Method and system for predicting service life of degradable drain board

Abstract

The invention relates to the technical field of drain board monitoring, and discloses a degradable drain board service life prediction method and system, and the method comprises the steps: collecting multi-dimensional environment original data, and carrying out the weighted fusion, and obtaining an integrated environment data set; performing feature extraction and spatio-temporal evolution analysis on the integrated environment data set to obtain a spatio-temporal feature evolution path; performing multi-factor correlation calculation on the spatio-temporal characteristic evolution path to obtain an interaction weight parameter; performing model training in combination with the weight parameters to obtain an initial attenuation model; performing regional differentiation calculation and historical data verification by using the model to obtain a standard attenuation rule; and dynamically calibrating the standard attenuation rule according to real-time monitoring data to obtain a final intensity attenuation report. According to the method, the attenuation law of the material in a complex environment can be accurately described, and high-precision dynamic life prediction is realized.

Description

Technical Field

[0001] This invention relates to the field of drainage board monitoring technology, and in particular to a method and system for predicting the service life of biodegradable drainage boards. Background Technology

[0002] Currently, biodegradable drainage boards are widely used in sponge city construction and soft soil foundation treatment as key components of green infrastructure. Their degradation rate in complex underground environments directly affects the drainage failure risk and long-term safety of engineering projects; therefore, accurately predicting their service life has significant engineering value.

[0003] In existing technologies, the assessment of drainage board lifespan primarily relies on accelerated aging tests under laboratory conditions or simple empirical models based on the time dimension. However, temperature fluctuations, humidity changes, and soil pH in actual service environments exhibit highly dynamic coupling characteristics, making it difficult for traditional methods to effectively simulate the nonlinear degradation process of materials under the synergistic effects of these multiple factors. Furthermore, existing engineering monitoring methods mostly employ discrete, single-point detection devices, lacking a distributed, real-time sensing network based on intelligent sensors. This makes it difficult to capture subtle changes in the deep soil microenvironment and their immediate impact on the local strength decay of the material, resulting in the inability to form a continuous and complete data chain to support high-precision lifespan modeling.

[0004] Therefore, existing technologies suffer from insufficient accuracy in predicting the service life of biodegradable drainage boards. Summary of the Invention

[0005] This invention provides a method and system for predicting the service life of biodegradable drainage boards, in order to solve the technical problem of insufficient accuracy in predicting the service life of biodegradable drainage boards in the prior art.

[0006] In a first aspect, to address the aforementioned technical problems, the present invention provides a method for predicting the service life of a biodegradable drainage board, comprising: Collect multidimensional raw environmental data of the target monitoring area, and perform denoising and multi-source weighted fusion processing on the multidimensional raw environmental data to obtain an integrated environmental dataset; Feature extraction and spatiotemporal evolution analysis are performed on the integrated environment dataset to obtain the spatiotemporal feature evolution path; Multi-factor correlation calculations are performed on the spatiotemporal feature evolution path to determine the interaction weight parameters; The spatiotemporal feature evolution path is used as input, and the interaction weight parameters are combined to train the model, resulting in an initial decay model. Based on the initial attenuation model, regional differential attenuation calculations are performed on the integrated environmental dataset to obtain the non-uniform attenuation distribution pattern. Historical degradation data is obtained, and the non-uniform decay distribution pattern is compared with the historical degradation data to verify the deviation and obtain the standard decay pattern. Real-time monitoring data is acquired, and dynamic calibration and lifetime prediction are performed based on the standard attenuation law and the real-time monitoring data to obtain the final intensity attenuation report.

[0007] Secondly, the present invention provides a lifespan prediction system for biodegradable drainage boards, comprising: The data processing module is used to collect multidimensional environmental raw data of the target monitoring area, and to perform noise reduction and multi-source weighted fusion processing on the multidimensional environmental raw data to obtain an integrated environmental dataset. The feature extraction module is used to extract features and perform spatiotemporal evolution analysis on the integrated environment dataset to obtain the spatiotemporal feature evolution path; The weight calculation module is used to perform multi-factor correlation calculations on the spatiotemporal feature evolution path and determine the interaction weight parameters; The model building module is used to take the spatiotemporal feature evolution path as input and combine it with the interaction weight parameters to train the model and obtain the initial decay model. The spatial analysis module is used to perform regional differential attenuation calculations on the integrated environmental dataset based on the initial attenuation model to obtain the non-uniform attenuation distribution pattern. The verification module is used to acquire historical degradation data and compare the deviation between the non-uniform decay distribution pattern and the historical degradation data to obtain the standard decay pattern. The prediction application module is used to acquire real-time monitoring data, perform dynamic calibration and lifetime prediction based on the standard attenuation law and the real-time monitoring data, and obtain the final intensity attenuation report.

[0008] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention collects multidimensional environmental raw data of the target monitoring area and performs noise reduction and multi-source weighted fusion processing to extract the spatiotemporal feature evolution path. This deep cleaning and fusion mechanism of multidimensional data effectively eliminates the interference of environmental noise, breaks the limitations of a single data source, and can accurately capture the spatiotemporal dynamic changes of temperature, humidity and acidity / alkalinity factors in complex environments, providing a high-fidelity and comprehensive data foundation for subsequent lifespan prediction.

[0009] (2) This invention analyzes the coupling correlation between multiple factors and calculates the interaction strength to determine the interaction weight parameters to correct the model training. This method innovatively integrates the environmental coupling mechanism of the physical world into the training process of the neural network. Through weight constraints, it effectively simulates the nonlinear influence of the synergistic effect of multiple factors on the material strength, solves the problem that traditional models are difficult to characterize the internal attenuation mechanism of materials under complex coupling environments, and significantly improves the generalization ability and prediction accuracy of the attenuation model.

[0010] (3) This invention utilizes a model to perform regionally differentiated attenuation calculations and spatial mapping analysis, and combines real-time monitoring data for dynamic calibration. This closed-loop feedback mechanism can not only accurately identify the uneven attenuation patterns of materials in different regions and locate potential local failure risks, but also dynamically correct the prediction results based on real-time environmental changes, eliminating the cumulative errors in long-term predictions and providing timely and targeted scientific decision-making basis for engineering maintenance. Attached Figure Description

[0011] Figure 1 This is a schematic flowchart of a method for predicting the service life of a biodegradable drainage board provided in the first embodiment of the present invention; Figure 2 This is a schematic diagram of a lifespan prediction system for a biodegradable drainage board provided in the second embodiment of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] Reference Figure 1 The first embodiment of the present invention provides a method for predicting the service life of a biodegradable drainage board, comprising the following steps: S11, Collect multidimensional environmental raw data of the target monitoring area, and perform denoising and multi-source weighted fusion processing on the multidimensional environmental raw data to obtain an integrated environmental dataset; S12, Perform feature extraction and spatiotemporal evolution analysis on the integrated environment dataset to obtain the spatiotemporal feature evolution path; S13, perform multi-factor correlation calculation on the spatiotemporal feature evolution path to determine the interaction weight parameters; S14, The spatiotemporal feature evolution path is used as input, and the interaction weight parameters are combined to train the model to obtain the initial decay model; S15, Based on the initial attenuation model, perform regional differential attenuation calculation on the integrated environmental dataset to obtain the non-uniform attenuation distribution law; S16, acquire historical degradation data, and compare the deviation between the uneven decay distribution law and the historical degradation data to verify the standard decay law; S17, acquire real-time monitoring data, perform dynamic calibration and lifetime prediction based on the standard attenuation law and the real-time monitoring data, and obtain the final intensity attenuation report.

