Real-time water quality detection system

Abstract

The invention relates to a water quality real-time detection system which comprises the following modules: a multi-source sensing module which is based on a multi-parameter sensing array, adopts a self-adaptive sampling strategy, realizes sensor time sequence synchronization through a state estimation algorithm, completes water body multi-dimensional parameter acquisition in combination with a micro-fluidic chip, generates a multi-modal sensing data set, and transmits the multi-modal sensing data set to a data processing module; the multi-source sensing module comprises a multi-source sensing sub-module, a signal conditioning sub-module, a time sequence synchronization sub-module and an anomaly capture sub-module. The method has the advantages that through the synergistic effect of the adaptive sampling strategy and the state estimation algorithm, the multi-sensor time sequence synchronization precision is remarkably improved, the phase deviation problem caused by traditional fixed frequency sampling is effectively eliminated, the sliding window polynomial fitting is combined with the wavelet threshold de-noising technology, and the multi-sensor time sequence synchronization precision is improved. High-frequency noise interference is greatly suppressed on the premise that effective components of the signals are reserved, and meanwhile, the abnormal value detection accuracy is improved through a dynamic threshold mechanism.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of environmental monitoring, in particular to a water quality real-time detection system. BACKGROUND

[0002] The technical field of environmental monitoring mainly involves a technical system for dynamically evaluating environmental quality and monitoring pollution, and covers the research and application of sensor technology, data collection and transmission, pollutant identification algorithms and intelligent early warning systems, and focuses on solving the real-time monitoring needs of environmental elements such as air, water quality and soil. This field integrates Internet of Things architecture, spectral analysis technology, biological sensing mechanism and microfluidic chips and other cutting-edge technologies, realizes multi-parameter synchronous collection through the construction of a distributed monitoring network, conducts pollution tracing analysis combined with machine learning algorithms, and completes the visualization of monitoring data and decision support on the cloud platform. The technical development trend is miniaturization of sensors, intelligentization of detection and real-time response. Among them, a water quality real-time detection system refers to a real-time monitoring system based on a multi-source sensing network and intelligent analysis algorithm, which is used for continuous monitoring of key parameters such as dissolved oxygen, chemical oxygen demand and heavy metal concentration in water bodies. The system realizes real-time collection, anomaly detection and pollution warning of water quality data by integrating high-precision sensor arrays and edge computing units.

[0003] The prior art has the problem of insufficient clock synchronization accuracy in multi-sensor collaborative collection, and the cumulative error is significant due to the dependence on a single time source, which easily produces phase aliasing when analyzing high-frequency change parameters. The traditional denoising method has limited suppression ability for non-stationary noise, and when there is complex interference in the water body, the signal baseline drift is obvious, which affects the detection accuracy. The feature extraction process does not fully consider the influence of dynamic changes of environmental factors on feature correlation, and the misjudgment rate increases in complex scenes. The early warning system usually adopts a single threshold alarm mechanism, lacks multi-level analysis of the evolution law of pollution events, and has a high false negative rate when dealing with complex pollution. SUMMARY

[0004] In order to solve the problem of insufficient clock synchronization accuracy in multi-sensor collaborative collection in the prior art, the dependence on a single time source leads to significant cumulative error, and phase aliasing is easily produced when analyzing high-frequency change parameters, the traditional denoising method has limited suppression ability for non-stationary noise, and when there is complex interference in the water body, the signal baseline drift is obvious, which affects the detection accuracy, the application provides a water quality real-time detection system.

