multi-objective function to generate the optimal maintenance decision. Step 4, Quantum Computing and Model Acceleration: Quantum computing is used to solve the equations of the numerical integration method described in Step 2 and to optimize the parameters and multi-objective functions of the deep reinforcement learning described in Step 3, utilizing quantum parallelism to process high-dimensional data; Step 5, Real-time Feedback and Intelligent Decision Support: Based on the real-time monitoring data described in Step 1, input it into the degradation prediction model described in Step 3. Combine the degradation state described in Step 2 and the optimal maintenance decision described in Step 3 to generate an early warning signal. Based on the real-time monitoring data described in Step 1, adaptively adjust the parameters of the degradation prediction model described in Step 3 through deep reinforcement learning described in Step 3.

2. The method according to claim 1, characterized in that, The formula for unifying the collected data into vector form as described in step 1 is as follows: （1）， In the formula, Indicates time Data record vector; Indicates time The stress value; Indicates time The vibration acceleration value; Indicates time Temperature value; Indicates time The humidity value; Indicates time Crack data; The denoising process includes Kalman filtering for removing noise from the sensor signal and wavelet transform for removing high-frequency noise. The formula for Kalman filtering is as follows: （2）， In the formula: Indicates time The estimated state; This represents the estimated state at time t-1; This represents the Kalman gain, which controls the tradeoff between predicted and actual observations. Indicates time The actual measured value; The measurement matrix represents the relationship between measurements and states. The data imputation includes linear interpolation for handling simple missing data cases and spline interpolation for filling more complex missing data cases. The formula for linear interpolation is as follows: （3）， In the formula: Indicates time The interpolated value; and The time for the known data points; The process of eliminating different dimensions employs standardization or normalization. Standardization is used to convert the data into a standard normal distribution with a mean of 0 and a standard deviation of 1, as shown in the following formula: （4）， In the formula: This is the original data; The mean of the data; The standard deviation of the data; Normalization is used to map the range of data values ​​to the interval [0, 1]. The formula is as follows: （5）， In the formula: This is the original data; and These are the minimum and maximum values ​​of the data.

3. The method according to claim 1, characterized in that, The nonlinear dynamic equations described in step 2 are as follows: （8）， In the formula: Indicates the bridge degradation state at time [time]. The rate of change of , i.e. the degradation rate, represents the rate at which the bridge degrades at this moment; Let be the degraded state vector of the bridge, representing the health state of the bridge; The nonlinear dynamic equations describing the bridge degradation process reflect the relationship between the degradation state and external factors such as load and temperature and humidity. External input variables represent environmental factors that affect bridge degradation, such as load, temperature and humidity, and traffic flow. The random perturbation term of the system represents the uncertainty in the degradation process, taking into account the unpredictable influence of the natural environment and usage conditions; The local degradation model is established based on nonlinear dynamic equations, as shown in the following formula: （10）， In the formula: This indicates the degree of degradation in a localized area of ​​the bridge, representing the extent of cracks or localized damage. Influencing factors include local loads and local stresses; The nonlinear dynamic equations are locally degenerate; This represents the random perturbation term during the local degradation process; The overall degradation model is established based on nonlinear dynamic equations, as shown in the following formula: （11）， In the formula: This represents the overall deterioration state of the bridge structure, indicating the overall health status of the bridge. Factors such as external loads, traffic volume, and ambient temperature and humidity during the overall use process; The nonlinear dynamic equations represent the overall degradation. This refers to the random perturbation term in the overall degradation process; The formula for combining the local degradation rate and the overall degradation rate for the weighting coefficients is as follows: （12）， In the formula: , The weighting coefficient controls the proportion of local and global degradation in the overall degradation process; The degradation state of the bridge is obtained by numerical integration, and the discretized equations are used for iterative solution. （13）， In the formula: Indicates in At time +1, the bridge's degraded state vector at the discrete time step is determined by the current time step. The predicted degradation state is obtained by updating the degradation state through numerical integration of the nonlinear dynamic equation. The time step controls the precision of the calculation; The nonlinear dynamic equations are calculated.

4. The numerical integration method is either the Euler method or the Runge-Kutta method.

5. The method according to claim 1, characterized in that, The formula for the state space described in step 3 is as follows: （15） In the formula: S t Representing the state space; Indicates time The stress; Indicates time acceleration (vibration); Indicates time Temperature; Indicates time humidity; Indicates time Crack data; The reward function R t Calculate using the following method: （16） In the formula: For a moment The accuracy of the prediction; To perform the action The resulting maintenance costs; , These are weighting coefficients used to balance the relationship between prediction accuracy and maintenance costs. The multi-objective function includes: （19）， In the formula: The degradation state predicted by the model; This is actual monitoring data; （20）， In the formula: For the execution of the first One maintenance action; The cost of this action; （21）， In the formula: This represents the probability of a bridge structure failing; if the structure is in a safe state, the risk is zero.

6. The method according to claim 1, characterized in that, The quantum computing described in step 4 includes using quantum annealing to solve the energy evolution problem of the parameter optimization problem of the local degradation model and the global degradation model described in step 2; and using a quantum approximation optimization algorithm to perform quantum circuit iterative optimization of the multi-objective function described in step 3 to obtain the global optimal solution.

7. The method according to claim 1, characterized in that, The real-time monitoring data described in step 5 undergoes noise removal and standardization preprocessing at the edge computing node; the generation of early warning signals includes analyzing the prediction results of the degradation prediction model described in step 3 and the historical changes of the real-time monitoring data, identifying abnormal changes in the degradation process, and generating risk alarms of different levels when the crack propagation rate exceeds the expected range or the stress value exceeds the safety threshold.

8. A bridge degradation prediction system implementing the method of any one of claims 1 to 6, characterized in that, include: The data acquisition module connects to strain gauges, accelerometers, temperature and humidity sensors, and crack sensors at key parts of the bridge to collect data on stress, vibration acceleration, temperature, humidity, and cracks. The data preprocessing module is connected to the data acquisition module. It receives the acquired real-time monitoring data and unifies it into vector form. It performs noise reduction, data filling, and elimination of different dimensions, and extracts time-domain features, frequency-domain features, and spatial features as feature vectors. The degradation modeling module, connected to the data preprocessing module, receives feature vectors as external input variables, establishes local degradation models and global degradation models based on nonlinear dynamic equations, and solves for the degradation state of the bridge. The quantum computing acceleration module is connected to the degradation modeling module and the deep reinforcement learning optimization module, respectively, and is used to perform quantum computing processing on the numerical integration solution process of the degradation modeling module and the parameter optimization process of the deep reinforcement learning optimization module. The deep reinforcement learning optimization module is connected to the data preprocessing module, the degradation modeling module, and the quantum computing acceleration module. It receives feature vectors and degradation states to construct a state space, uses maintenance decisions as the action space, establishes and optimizes a degradation prediction model based on the degradation state, and generates the optimal maintenance decision. The real-time feedback module is connected to the data acquisition module and the deep reinforcement learning optimization module. It inputs the data collected in real time by the acquisition module into the degradation prediction model, combines the degradation state with the optimal maintenance decision to generate an early warning signal, and adaptively adjusts the parameters of the degradation prediction model according to the real-time data. The intelligent decision support module is connected to the real-time feedback module, receives early warning signals, and outputs maintenance decision instructions.

Images