[0014] In step S11, multidimensional environmental raw data of the target monitoring area is collected, and the multidimensional environmental raw data is subjected to denoising and multi-source weighted fusion processing to obtain an integrated environmental dataset, including: Collect multidimensional raw environmental data of the target monitoring area, and perform format standardization and outlier removal to obtain a standardized dataset; The standardized dataset is subjected to time-series smoothing filtering to obtain denoised environmental data; The denoised environmental data is subjected to multi-source feature weighting calculation, and timestamps and regional labels are added to obtain the integrated environmental dataset.

[0015] It should be noted that the collection of multidimensional environmental raw data of the target monitoring area is achieved by deploying a distributed sensor network in the area where drainage boards are buried (such as deep in soft soil foundations or underground drainage channels). This network integrates temperature sensors (such as thermocouples), humidity sensors (such as frequency domain reflectance (FDR) sensors), and soil pH sensors. These sensors continuously collect analog signals at a preset sampling frequency (e.g., once every 5 minutes) and convert them into digital signals through an analog-to-digital converter. The collected data includes sensor IDs, raw readings, and the time of collection. Format unification processing refers to the process where edge computing nodes receive raw data packets from different manufacturers and using different protocols (such as Modbus and ZigBee), parse them, and map them into a unified JSON key-value pair format (e.g., {"type":"pH", "val":6.5, "unit":"pH"}). Outlier removal is handled using... The criterion (Raida criterion) is used to clean the data. The system calculates the average readings of similar sensors within the same area over a short time window. and standard deviation It will fall within the interval Any readings other than those indicated are considered gross errors caused by sensor malfunction or electromagnetic interference and are therefore discarded.

[0016] It is worth noting that the determination of the preset sampling frequency is based on statistical analysis of the degradation rate of historical drainage boards and their sensitivity to environmental changes. By analyzing the fluctuation frequency of environmental parameters (temperature, humidity) in historical data (i.e., spectrum analysis), a frequency value greater than twice the highest frequency of environmental changes is selected according to the Nyquist sampling theorem to ensure that the dynamic details of environmental changes can be recovered without distortion.

[0017] It should be noted that for small sample data (e.g., data volume less than 10), 2 can be used. The criteria aim to improve the sensitivity of outlier detection; for example, an edge node in region A collects readings from a set of 5 pH sensors, [6.4, 6.5, 6.4, 8.2, 6.5]. The calculated mean is 6.8, and the standard deviation is 0.7. According to 2 According to criterion 8.2, values ​​falling within the boundary are considered outliers and removed. The remaining valid data, after being formatted, constitute the standardized dataset.

[0018] It should be noted that time-series smoothing filtering is applied to the standardized dataset to eliminate the impact of minor environmental fluctuations and quantization noise on the data trend. In this embodiment, a Savitzky-Golay filter (SG filter) is used. This filter uses the least squares method to fit a low-order polynomial to smooth the data points within the sliding window. For each sampling point... its smoothed value Through formula The calculation yielded, where To adjust the sliding window size, The convolution coefficients are pre-calculated based on the polynomial order. The SG filter effectively preserves the peak and trough characteristics of the signal while removing high-frequency noise, avoiding the data clipping phenomenon caused by the traditional moving average method.

[0019] It is worth noting that the convolution coefficients The determination of the value is based on the least squares fitting principle. It is obtained by constructing a system of normal equations with a preset polynomial order (e.g., order 2) and a sliding window size (e.g., 5), and then solving for its generalized inverse matrix. The sliding window size... The determination of the optimal window size is based on autocorrelation analysis of historical environmental data. The autocorrelation function of the data is calculated, and the lag time at which the autocorrelation coefficient decays to an autocorrelation decay threshold (e.g., 0.5) is selected as the optimal window size. This autocorrelation decay threshold is derived from the coherence time statistics of the signal's dominant frequency components in historical data, ensuring that the filtering window covers the main region of the signal, smoothing noise without obscuring the true rapid changing trend.

[0020] For example, a continuous temperature data sequence is processed using an SG filter with a window size of 5 and a polynomial order of 2. After processing, the high-frequency jitter of ±0.1°C caused by circuit noise in the original data is filtered out, resulting in a smooth temperature change curve, which is the denoised environmental data.

[0021] It should be noted that performing multi-source feature weighting calculation on the denoised environmental data refers to fusing data from multiple redundant sensors into a single value representing the overall state of the area within the same monitoring sub-region. This embodiment employs inverse variance weighting. First, the variance of each sensor over a period of time is calculated. A smaller variance indicates a more stable sensor. Then, weights are assigned to each sensor. The final fusion value Adding timestamps and geographic tags means binding the merged data with the precise system time (such as NTP synchronization time) and the preset geographic grid ID (such as Zone-A-01) to form a complete record.

[0022] It is worth noting that the preset geographic grid ID is determined based on spatial discretization processing of the Geographic Information System (GIS) coordinates at the engineering site. The system divides the entire monitoring area into several independent rectangular grids according to a preset spatial resolution (e.g., 5m x 5m), and assigns a unique identifier to each grid as the geographic grid ID. The preset spatial resolution is determined based on spatial autocorrelation analysis of environmental parameters (such as soil moisture) within the monitoring area. Specifically, this analysis first collects data at different intervals within the monitoring area. The following data point pairs Then, using the semivariogram formula... Calculate the experimental variance, where The spacing is Number of point pairs For position The parameter values ​​at the specified location are then determined. Next, a spherical model is used to fit the experimental variability values, and the lag distance when the variability function reaches the sill value is extracted from the fitted curve, i.e., the range. Finally, this range value is set as the upper limit of the grid side length to ensure that the environmental state inside the grid has statistical uniformity.

[0023] For example, in region A, the variance of sensor 1 is 0.01, and the variance of sensor 2 is 0.04. According to the inverse variance weighting method, the weight of sensor 1 is 0.8, and the weight of sensor 2 is 0.2. If the reading of sensor 1 is 25.0°C and the reading of sensor 2 is 25.5°C, then the fused temperature value is... °C. This value, along with the timestamp "2025-11-20 10:00:00" and the region A label, constitutes a record in the integrated environmental dataset.