[0005] The water quality real-time detection system provided by the application adopts the following technical scheme: A water quality real-time detection system includes the following modules: The multi-source perception module is based on a multi-parameter sensing array, adopts an adaptive sampling strategy, realizes sensor time sequence synchronization through a state estimation algorithm, combines a microfluidic chip to complete multi-dimensional parameter acquisition of a water body, and generates a multi-modal sensing data set; The multi-source perception module includes a multi-source sensing sub-module, a signal conditioning sub-module, a time sequence synchronization sub-module, and an anomaly capturing sub-module. The data purification module is based on the multi-modal sensing data set, adopts a sliding window polynomial fitting to eliminate baseline drift, uses a wavelet threshold denoising algorithm to suppress high-frequency interference, and uses a Mahalanobis distance detection method to remove outliers, to generate a purified feature matrix. The data purification module includes a signal enhancement sub-module, a noise suppression sub-module, an anomaly cleaning sub-module, and a data alignment sub-module. The feature analysis module is based on the purified feature matrix, adopts an improved principal component analysis for dimension compression, realizes multi-source feature alignment through dynamic time warping, extracts key water quality indicators through mutual information entropy algorithm, and generates a feature fingerprint spectrum. The feature analysis module includes a dimension compression sub-module, a time sequence alignment sub-module, a feature optimization sub-module, and a spectrum construction sub-module. The intelligent diagnosis module is based on the feature fingerprint spectrum, adopts a deep belief network for pollutant identification, realizes water quality grading through a support vector machine, completes comprehensive health degree evaluation through a fuzzy reasoning system, and generates a water quality diagnosis report. The intelligent diagnosis module includes a pollutant identification sub-module, a grade determination sub-module, a health evaluation sub-module, and a report generation sub-module. The dynamic early warning module is based on the water quality diagnosis report, adopts a long short-term memory network for trend prediction, identifies mutation signals through Bayesian change point detection, triggers graded alarms through a multi-level threshold determination method, and generates a risk early warning instruction. The dynamic early warning module includes a trend prediction sub-module, a mutation detection sub-module, a threshold determination sub-module, and an instruction issuing sub-module. The visualization module is based on the risk early warning instruction, adopts a WebGL rendering engine to construct a three-dimensional water quality map, displays pollutant migration paths through a dynamic Sankey diagram, realizes immersive monitoring through virtual reality technology, and generates a decision support dashboard. The visualization module includes a space modeling sub-module, a path tracking sub-module, a situation deduction sub-module, and a dashboard generation sub-module. The self-maintenance module is based on the decision support dashboard, adopts a reinforcement learning algorithm to optimize sensor deployment, predicts equipment health degree through a digital twin model, realizes fault self-diagnosis through self-organizing mapping, and generates a system health record. The self-maintenance module includes a deployment optimization sub-module, a fault prediction sub-module, a self-repair sub-module, and a record update sub-module.

[0006] Preferably, the multi-source sensing sub-module, based on a micro-electrode array, collects pH / ORP parameters using adaptive pulse width modulation, eliminates polarization effects using a three-electrode differential detection method, and generates a set of raw electrochemical signals; The signal conditioning sub-module, based on the set of raw electrochemical signals, uses an instrument amplifier circuit for signal amplification, suppresses high-frequency noise using a Butterworth low-pass filter, and generates pre-processed analog signals; The time synchronization sub-module, based on the pre-processed analog signals, uses a state estimation algorithm to achieve multi-sensor clock synchronization, aligns data using a sliding time window, and generates time-synchronized data packets; The anomaly capture sub-module, based on the time-synchronized data packets, uses the 3σ rule to detect sudden anomalies and uses a sliding t-test method to identify gradual drift, and generates a multi-modal sensing data set.

[0007] Preferably, the signal enhancement sub-module, based on the multi-modal sensing data set, uses a sliding window polynomial fitting to reconstruct the signal baseline, uses adaptive gain control to improve the signal-to-noise ratio, and generates a baseline-corrected signal; The noise suppression sub-module, based on the baseline-corrected signal, uses a wavelet threshold denoising algorithm to decompose high-frequency components, uses an improved threshold function to retain effective features, and generates a denoised feature vector; The anomaly cleaning sub-module, based on the denoised feature vector, uses the Mahalanobis distance detection method to identify outliers, uses the KNN interpolation algorithm to repair missing data, and generates a cleaned data set; The data alignment sub-module, based on the cleaned data set, uses dynamic time warping to align multi-source time series, uses the Pearson correlation coefficient to verify data consistency, and generates a purified feature matrix.