[0024] In step S12, feature extraction and spatiotemporal evolution analysis are performed on the integrated environment dataset to obtain the spatiotemporal feature evolution path, including: The integrated environment dataset is segmented according to a preset time window to obtain segmented time-series data; The fluctuation amplitude and abrupt change points in the segmented time series data are extracted to obtain the nonlinear change characteristics; Spatial dimension distribution analysis and difference measurement calculation are performed on the integrated environment dataset to obtain regional distribution characteristics; The spatiotemporal feature evolution path is obtained by temporally associating the nonlinear change characteristics with the regional distribution characteristics.

[0025] It should be noted that the segmentation of the integrated environmental dataset according to a preset time window is achieved using a sliding window processing method. This method defines a fixed-length time window and a step size, which slides along the data sequence on the time axis. For each window position, all environmental parameter data contained therein (specifically including temperature, humidity, and soil pH sequences) are extracted to form an independent time slice sample, i.e., segmented time-series data. This operation transforms long-period continuous monitoring data into a series of short-period local data blocks to capture local dynamic characteristics.

[0026] It is worth noting that the length of the preset time window (e.g., 24 hours) is determined based on spectral analysis of historical environmental data. By performing a Fast Fourier Transform (FFT) on the historical data, the main cycles of environmental parameter changes (such as the diurnal temperature range cycle) are identified. Based on the engineering application principles of the Nyquist sampling theorem, a duration that can completely cover at least one main cycle of change is selected as the window length to ensure that the window contains complete fluctuation pattern information.

[0027] For example, suppose historical data shows that the main period of geothermal change is 24 hours. The system sets the sliding window length to 24 hours and the step size to 1 hour. For a 30-day integrated environmental dataset, it is segmented into... A number of overlapping segmented time-series data blocks.

[0028] It should be noted that extracting fluctuation amplitudes and abrupt change points from the segmented time-series data aims to quantify the nonlinear dynamic behavior of environmental parameters. For fluctuation amplitudes, the system calculates the root mean square error (RMSE) of the data within each segment to measure the dispersion and fluctuation energy of the data. For abrupt change points, the system employs the Pettitt mutation test algorithm. This is a nonparametric test method used to detect time points in a time series where the mean changes significantly. The algorithm calculates any time point in the sequence... Order column statistics of the data before and after ,like Corresponding statistical significance If the value is less than the preset significance level (e.g., 0.05), then that moment is considered significant. These are the mutation points. The calculated RMSE values, mutation point times, and mutation magnitudes together constitute the nonlinear variation characteristics.

[0029] It is worth noting that the preset significance level (e.g., 0.05) is determined based on the Type I error rate control standard in statistical hypothesis testing. In the field of environmental monitoring, a 95% confidence level (corresponding to a significance level of 0.05) is typically used to balance the sensitivity and false alarm rate of mutation detection, ensuring that the identified mutations are statistically significant, rather than noise generated by random fluctuations.

[0030] For example, in a 24-hour temperature segmentation data set, the calculated RMSE is 2.5°C, indicating significant temperature fluctuations. The Pettitt test found that the statistic at the 10th hour... Reaching peak and This indicates that a significant temperature jump occurred at that moment (e.g., caused by rainfall). This information was extracted as nonlinear variation features, {rms: 2.5, mutation_time: 10h, mutation_sig: 0.01}.

[0031] It should be noted that spatial distribution analysis and difference measurement calculation of the integrated environmental dataset are crucial for assessing environmental heterogeneity among different monitoring areas. Distribution analysis employs Moran's I to calculate global spatial autocorrelation. This autocorrelation differs from the spatial autocorrelation analysis mentioned in step S11, which determines the distance range (range) of spatial correlation based on a semi-variogram to set the grid size; this step, however, identifies the clustering patterns of environmental parameters (such as humidity) in spatial distribution based on the established grid. Specifically, the calculated Moran's I... The range is ,like Furthermore, it passed the significance test and was determined to be a clustered distribution (i.e., high values ​​are adjacent to high values, and low values ​​are adjacent to low values); if It is determined to be a discrete distribution (i.e., a distribution with alternating high and low values); if If so, it is determined to be a random distribution.

[0032] In one alternative implementation, the difference metric is calculated using the Mahalanobis distance method. For each predefined geographic grid cell, the system first constructs its multidimensional environmental feature vector. This includes the average temperature, average humidity, and average pH within the grid; then, the average feature vector of the entire monitoring area is calculated. Covariance Matrix Finally, through the formula Calculate the Mahalanobis distance for this grid. This distance, through the covariance matrix, eliminates the influence of different variable units (such as °C and %) and the correlation between variables (such as the coupling of temperature and humidity), accurately quantifying the degree of deviation (i.e., variability) of this region relative to the overall environment. The calculated Moran index and the Mahalanobis distance values ​​for each grid represent the regional distribution characteristics.

[0033] It is worth noting that the determination of the preset geographic grid unit is consistent with that described in step S11. It is a unique spatial identifier unit obtained by discretizing the geographic information system (GIS) coordinates of the engineering site according to the spatial resolution (e.g., 5 meters × 5 meters) determined in S11.

[0034] For example, when calculating the regional distribution characteristics at a certain moment, the global Moran index is 0.6 ( This indicates that high humidity areas exhibit significant spatial clustering (such as groundwater accumulation zones). For grid cell "Zone-A-05", the Mahalanobis distance between its environmental feature vector and the global mean is 3.5, significantly higher than the average level, indicating that the environmental conditions in this area are highly specific.

[0035] It should be noted that mapping the nonlinear variation characteristics to the regional distribution characteristics over time is a step in constructing a four-dimensional (three-dimensional space + one-dimensional time) evolution model. This embodiment employs the tensor construction method. The system constructs a fourth-order tensor. ,in Spatial grid coordinates, For time step, For feature dimensions. The system will use each time step... Each grid The corresponding nonlinear variation characteristics (such as RMSE) and regional distribution characteristics (such as Mahalanobis distance) are filled into the corresponding positions of the tensor. Subsequently, the tensor is expanded into a sequence of state vectors in the time dimension, forming a trajectory describing the dynamic evolution of the environmental state over time and space, i.e., the spatiotemporal feature evolution path.

[0036] For example, at time , grid The state vector is At any moment The state vector of the grid evolves into Linking these continuous state vectors in chronological order constitutes the spatiotemporal evolution path of the grid point.

[0037] In step S13, multi-factor correlation calculations are performed on the spatiotemporal feature evolution path to determine the interaction weight parameters, including: The spatiotemporal feature evolution path is subjected to trajectory extraction processing to obtain the changing trajectory; Correlation analysis was performed on the aforementioned change trajectory to obtain a coupling correlation index; If the coupling correlation index exceeds the preset correlation threshold, then the multi-factor superposition influence calculation is performed to obtain the interaction strength. Based on the interaction strength, parameter mapping is performed to obtain the interaction weight parameters.

[0038] It should be noted that the trajectory extraction processing of the spatiotemporal feature evolution path is achieved using a feature dimension separation method. Since the spatiotemporal feature evolution path output by S12 is a high-dimensional state sequence containing multi-dimensional features (specifically including root mean square error of temperature, soil pH mutation value, and Mahalanobis distance of humidity), this method first loads a pre-set feature index mapping table. This mapping table is constructed during the feature engineering initialization phase. The construction process is as follows: the system traverses all feature calculation logic defined in S12, specifically including root mean square error calculation for temperature, humidity, and soil pH data, Pettitt mutation detection, Moran's index calculation, and Mahalanobis distance calculation. According to the order in which the calculation results are written into the tensor, a unique index number is assigned to each feature dimension, and the correspondence between the index and physical environmental factors (such as temperature, humidity, and pH) and statistical attributes is recorded to form a key-value pair structure. Subsequently, the system traverses the state vector at each time step in the evolution path, slices the vector according to the index mapping table, extracts the feature values ​​belonging to the same physical factor, and reassembles them in chronological order, thereby decoupling the mixed high-dimensional sequence into several independent time series representing the changes of a single environmental factor over time. These independent time series are the change trajectories, such as temperature fluctuation trajectories and pH change trajectories.

[0039] For example, suppose the spatiotemporal feature evolution path is a time series of length 24, where each time point is a 5-dimensional vector. Index mapping table definition This is a temperature characteristic. This is a humidity-related feature. The trajectory extraction process will traverse the sequence and extract all... Construct a temperature fluctuation trajectory sequence and extract all of them. This forms a sequence of humidity change trajectories, thereby achieving physical decoupling.

[0040] It should be noted that the correlation analysis of the aforementioned change trajectories aims to quantify the temporal correlation characteristics (covering synchronous changes and lagged responses) of different environmental factors during dynamic changes. This embodiment employs a cross-correlation function (CCF). For the change trajectories of two environmental factors... and The system calculates their lag at different times. Cross-correlation coefficients under Its calculation formula is, in, For the trajectory length, and These are the average values ​​of the two trajectories. The system iterates through all lag times within a preset range. Select The maximum value of the coefficient is used as a scalar value to measure the degree of correlation between the two factors; this maximum value is the coupling correlation index. This method can not only identify factors that change synchronously, but also capture the causal coupling relationship where a change in one factor leads to a delayed change in another factor.

[0041] It is worth noting that the preset association threshold (e.g., 0.7) is determined based on statistical significance analysis of historical multi-source environmental data. The system calculates the cross-correlation coefficients of unrelated environmental factor pairs (such as random noise and temperature) in a large amount of historical data, constructing a probability distribution model of invalid correlations. The upper bound of the 99% confidence interval of this distribution (i.e., the critical value corresponding to p-value < 0.01) is selected as the association threshold. This ensures that only statistically significant factor pairs are considered to have physical coupling.

[0042] For example, the system calculates the cross-correlation coefficient between the temperature fluctuation trajectory and the pH change trajectory, and finds that in the lag time... At the hour mark, the coefficient reached a peak of 0.85. Since 0.85 is greater than the preset threshold of 0.7, the system determined that there is a significant strong coupling relationship between temperature and pH (e.g., increased temperature accelerates the release of acidic substances).

[0043] It should be noted that the calculation of the combined effects of multiple factors is to quantify the nonlinear enhancement effect (i.e., the "1+1>2" effect) on material degradation when strongly coupled factors act together. This embodiment uses a coupling degree model. For factors whose correlation index exceeds a threshold... and The strength of their interaction Through formula Calculation. Among them, It is a coupling correlation index. These are the normalized fluctuation amplitudes of the two factors within the current time window. Specifically, and This refers to the nonlinear change characteristics extracted in step S12 (such as the root mean square error (RMSE) of temperature and the abrupt change amplitude of pH value, for example, The dimensionless values ​​obtained after performing Min-Max Normalization are in the range of [0, 1], thus eliminating the dimensional differences between different environmental parameters (such as °C and dimensionless pH). This is a preset gain coefficient. The formula shows that the higher the correlation and the more violent the fluctuations of each component, the more exponentially the destructive force of their interaction increases.

[0044] It is worth noting that the gain coefficient The determination of the coefficients was based on fitting experimental data of material degradation kinetics. By simulating different levels of environmental coupling conditions in the laboratory (such as simultaneous high temperature and high humidity versus individual conditions), the differences in the material strength decay rate were measured, and nonlinear regression analysis was used to solve for the coefficients that best fit the experimental data. Specifically, temperature-humidity coupled conditions (e.g., 20–40°C, 60–90%RH) were simulated in a controlled environment chamber, the strength decay rate of the drainage board was measured, and the least squares method was used for fitting. The value is set such that it satisfies the criterion of minimizing the variance of the prediction error. The value range is usually 0.5–2.0, with 1.0 being preferred.

[0045] It should be noted that parameter mapping based on the interaction strength converts the physical strength values ​​into weight coefficients usable in the neural network loss function. This embodiment uses the Sigmoid activation mapping function. The system will calculate the interaction strength... Input function ,in This represents the maximum saturation value of the weight (e.g., 2.0). The center offset, This is the slope coefficient. This mapping smoothly maps unbounded intensity values ​​to a bounded interval of positive numbers, and the resulting value is the interaction weight parameter. A global feature interaction weight matrix is ​​constructed, which will be used in S14 to calculate the physical constraint weights for each training sample, forcing the model to focus on environmental features with strong interactions.

[0046] For example, the strength of the interaction between temperature and pH is calculated. After Sigmoid mapping (let's assume...) The weight parameters are calculated. This means that in subsequent model training, the contribution of features involving the combined fluctuations of temperature and pH to the loss function will be amplified by this parameter (by approximately 1.64 times), thereby improving the model's sensitivity to this combined risk factor.

[0047] In step S14, the spatiotemporal feature evolution path is used as input, and the interaction weight parameters are combined to train the model, resulting in an initial decay model, including: A model structure for predicting intensity decay is constructed. The spatiotemporal feature evolution path is processed by vector mapping to obtain an input vector sequence, and the input vector sequence is input into the model structure for predicting intensity decay. The interaction weight parameters are constrained and fused with a preset loss function to obtain a weighted objective function; The intensity attenuation prediction model structure is iteratively trained based on the weighted objective function until the model converges, thus obtaining the initial attenuation model.

[0048] It should be noted that the intensity decay prediction model structure is constructed using a Long Short-Term Memory (LSTM) network architecture. The LSTM model constructed in this embodiment includes an input layer, hidden layers, and an output layer. The input layer accepts time-series data with dimensions of (time step, feature dimension); the hidden layer consists of several stacked LSTM units, each containing a forget gate, an input gate, and an output gate. The activation function controls the flow and memory of information, and a Dropout layer (with a dropout rate of 0.2) is set between every two LSTM layers to prevent overfitting; the output layer is a fully connected (Dense Layer) used to map the output of the hidden layer to the final intensity decay prediction value (i.e., the normalized percentage of intensity retention).