[0008] Preferably, the dimension compression sub-module, based on the purified feature matrix, uses an improved principal component analysis for feature dimension reduction, uses a variance contribution rate threshold method to retain feature information, and generates a reduced dimension feature set; The time alignment sub-module, based on the reduced dimension feature set, uses dynamic time warping to correct the phase difference of multi-source data, uses the cosine similarity to verify the alignment effect, and generates a time alignment matrix; The feature selection sub-module, based on the time alignment matrix, uses the mutual information entropy algorithm to calculate the importance of features, uses the forward search strategy to select the TOP10 key indicators, and generates an optimized feature vector; The atlas construction sub-module, based on the optimized feature vector, uses the radial basis function interpolation to construct a three-dimensional feature space, uses the isosurface extraction algorithm to generate a visual atlas, and generates a feature fingerprint atlas.

[0009] Preferably, the pollutant identification sub-module, based on the feature fingerprint atlas, uses a deep belief network for feature learning, uses a restricted Boltzmann machine stack for pollutant classification, and generates a pollutant classification result; a grade determination sub-module, based on the pollutant classification result, using a support vector machine to build a multi-classification model, realizing five-level division of water quality through a one-versus-one strategy, and generating a water quality grade label; a health assessment sub-module, based on the water quality grade label, using a fuzzy inference system to build an evaluation rule base, calculating a comprehensive health index through a defuzzification algorithm, and generating a health degree index; a report generation sub-module, based on the health degree index, using a template engine technology to integrate multi-dimensional data, automatically generating a diagnostic conclusion through natural language generation, and generating a water quality diagnostic report.

[0010] Preferably, a trend prediction sub-module, based on the water quality diagnostic report, using a long short-term memory network to build a time series prediction model, capturing long-term dependencies through an attention mechanism, and generating a trend prediction curve; a mutation detection sub-module, based on the trend prediction curve, using a Bayesian change point detection to analyze data distribution mutations, identifying risk events through a posterior probability threshold method, and generating a risk event list; a threshold determination sub-module, based on the risk event list, using a multi-level threshold determination method to set three-level alarm rules, managing alarm life cycles through a state machine model, and generating a pre-warning level signal; an instruction issuing sub-module, based on the pre-warning level signal, using a publish-subscribe mode to push alarm information, implementing Internet of Things device linkage through an MQTT protocol, and generating a risk pre-warning instruction.

[0011] Preferably, a spatial modeling sub-module, based on the risk pre-warning instruction, using a WebGL rendering engine to build a three-dimensional river model, generating a water quality distribution surface through a Gaussian process interpolation, and generating a water quality spatial model; a path tracking sub-module, based on the water quality spatial model, using a dynamic Sankey diagram to visualize pollutant migration, simulating diffusion paths through a particle tracking algorithm, and generating a migration path atlas; a situation deduction sub-module, based on the migration path atlas, using virtual reality technology to build an immersive scene, predicting pollution diffusion trends through a fluid dynamics model, and generating a situation deduction animation; a dashboard generation sub-module, based on the situation deduction animation, using data dashboard technology to integrate multi-source views, adapting to different terminal devices through a responsive layout, and generating a decision support dashboard.

[0012] Preferably, a deployment optimization sub-module, based on the decision support dashboard, using deep reinforcement learning to build an optimization model, calculating an optimal sensor layout through a Q-learning algorithm, and generating an optimized deployment scheme; a fault prediction sub-module, based on the optimized deployment scheme, using digital twinning technology to build a device mirror, predicting remaining useful life through a Weibull distribution model, and generating a fault prediction report; The self-repairing submodule, based on the fault prediction report, uses self-organizing mapping for fault mode identification, generates a repair strategy through fault tree analysis, and generates a repair strategy set; The archive updating submodule, based on the repair strategy set, uses blockchain technology to record maintenance logs, updates device lifecycle data through a time series database, and generates a system health archive.