[0049] It should be noted that vector mapping processing of the spatiotemporal feature evolution path refers to converting the evolution path data output by S12 into a tensor format adapted to the model input. The system first loads a historical intensity monitoring dataset, which contains the true values ​​of material strength decay obtained through regular laboratory testing or accelerated aging tests, synchronized with environmental data timestamps. Then, using the Z-Score normalization method, environmental feature values ​​(such as temperature and humidity) and historical intensity values ​​of different dimensions in the evolution path are uniformly mapped to a distribution interval with a mean of 0 and a standard deviation of 1. Finally, a sliding time window technique is used to reconstruct the continuous time series data. Specifically, the reconstruction logic is as follows: the time window length is set to... (e.g., 24) and prediction step size (For example, 1). For a length of Standardized environmental feature sequence and the corresponding intensity label sequence From the index Start sliding to extract environmental feature subsequences. As input samples, the corresponding future intensity values ​​are obtained. This serves as the corresponding label value. Through this operation, the two-dimensional time series is transformed into a three-dimensional input vector sequence with shape (number of samples, time step, number of features) and its corresponding intensity label, thereby constructing a complete supervised learning dataset.

[0050] It is worth noting that the determination of the number of LSTM hidden layers (e.g., 2 layers) and the number of neurons per layer (e.g., 64) is based on the optimization results of a grid search on a historical dataset. The system performs traversal training within a preset hyperparameter space, evaluates model performance using K-fold cross-validation, and finally selects the set of hyperparameters that minimizes the average error on the validation set as the preset configuration of the model structure. The preset hyperparameter space is set according to the principle of matching model capacity with dataset size. According to statistical learning theory, to avoid underfitting or overfitting, the number of parameters should be linearly or logarithmically proportional to the number of training samples. Therefore, based on the sample size of historical data (e.g., 100,000 samples), the system limits the search range for the number of layers to [1, 3] and the search range for the number of neurons to [32, 128] to ensure that the search process covers the optimal solution within the limits of computational resources; during grid search, 5-fold cross-validation is used to select the configuration with the smallest MAPE on the validation set.

[0051] For example, the spatiotemporal feature evolution path includes 10 specific feature dimensions: real-time temperature, real-time humidity, real-time pH value, root mean square error (RMSE) of temperature fluctuation, RMSE of humidity fluctuation, RMSE of pH fluctuation, amplitude of temperature abrupt change, amplitude of pH abrupt change, Moran's index of spatial autocorrelation, and Mahalanobis distance of spatial heterogeneity. The system sets the time step to 24 (i.e., predicting future states from data of the past 24 hours). After vector mapping processing, the input data is transformed into a shape... The tensor sequence is input into a model structure containing two LSTM layers (64 units per layer).

[0052] It should be noted that the constraint fusion of the interaction weight parameters with the preset loss function aims to forcibly inject the coupling mechanism of the physical world (the weights determined by S13) into the data-driven neural network. This embodiment uses mean squared error (MSE) as the basis for the preset loss function, constructing the following physically constrained weighted objective function. , in, The total number of samples, This represents the actual attenuation value. These are the model's predicted values. These are the physical constraint scalar weights of the sample during training. The calculation logic is as follows: First, obtain the interaction weight parameters determined in step S13 (i.e., the global coupling weight matrix between feature pairs). Then, for the first... The nth sample (i.e., the nth sample) Feature vectors at each time step ), calculate the instantaneous coupling stress of all feature pairs at that moment. This value reflects whether strongly coupled factors (such as high temperature and high humidity) are simultaneously at a high level at the current moment; finally, for Perform normalization processing (mapped to) interval, for example ), to obtain the final sample weights By minimizing this weighted objective function, the model is forced to prioritize fitting samples in high-coupling-risk environments during training, thereby significantly improving the model's prediction accuracy under complex conditions.

[0053] It should be noted that the iterative training of the intensity decay prediction model structure based on the weighted objective function is implemented using the Adam optimizer and the backpropagation algorithm (BPTT). During training, the system sets the initial learning rate to 0.001 and the batch size to 32. The input vector sequence is input into the model in batches, the predicted value is calculated through forward propagation, the weighted error is calculated using the weighted objective function, and the gradient is calculated using the chain rule to update the model parameters. The model converges, i.e., the early stopping condition is met. When the loss function value on the validation set no longer decreases within a preset number of rounds (e.g., 10 rounds), the model is considered to have converged, training is stopped, and the current parameters are saved, thus obtaining the initial decay model.

[0054] For example, in one training iteration, a sample is in a high-temperature, high-humidity, and strongly coupled environment, and S13 calculates its interaction weights. If the model's prediction of the sample is significantly off, the error term in the weighted objective function will be amplified by a factor of 2, resulting in a larger gradient value, which in turn significantly adjusts the weight parameters during backpropagation. After 100 epochs of training, the validation set loss stabilizes below 0.001, and training automatically terminates. The resulting model is the initial decay model that accurately reflects the coupled effects of multiple factors.

[0055] In step S15, based on the initial attenuation model, regionally differentiated attenuation calculations are performed on the integrated environment dataset to obtain the non-uniform attenuation distribution pattern, including: Based on the initial decay model, a theoretical rate estimate is performed on the integrated environment dataset to obtain a set of regional decay rates; Gradient calculation is performed on the set of regional attenuation rates to obtain regional difference characteristics; The regional difference features are mapped onto a preset coordinate grid, and high difference is marked to obtain the non-uniform attenuation distribution law.

[0056] It should be noted that the theoretical rate estimation of the integrated environment dataset based on the initial decay model is achieved through sliding inference. The system iterates through each spatial monitoring node in the integrated environment dataset, inputting its corresponding historical time-series data into the initial decay model trained in S14, to predict the residual strength of the material after a preset time step (e.g., 30 days). Subsequently, the residual strength is calculated using the formula... Calculate the theoretical decay rate of this node, where This is the current intensity value. To predict the intensity value, The time step is defined as . The calculation results of all nodes are organized into a two-dimensional matrix that corresponds one-to-one with their physical space coordinates, i.e., the set of regional decay rates.

[0057] For example, for coordinates... At the monitoring point, the model predicts that the intensity will decrease from 90 MPa to 87 MPa after 30 days. The theoretical decay rate at this point is calculated to be 0.1 MPa / day. The system performs this operation on all monitoring points in the coverage area, forming a rate matrix describing the distribution of degradation rates across the entire engineering area.

[0058] It should be noted that gradient calculation on the set of decay rates in the regions aims to quantify the degree of abrupt change in decay rates between adjacent regions, i.e., to identify differentiated physical boundaries. In this embodiment, the Sobel operator is used for spatial gradient calculation. The system calculates the first-order difference approximations of the rate matrix along the X and Y axes, respectively. and And then through the formula Calculate the gradient magnitude at each grid point. This gradient magnitude directly reflects the degree of non-uniformity in degradation rate between that point and the surrounding environment, i.e., the regional difference characteristic; the gradient calculation is based on a preset coordinate grid, with the distance unit uniformly in meters, so the gradient unit is (MPa / day) / meter.

[0059] For example, in a local region, the decay rate at the center point is 0.1, while the rate at the adjacent point to its right increases sharply to 0.5 (possibly due to the passage of groundwater flow). The Sobel operator calculates that the gradient magnitude at this location is large, indicating the presence of a significant non-uniform decay boundary.

[0060] It should be noted that mapping the regional difference features to a preset coordinate grid and marking high-difference areas is achieved using a threshold segmentation method. The system first acquires the preset coordinate grid (i.e., the GIS grid system at the engineering site) and fills the calculated gradient magnitude into the corresponding grid cells. Then, the gradient value of each cell is compared with a preset difference threshold. If the gradient value exceeds the threshold, the grid cell is marked as a high-risk non-uniform area, and its coordinate range is recorded. These marked grids and their gradient data together constitute the non-uniform attenuation distribution pattern.

[0061] It is worth noting that the preset difference threshold was determined based on survival analysis of historical engineering failure cases. The system collected a large amount of environmental data surrounding the locations of historical drainage board fractures and failures, and retroactively calculated their decay rate gradients. Statistical analysis revealed that when the spatial gradient of the decay rate exceeds a certain critical value, the probability of material fracture due to internal stress difference increases significantly (e.g., exceeding 80%). Selecting this critical value (e.g., the 90th percentile of the gradient amplitude distribution) as the preset difference threshold ensures that the marking results have clear engineering safety warning significance.

[0062] For example, in a preset coordinate grid, the gradient magnitude calculated for grid cell "Zone-C-12" is 0.05. The preset difference threshold determined based on survival analysis is 0.03. Since 0.05 is greater than 0.03, the system marks this "Zone-C-12" grid cell as a high-risk non-uniform region and records its state as abnormal in the final distribution data, indicating a potential risk of fracture due to uneven degradation rates in this area.

[0063] In step S16, historical degradation data is acquired, and the non-uniform decay distribution pattern is compared and verified with the historical degradation data to obtain the standard decay pattern, including: Obtain historical degradation data that includes actual attenuation values; The predicted attenuation value in the uneven attenuation distribution law is numerically compared with the actual attenuation value to calculate the prediction deviation value. The distribution characteristics of the predicted deviation values ​​are analyzed to obtain deviation statistical indicators; If the deviation statistics meet the preset verification conditions, then the non-uniform attenuation distribution law is determined to be the standard attenuation law.

[0064] It should be noted that obtaining historical degradation data containing actual attenuation values ​​refers to retrieving historical records of similar projects (including completed and under-construction projects) with similar geological conditions to the current monitoring area from a pre-set engineering quality database. The pre-set engineering quality database is constructed by the system collecting a large amount of drainage board application engineering data under different geological backgrounds, and establishing a structured record for each project containing geological feature vectors (such as soil moisture content, porosity, and average pH value) and periodic sampling intensity data. Retrieving historical records similar to the current monitoring area's geological conditions is achieved using a geological feature similarity matching operation. This operation first constructs the geological feature vector of the current area. Then, calculate the vector and the feature vector of each project record in the database. The Euclidean distance between the samples was used to determine the target historical data. Finally, engineering records with an Euclidean distance less than a preset similarity threshold were selected. This data was obtained through a combination of periodic physical sampling and laboratory tensile testing. Specifically, technicians excavated drainage board samples from the engineering site at preset sampling points (e.g., 30 days and 60 days after installation) and measured their remaining tensile strength using a universal testing machine in a standard laboratory environment. These metrologically verified measured physical values ​​are the actual attenuation values.

[0065] For example, the system constructs a feature vector for the current region [water content: 45%, pH: 6.5] and retrieves the feature vector for project number 2024 from the database [water content: 44%, pH: 6.6]. The Euclidean distance between the two is extremely small, indicating they are similar projects. The system extracts the record for this project on day 180, including the coordinates... The actual measured strength of the drainage board at that location was 65.5 MPa.

[0066] It should be noted that the numerical comparison between the predicted attenuation value and the actual attenuation value in the described non-uniform attenuation distribution pattern is achieved using a spatial coordinate matching operation. The system first indexes the corresponding grid cell in the non-uniform attenuation distribution pattern (i.e., the prediction grid covering the entire region) output by S15 based on the sampling coordinates in the historical degradation data, and extracts the predicted attenuation value of that cell. Then, it uses the relative error formula... Calculate the difference for each sampling point, where For predicted values, The actual value is the calculated set of relative errors, which represents the prediction deviation.

[0067] For example, in coordinates At this point, the model predicted a value of 68.0 MPa, while the actual value was 65.5 MPa. The prediction deviation for this point was calculated. .

[0068] It should be noted that the distribution characteristic analysis of the predicted deviation values ​​aims to evaluate the overall reliability of the model from a statistical perspective. In this embodiment, the system calculates the mean absolute percentage error (MAPE) and standard deviation of the predicted deviation value set. MAPE measures the average accuracy of the prediction results, while standard deviation measures the dispersion (i.e., stability) of the prediction error. These two statistics together constitute the deviation statistics index.

[0069] For example, for a set of deviations containing 50 sampling points, the calculated MAPE is 4.2% and the standard deviation is 1.5%. This set of data {MAPE: 4.2%, Standard Deviation: 1.5%} is the deviation statistic.

[0070] It should be noted that if the deviation statistics meet the preset verification conditions, the non-uniform decay distribution law is determined to be the standard decay law, which is a threshold-based binary determination process. The system compares the calculated MAPE with a preset accuracy threshold and the standard deviation with a preset stability threshold. Only when both are below their respective thresholds (e.g., MAPE < 5% and Std < 2%), the model prediction is deemed to have passed verification. At this time, the distribution law generated in S15 is confirmed as a true value with engineering reference value, i.e., the standard decay law.

[0071] It is worth noting that the preset verification conditions (i.e., setting an accuracy threshold of 5% and a stability threshold of 2%, requiring MAPE < 5% and Std < 2%) are determined based on confidence interval analysis of the safety redundancy coefficient of historical projects. The system statistically analyzed the prediction error data of all engineering projects that had not experienced drainage failure accidents in the past ten years and constructed an error probability distribution. The upper bound of the 95% confidence interval of this distribution was selected as the threshold to ensure that within this error range, the actual performance of the drainage board can still meet the minimum safety factor requirements of the engineering design, thereby guaranteeing the safety and usability of the prediction results.

[0072] In step S17, real-time monitoring data is acquired, and dynamic calibration and lifetime prediction are performed based on the standard attenuation law and the real-time monitoring data to obtain a final intensity attenuation report, including: Obtain real-time monitoring data; The theoretical prediction value is obtained by performing theoretical prediction calculations based on the standard attenuation law. The difference between the theoretical predicted value and the theoretical reference value calculated based on the real-time monitoring data is used to obtain the real-time deviation; If the real-time deviation exceeds the preset calibration threshold, the theoretical prediction value is recalculated to obtain the corrected prediction data. Based on the corrected prediction data, a report generation process is performed to obtain the final intensity decay report.