[0013] In summary, the present application includes at least one of the following beneficial technical effects: Through the synergistic effect of adaptive sampling strategy and state estimation algorithm, the multi-sensor time synchronization accuracy is significantly improved, the phase deviation problem caused by traditional fixed frequency sampling is effectively eliminated, the sliding window polynomial fitting combined with wavelet threshold denoising technology is used to significantly suppress high-frequency noise interference while preserving the effective components of the signal, and the dynamic threshold mechanism is used to improve the accuracy of outlier detection. The improved principal component analysis combined with the time alignment algorithm significantly improves the analysis efficiency of multi-dimensional feature data, and the quantitative feature correlation enhances the key indicator extraction capability. The cascaded model of deep belief network and support vector machine achieves higher recognition accuracy in the pollutant classification task, and the nonlinear compensation mechanism optimizes the accuracy of water quality health evaluation. The long short-term memory network integrates the change point detection technology to shorten the warning response time of water quality sudden change events, discover potential risks in advance, and improve the pollution traceability efficiency through multi-dimensional visualization technology. Combined with the optimized sensor deployment strategy, the monitoring network coverage is expanded and the operation and maintenance cost is reduced. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 The system block diagram of the present application; Figure 2 The multi-source perception module schematic diagram of the present application; Figure 3 The data purification module schematic diagram of the present application; Figure 4 The feature analysis module schematic diagram of the present application; Figure 5 The intelligent diagnosis module schematic diagram of the present application; Figure 6 The dynamic warning module schematic diagram of the present application; Figure 7 The visualization module schematic diagram of the present application; Figure 8 The self-maintenance module schematic diagram of the present application. DETAILED DESCRIPTION

[0015] The following will be described in detail in conjunction with the accompanying Figures 1-8 The present application will be further described in detail.

[0016] Reference is made to Figure 1The utility model relates to a water quality real-time detection system characterized by comprising the following modules: A multi-source perception module based on a multi-parameter sensing array adopts an adaptive sampling strategy, realizes sensor timing synchronization through a state estimation algorithm, combines a microfluidic chip to complete water body multi-dimensional parameter acquisition, and generates a multi-modal sensing data set; The multi-source perception module comprises a multi-source sensing sub-module, a signal conditioning sub-module, a timing synchronization sub-module, and an anomaly capture sub-module; A data purification module based on the multi-modal sensing data set adopts a sliding window polynomial fitting to eliminate baseline drift, uses a wavelet threshold denoising algorithm to suppress high-frequency interference, and uses a Mahalanobis distance detection method to eliminate outliers, thereby generating a purified feature matrix; The data purification module comprises a signal enhancement sub-module, a noise suppression sub-module, an anomaly cleaning sub-module, and a data alignment sub-module; A feature analysis module based on the purified feature matrix adopts an improved principal component analysis for dimension compression, realizes multi-source feature alignment through dynamic time warping, extracts key water quality indicators using a mutual information entropy algorithm, and generates a feature fingerprint spectrum; The feature analysis module comprises a dimension compression sub-module, a timing alignment sub-module, a feature optimization sub-module, and a spectrum construction sub-module; An intelligent diagnosis module based on the feature fingerprint spectrum adopts a deep belief network for pollutant identification, uses a support vector machine to realize water quality grading, combines a fuzzy reasoning system to complete comprehensive health degree evaluation, and generates a water quality diagnosis report; The intelligent diagnosis module comprises a pollutant identification sub-module, a grade determination sub-module, a health evaluation sub-module, and a report generation sub-module; A dynamic early warning module based on the water quality diagnosis report adopts a long short-term memory network for trend prediction, identifies mutation signals through Bayesian change point detection, triggers graded alarms using a multi-level threshold determination method, and generates a risk early warning instruction; The dynamic early warning module comprises a trend prediction sub-module, a mutation detection sub-module, a threshold determination sub-module, and an instruction issuing sub-module; A visualization module based on the risk early warning instruction adopts a WebGL rendering engine to construct a three-dimensional water quality map, displays pollutant migration paths through a dynamic Sankey diagram, realizes immersive monitoring using virtual reality technology, and generates a decision support dashboard; The visualization module comprises a space modeling sub-module, a path tracking sub-module, a situation deduction sub-module, and a dashboard generation sub-module; A self-maintenance module based on the decision support dashboard adopts a reinforcement learning algorithm to optimize sensor deployment, uses a digital twin model to predict equipment health degree, combines a self-organizing map to realize fault self-diagnosis, and generates a system health profile; The self-maintenance module comprises a deployment optimization submodule, a fault prediction submodule, a self-repairing submodule, and an archive updating submodule.