[0073] It should be noted that real-time monitoring data acquisition is achieved through IoT data subscription operations. The system subscribes to topic messages published by the field sensor network in real time via the MQTT protocol to obtain the current temperature, humidity, and pH data streams. The theoretical prediction calculation based on the aforementioned standard attenuation law refers to the system using the current time... For indexing, the standard attenuation law (i.e., the verified attenuation curve function) determined in S16. By querying within the range of [reference needed], the estimated residual intensity value at that specific moment can be directly retrieved, which is the theoretical prediction value. .

[0074] It should be noted that calculating the difference between the theoretical prediction and the kinetic estimate calculated based on the real-time monitoring data is a crucial step in identifying whether there is physical drift in the data-driven model. In this step, the system employs soft sensing technology and utilizes the Arrhenius kinetic calculation method to construct a reference benchmark based on chemical mechanisms, i.e., the kinetic estimate. This method is based on the principles of chemical reaction kinetics, and utilizes real-time collected environmental parameters (especially temperature). and humidity ), using formula Calculate the degradation rate constant at the current time step. By integrating over time to calculate cumulative losses, the estimated material strength at the current moment under the influence of the actual environment is derived. Then, the real-time deviation was calculated. This bias reflects the degree of inconsistency between data-driven predictions and physical mechanism constraints.

[0075] For example, the LSTM model predicts the current intensity to be 80 MPa. However, real-time monitoring showed a sudden increase in temperature to 35°C over the past 24 hours. Arrhenius kinetic model calculations indicate that the high temperature accelerated degradation; the theoretical kinetic estimate should be 78 MPa. At this point, the real-time deviation MPa.

[0076] It should be noted that if the real-time deviation exceeds the preset calibration threshold, a secondary correction calculation is performed on the theoretical prediction value, which is achieved using a feedback linearization compensation method. When When the calibration threshold is reached, the system considers that the standard law can no longer accurately reflect the current acceleration (or deceleration) decay state. The correction operation introduces a dynamic correction factor. Through formula Calculate the corrected forecast data.

[0077] It is worth noting that the dynamic correction factor... The determination was based on historical model-mechanism deviation analysis. The system statistically analyzed historical data during operation. and The error distribution between the actual physical value and the actual value is fitted with an optimal weighting function using the least squares method. This ensures that the corrected error variance is minimized. The preset calibration threshold is determined based on statistical process control (SPC) analysis of sensor measurement noise and short-term environmental fluctuations. The system analyzes the distribution of calculation deviations caused by normal environmental fluctuations in historical data and calculates their standard deviation. Set the threshold to (Three standard deviations) to ensure that corrections are triggered only when significant environmental changes occur or accumulated errors become too large, thus avoiding system oscillations caused by random sensor noise.

[0078] For example, the calibration threshold is set to 0.5 MPa. In the example above, the deviation is 2 MPa, exceeding the threshold. The system calculates the correction factor. The corrected prediction data MPa.

[0079] It should be noted that the report generation process based on the revised prediction data is implemented using template data binding technology. This technology first loads a pre-built JSON data template, constructed during system initialization according to industry standards for engineering monitoring reports (such as ISO 19901). This template explicitly defines placeholder structures for four key fields: current remaining strength value, predicted remaining useful life (RUL), current health status rating, and maintenance recommendations. Subsequently, the system performs a health status logic judgment. If the revised prediction data (i.e., remaining strength) is higher than a preset safety threshold A (e.g., 80% of design strength), the rating is excellent; if it is between threshold A and a preset warning threshold B (e.g., 50% of design strength), the rating is good; if it is lower than threshold B, the rating is high risk. The system maps the revised prediction data and the determined rating results to the corresponding fields in the template, generating a structured report data stream. Finally, a PDF document generation library (such as iText) is used to convert this data stream into visual charts and text descriptions, resulting in the final strength decay report containing immediate risk warnings and maintenance recommendations.

[0080] It is worth noting that the preset safety threshold A and preset warning threshold B are determined based on reliability statistical analysis of drainage board materials in historical projects. The system collects a large amount of drainage board strength data and their corresponding engineering drainage effects at different service years. Maximum likelihood estimation is used to statistically fit the above data to construct a strength-failure probability distribution function. The strength value corresponding to a failure probability below 0.1% is selected as the safety threshold A (e.g., 80% of the design strength), and the inflection point where the failure probability rises sharply (i.e., the point where the derivative of the probability density function changes abruptly, e.g., 50% of the design strength) is selected as the warning threshold B to ensure that the rating results have statistically significant risk indication.

[0081] In summary, this invention deeply integrates the environmental coupling mechanism of the physical world with data-driven intelligent algorithms by constructing a full-process prediction method, from the acquisition and weighted fusion of multi-dimensional environmental data and the extraction of spatiotemporal feature evolution paths, to the introduction of interaction weight parameters to optimize the neural network model, and finally to differential analysis and real-time dynamic calibration based on spatial mapping. This invention innovatively utilizes the intensity of multi-factor interactions to constrain the attenuation model and combines historical verification and real-time feedback mechanisms, effectively solving the technical problem of insufficient prediction accuracy in complex environments due to the inability to accurately capture nonlinear degradation laws and the uneven attenuation characteristics of materials. This significantly improves the accuracy, robustness, and engineering practical value of predicting the service life of biodegradable drainage boards.

[0082] Reference Figure 2 The second embodiment of the present invention provides a lifespan prediction system for biodegradable drainage boards, comprising: The data processing module is used to collect multidimensional environmental raw data of the target monitoring area, and to perform noise reduction and multi-source weighted fusion processing on the multidimensional environmental raw data to obtain an integrated environmental dataset. The feature extraction module is used to extract features and perform spatiotemporal evolution analysis on the integrated environment dataset to obtain the spatiotemporal feature evolution path; The weight calculation module is used to perform multi-factor correlation calculations on the spatiotemporal feature evolution path and determine the interaction weight parameters; The model building module is used to take the spatiotemporal feature evolution path as input and combine it with the interaction weight parameters to train the model and obtain the initial decay model. The spatial analysis module is used to perform regional differential attenuation calculations on the integrated environmental dataset based on the initial attenuation model to obtain the non-uniform attenuation distribution pattern. The verification module is used to acquire historical degradation data and compare the deviation between the non-uniform decay distribution pattern and the historical degradation data to obtain the standard decay pattern. The prediction application module is used to acquire real-time monitoring data, perform dynamic calibration and lifetime prediction based on the standard attenuation law and the real-time monitoring data, and obtain the final intensity attenuation report.

[0083] It should be noted that the service life prediction system for a biodegradable drainage board provided in this embodiment of the invention is used to execute all the process steps of the service life prediction method for a biodegradable drainage board in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0084] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a lifespan prediction program for a biodegradable drainage board. When the processor executes the computer program, it implements the steps in the various embodiments of the lifespan prediction method for a biodegradable drainage board described above, for example... Figure 1 The step S11 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above system embodiments, such as the data processing module.