[0017] Through the synergy of adaptive sampling strategy and state estimation algorithm, the multi-sensor time synchronization accuracy is significantly improved, the phase deviation problem caused by traditional fixed frequency sampling is effectively eliminated, the sliding window polynomial fitting combined with wavelet threshold denoising technology is adopted to significantly suppress high-frequency noise interference while preserving the effective components of the signal, the dynamic threshold mechanism is used to improve the accuracy of outlier detection, the improved principal component analysis combined with time alignment algorithm significantly improves the analysis efficiency of multi-dimensional feature data, and the key indicator extraction capability is enhanced through quantitative feature correlation, the cascaded model of deep belief network and support vector machine realizes higher recognition accuracy in the pollutant classification task, the nonlinear compensation mechanism is combined to optimize the accuracy of water quality health evaluation, the long short-term memory network integrates the change point detection technology to shorten the warning response time of water quality sudden change events, and potential risks are found in advance, the multi-dimensional visualization technology improves the pollution tracing efficiency, and the optimized sensor deployment strategy expands the monitoring network coverage and reduces the operation and maintenance cost.

[0018] Referring to Figure 2 , a multi-source sensing submodule based on a microelectrode array adopts adaptive pulse width modulation to collect pH / ORP parameters, eliminates polarization effect through three-electrode differential detection method, and generates an original electrochemical signal set; Based on a microelectrode array, in an industrial wastewater treatment scene, adaptive pulse width modulation dynamically adjusts the sampling frequency according to the solution conductivity, automatically shortens the pulse interval to improve the sampling density when detecting pH value fluctuation, three-electrode differential detection method obtains the reference potential through the reference electrode, applies a step scan voltage between the working electrode and the counter electrode, and collects the redox current difference, for example, in the detection of chromium-containing wastewater, the working electrode surface is modified with nano gold particles to enhance the electron transfer efficiency, the counter electrode adopts a platinum mesh structure to expand the reaction area, the reference electrode selects the Ag / AgCl system to ensure the potential stability, and the median value is obtained through three times of differential measurement to eliminate accidental interference, and finally the multi-period sampling data is integrated to form the original electrochemical signal set.

[0019] A signal conditioning submodule based on the original electrochemical signal set adopts an instrument amplifier circuit for signal amplification, suppresses high-frequency noise through a Butterworth low-pass filter, and generates a preprocessed analog signal; When the raw electrochemical signal set is input to the instrumentation amplifier, the gain level is selected according to the signal amplitude range. The weak current signal is converted into a voltage signal by the transimpedance amplifier. The Butterworth low-pass filter is set to the cutoff frequency to filter out high-frequency noise. In the wastewater treatment plant application, a fourth-order filter is configured to attenuate the 50Hz power frequency interference caused by pump vibration. The signal conditioning stage adopts a segmented gain control strategy. The front stage fixed gain amplification ensures the signal-to-noise ratio, and the rear stage programmable gain adapts to different ranges. For example, a high-gain mode is enabled when detecting low concentration ammonia nitrogen, and a low-gain state is switched when detecting high concentration COD. After multi-stage conditioning, a smooth pre-processed analog signal is output.

[0020] The timing synchronization submodule, based on preprocessed analog signals, uses a state estimation algorithm to achieve multi-sensor clock synchronization, combines a sliding time window to complete data alignment, and generates timing synchronization data packets. After preprocessing the analog signal input, the state estimation algorithm establishes a sensor clock deviation model, compensates for transmission delay through timestamp alignment, and uses a sliding time window to truncate data segments at fixed lengths. In the river cross-section monitoring scenario, a dynamic window adjustment mechanism is adopted to address the time sequence misalignment caused by flow velocity differences. When the phase difference between the turbidity sensor and the conductivity sensor data exceeds a set threshold, the window length is automatically expanded for interpolation alignment. Time stamp information is retained during data reconstruction. In the event of sudden data loss, a linear prediction algorithm is used to fill in the missing points, ultimately generating a time-series synchronization data packet with precise time dimension matching.