[0085] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0086] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0087] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0088] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0089] If the modules / units integrated into the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or system capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0090] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0091] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for predicting the service life of a biodegradable drainage board, characterized in that, include: Collect multidimensional raw environmental data of the target monitoring area, and perform denoising and multi-source weighted fusion processing on the multidimensional raw environmental data to obtain an integrated environmental dataset; Feature extraction and spatiotemporal evolution analysis are performed on the integrated environment dataset to obtain the spatiotemporal feature evolution path; Multi-factor correlation calculations are performed on the spatiotemporal feature evolution path to determine the interaction weight parameters; The spatiotemporal feature evolution path is used as input, and the interaction weight parameters are combined to train the model, resulting in an initial decay model. Based on the initial attenuation model, regional differential attenuation calculations are performed on the integrated environmental dataset to obtain the non-uniform attenuation distribution pattern. Historical degradation data is obtained, and the non-uniform decay distribution pattern is compared with the historical degradation data to verify the deviation and obtain the standard decay pattern. Real-time monitoring data is acquired, and dynamic calibration and lifetime prediction are performed based on the standard attenuation law and the real-time monitoring data to obtain the final intensity attenuation report.

2. The method for predicting the service life of the biodegradable drainage board according to claim 1, characterized in that, The system collects multidimensional raw environmental data of the target monitoring area, and performs denoising and multi-source weighted fusion processing on the raw multidimensional environmental data to obtain an integrated environmental dataset, including: Collect multidimensional raw environmental data of the target monitoring area, and perform format standardization and outlier removal to obtain a standardized dataset; The standardized dataset is subjected to time-series smoothing filtering to obtain denoised environmental data; The denoised environmental data is subjected to multi-source feature weighting calculation, and timestamps and regional labels are added to obtain the integrated environmental dataset.

3. The method for predicting the service life of the biodegradable drainage board according to claim 1, characterized in that, The step of performing feature extraction and spatiotemporal evolution analysis on the integrated environment dataset to obtain the spatiotemporal feature evolution path includes: The integrated environment dataset is segmented according to a preset time window to obtain segmented time-series data; The fluctuation amplitude and abrupt change points in the segmented time series data are extracted to obtain the nonlinear change characteristics; Spatial dimension distribution analysis and difference measurement calculation are performed on the integrated environment dataset to obtain regional distribution characteristics; The spatiotemporal feature evolution path is obtained by temporally associating the nonlinear change characteristics with the regional distribution characteristics.

4. The method for predicting the service life of the biodegradable drainage board according to claim 1, characterized in that, The step of performing multi-factor correlation calculations on the spatiotemporal feature evolution path to determine the interaction weight parameters includes: The spatiotemporal feature evolution path is subjected to trajectory extraction processing to obtain the changing trajectory; Correlation analysis was performed on the aforementioned change trajectory to obtain a coupling correlation index; If the coupling correlation index exceeds the preset correlation threshold, then the multi-factor superposition influence calculation is performed to obtain the interaction strength. Based on the interaction strength, parameter mapping is performed to obtain the interaction weight parameters.

5. The method for predicting the service life of the biodegradable drainage board according to claim 1, characterized in that, The step of using the spatiotemporal feature evolution path as input and combining it with the interaction weight parameters to train the model and obtain the initial decay model includes: A model structure for predicting intensity decay is constructed. The spatiotemporal feature evolution path is processed by vector mapping to obtain an input vector sequence, and the input vector sequence is input into the model structure for predicting intensity decay. The interaction weight parameters are constrained and fused with a preset loss function to obtain a weighted objective function; The intensity attenuation prediction model structure is iteratively trained based on the weighted objective function until the model converges, thus obtaining the initial attenuation model.

6. The method for predicting the service life of the biodegradable drainage board according to claim 1, characterized in that, The step of calculating regionally differentiated attenuation on the integrated environmental dataset based on the initial attenuation model to obtain the non-uniform attenuation distribution pattern includes: Based on the initial decay model, a theoretical rate estimate is performed on the integrated environment dataset to obtain a set of regional decay rates; Gradient calculation is performed on the set of regional attenuation rates to obtain regional difference characteristics; The regional difference features are mapped onto a preset coordinate grid, and high difference is marked to obtain the non-uniform attenuation distribution law.

7. The method for predicting the service life of the biodegradable drainage board according to claim 1, characterized in that, The process of acquiring historical degradation data and comparing the non-uniform decay distribution pattern with the historical degradation data to obtain a standard decay pattern includes: Obtain historical degradation data that includes actual attenuation values; The predicted attenuation value in the uneven attenuation distribution law is numerically compared with the actual attenuation value to calculate the prediction deviation value. The distribution characteristics of the predicted deviation values ​​are analyzed to obtain deviation statistical indicators; If the deviation statistics meet the preset verification conditions, then the non-uniform attenuation distribution law is determined to be the standard attenuation law.

8. The method for predicting the service life of the biodegradable drainage board according to claim 1, characterized in that, The process of acquiring real-time monitoring data, performing dynamic calibration and lifetime prediction based on the standard attenuation law and the real-time monitoring data, and obtaining a final intensity attenuation report includes: Obtain real-time monitoring data; The theoretical prediction value is obtained by performing theoretical prediction calculations based on the standard attenuation law. The difference between the theoretical predicted value and the theoretical reference value calculated based on the real-time monitoring data is used to obtain the real-time deviation; If the real-time deviation exceeds the preset calibration threshold, the theoretical prediction value is recalculated to obtain the corrected prediction data. Based on the corrected prediction data, a report generation process is performed to obtain the final intensity decay report.

9. A lifespan prediction system for biodegradable drainage boards, characterized in that, include: The data processing module is used to collect multidimensional environmental raw data of the target monitoring area, and to perform noise reduction and multi-source weighted fusion processing on the multidimensional environmental raw data to obtain an integrated environmental dataset. The feature extraction module is used to extract features and perform spatiotemporal evolution analysis on the integrated environment dataset to obtain the spatiotemporal feature evolution path; The weight calculation module is used to perform multi-factor correlation calculations on the spatiotemporal feature evolution path and determine the interaction weight parameters; The model building module is used to take the spatiotemporal feature evolution path as input and combine it with the interaction weight parameters to train the model and obtain the initial decay model. The spatial analysis module is used to perform regional differential attenuation calculations on the integrated environmental dataset based on the initial attenuation model to obtain the non-uniform attenuation distribution pattern. The verification module is used to acquire historical degradation data and compare the deviation between the non-uniform decay distribution pattern and the historical degradation data to obtain the standard decay pattern. The prediction application module is used to acquire real-time monitoring data, perform dynamic calibration and lifetime prediction based on the standard attenuation law and the real-time monitoring data, and obtain the final intensity attenuation report.

Images