[0021] The anomaly capture submodule, based on time-series synchronous data packets, uses the 3σ criterion to detect sudden anomalies, identifies gradual drifts using the sliding t-test, and generates a multimodal sensing dataset.

[0022] After the time-series synchronization data packet is input, the 3σ criterion calculates the mean and standard deviation based on a sliding window. When the dissolved oxygen reading exceeds three times the standard deviation for three consecutive times, it is marked as a sudden anomaly. The sliding t-test method is set up with a double window to compare the differences in data distribution. In the monitoring of lake eutrophication, for the gradual trend of chlorophyll concentration, a 300-second window is used to calculate the difference in the mean before and after the period. A confidence judgment threshold is set to identify slow drift. A multi-level verification mechanism is implemented in the anomaly judgment stage. After the initial screening, cross-validation is performed through adjacent sensor data to eliminate false alarms due to single-point faults. Finally, the spatiotemporal dimension anomaly features are integrated to form a multimodal sensor dataset.

[0023] See Figure 3 The signal enhancement submodule, based on a multimodal sensor dataset, reconstructs the signal baseline using a sliding window polynomial fitting, improves the signal-to-noise ratio through adaptive gain control, and generates a baseline correction signal. Based on the multi-modal sensor data set, in the industrial wastewater monitoring scene, the sliding window polynomial fitting adopts the piecewise approximation strategy, dynamically adjusts the window length according to the fluctuation characteristics of the water body parameters, automatically reduces the window when the conductivity changes rapidly to improve the local fitting precision, and the adaptive gain control module calculates the signal fluctuation amplitude in real time, increases the gain coefficient when the heavy metal ion concentration suddenly changes to strengthen feature extraction, for example, when processing copper-containing wastewater, the redox potential signal is fitted with a cubic polynomial to eliminate the baseline drift caused by electrode aging, and the amplification multiple is dynamically adjusted according to the signal peak-valley difference, and finally the corrected signal with stable baseline is output.

[0024] The noise suppression submodule is based on the baseline corrected signal, and uses a wavelet threshold denoising algorithm to decompose high-frequency components, and uses an improved threshold function to retain effective features to generate a denoised feature vector. After the baseline correction signal is input, the wavelet threshold denoising selects an appropriate basis function to decompose the signal level, and uses a soft threshold function to process the detail coefficients for high-frequency noise generated by the water pump, and the improved threshold function introduces a scale adaptive mechanism. In the application of sewage treatment plants, the threshold proportion coefficient is dynamically adjusted according to the noise energy distribution, for example, after the turbidity signal is decomposed by five layers of wavelets, the high-frequency coefficients above the third layer are nonlinearly compressed to retain the effective fluctuation characteristics reflecting the suspended solids concentration, and the denoised signal vector retaining key features is generated by multi-scale reconstruction to eliminate random spike interference.

[0025] The anomaly cleaning submodule is based on the denoised feature vector, and uses the Mahalanobis distance detection method to identify outliers, and uses the KNN interpolation algorithm to repair missing data to generate a cleaned data set. After the denoised feature vector is input, the Mahalanobis distance detection establishes a multi-dimensional feature space distribution model to calculate the deviation of each data point from the cluster center. In the monitoring of lake eutrophication, a dynamic threshold boundary is set for the combined distribution of multiple parameters such as chlorophyll and dissolved oxygen. When the combined features of ammonia nitrogen concentration and pH value exceed the normal distribution range, it is marked as an outlier. The KNN interpolation selects historical data from adjacent monitoring points in the same watershed to repair missing values, for example, when the dissolved oxygen data of a monitoring point is missing, the weighted average of the data of three upstream and downstream stations at the same time is taken to fill in the missing values, generating a time and space continuous data set.

[0026] The data alignment submodule is based on the cleaned data set, and uses dynamic time warping to align multi-source time series, and verifies the data consistency by combining the Pearson correlation coefficient to generate a purified feature matrix.

[0027] After the data set is input after cleaning, the dynamic time warping algorithm constructs a multi-source data alignment path matrix. In the river section monitoring scene, in view of the phase difference of conductivity and turbidity signals caused by flow rate difference, the optimal matching path is found by minimizing the bending path cost, the correlation coefficient is verified in the Pearson correlation coefficient verification stage, and when the correlation coefficient of the upstream and downstream pH value sequences is lower than the set standard, the data review mechanism is triggered, for example, the dissolved oxygen sequence of two monitoring stations in a river section is aligned after normalization, the average of the correlation coefficient in the sliding window is calculated, and when the threshold is lower than the threshold, the redundant sensor data is automatically started for cross verification, and finally the feature matrix meeting the spatial and temporal consistency is output.

[0028] Referring to Figure 4 , the dimension compression submodule, based on the purified feature matrix, uses an improved principal component analysis for feature dimension reduction, retains feature information by the variance contribution rate threshold method, and generates a reduced feature set; Based on the purified feature matrix, in the industrial wastewater multi-parameter monitoring scene, the improved principal component analysis introduces a feature weight adjustment mechanism, dynamically allocates load coefficients according to the correlation between parameters, and when processing wastewater containing heavy metals, higher weights are given to strongly correlated parameters such as conductivity and oxidation-reduction potential. The threshold value of the variance contribution rate is set to consider the monitoring target demand, and a higher threshold value is set in drinking water source monitoring to retain more safety-related features. The inflection point of the cumulative contribution rate curve is used to determine the dimension reduction amplitude, for example, when processing twelve-dimensional sensor data, the first four principal components are selected to cover the main information variation, and finally a low-dimensional data set retaining key pollution features is output.

[0029] The time series alignment submodule, based on the reduced feature set, uses dynamic time warping to correct the phase difference of multi-source data, and combines cosine similarity to verify the alignment effect to generate a time series alignment matrix; After the reduced feature set is input, the dynamic time warping algorithm constructs an optimal matching path for multi-source data, in the river basin monitoring scene, for the flow rate difference between upstream and downstream stations, the minimum bending cost path is found by using the cumulative distance matrix, the sliding window local alignment strategy is used in the phase correction stage, and the time delay difference between turbidity and dissolved oxygen signals is compensated in segments. The cosine similarity verification sets a dynamic threshold, when the similarity of the pH sequences of the two monitoring points at the river section is lower than the standard after alignment, the interpolation algorithm is automatically started to smooth the transition, for example, the conductivity sequence at the confluence of a tributary into the main stream is stretched and compressed three times locally, and the time series synchronization with the main monitoring station data is achieved.

[0030] The feature optimization submodule, based on the time series alignment matrix, uses the mutual information entropy algorithm to calculate the feature importance, selects the top 10 key indicators by the forward search strategy, and generates an optimized feature vector; After the time alignment matrix is input, the mutual information entropy algorithm calculates the correlation strength of the features and the pollution categories. In the chemical industry park drainage monitoring, the mutual information values of parameters such as COD and ammonia nitrogen and the pollution types are sorted, and a forward search strategy adopts a stepwise regression method to add the feature that improves the model performance most in each iteration. For example, when screening the key indicators of heavy metal pollution, the redox potential feature with strong correlation with lead and chromium ion concentration is preferentially retained. Through five rounds of iteration, the optimal feature subset is determined, the classification accuracy is re-evaluated after removing redundant parameters each time, and the final core parameter combination with the highest impact weight is locked.

[0031] The atlas construction submodule constructs a three-dimensional feature space based on the preferred feature vector using radial basis function interpolation, and generates a visual atlas through an isosurface extraction algorithm to generate a feature fingerprint atlas.

[0032] After the preferred feature vector is input, a radial basis function interpolation selects a Gaussian kernel function to construct a three-dimensional spatial distribution model. In the monitoring of lake eutrophication, a continuous distribution surface is generated according to the spatial sampling points of parameters such as chlorophyll and total phosphorus. The isosurface extraction adopts a moving cube algorithm, and a concentration gradient threshold is set to divide different pollution areas. For example, in a water area where algae has exploded, the chlorophyll concentration gradient change rate is used as the basis for generating an isosurface, and an interactive three-dimensional thermal atlas is generated through triangular meshing processing, and finally a feature fingerprint model reflecting the spatial distribution of pollution is formed.

[0033] Referring to Figure 5 , the pollutant identification submodule uses a deep belief network to learn features based on the feature fingerprint atlas, and realizes pollutant classification through a stack of restricted Boltzmann machines to generate a pollutant classification result. Based on the feature fingerprint atlas, in the monitoring of electroplating wastewater, a deep belief network constructs a multi-layer feature abstraction structure. The first layer of restricted Boltzmann machine extracts the spatial distribution pattern of spectral features, and the second layer captures the concentration change rule in the time dimension. For the identification task of hexavalent chromium pollution, the input layer receives the ultraviolet-visible spectrum feature vector, and the hidden layer nodes update the weight parameters through the contrastive divergence algorithm. In the pre-training stage, a layer-by-layer greedy strategy is used to initialize the network. For example, when processing nickel-containing wastewater, the absorption peak shape in the wavelength interval of 300-400 nm in the feature atlas is used to adjust the hidden layer node activation threshold, and the classification boundary is optimized in the fine-tuning stage combined with label data, and finally the heavy metal pollutant type code is output.

[0034] The grade determination submodule uses a support vector machine to build a multi-classification model based on the pollutant classification result, realizes five-level division of water quality through a one-versus-one strategy, and generates a water quality grade label. After the pollution classification result is input, the support vector machine uses a Gaussian kernel function to map a feature space, in drinking water source monitoring, ten binary classifiers are constructed for five-level classification of water quality, each classifier focuses on the decision boundary between two adjacent levels, for the problem of dividing Ⅲ and Ⅳ water, dissolved oxygen and permanganate index are selected as key discriminant features, the width parameter of the kernel function is adaptively adjusted according to the distance between feature vectors, in the model training stage, the grid search method is used to find the optimal penalty coefficient, for example, when processing lake eutrophication data, total phosphorus and chlorophyll concentration are standardized and input into the classifier, and the final level label is determined through the majority voting mechanism.

[0035] The health assessment submodule, based on the water quality level label, uses a fuzzy reasoning system to construct an evaluation rule base, calculates a comprehensive health index through a defuzzification algorithm, and generates a health degree index; After the water quality level label is input, the fuzzy reasoning system defines the membership function of health degree evaluation, in river health assessment, dissolved oxygen, ammonia nitrogen and other parameters are divided into "excellent", "medium" and "poor" fuzzy sets, the rule base includes twenty IF-THEN type judgment logics, for example, when the pH value belongs to the "normal" set and the COD belongs to the "low pollution" set, the health degree improvement rule is activated, the barycenter method is used to calculate the accurate output value in the defuzzification stage, for the reservoir water quality evaluation task, the weight coefficients of each parameter are set to reflect its influence on the ecology, the weighted average fusion of the multi-rule output value is used to generate a continuous health index.

[0036] The report generation submodule, based on the health degree index, uses template engine technology to integrate multi-dimensional data, automatically generates a diagnosis conclusion through natural language generation, and generates a water quality diagnosis report.

[0037] After the health degree index is input, the template engine parses the XML format report structure, in the industrial wastewater monitoring report, the preset conclusion paragraph includes pollution type, level distribution and trend prediction three modules, the natural language generator selects the preset corpus according to the health degree value range, when the index is in the critical interval, multiple description segments are automatically combined to form a coherent conclusion, for example, when the data of the drainage outlet of a chemical plant triggers a heavy metal overlimit warning, the system selects keywords such as "chromium pollution", "continuous deterioration" and "suggestion for emergency treatment", and generates a professional diagnosis report text conforming to environmental protection specifications through syntax tree reorganization.

[0038] Referring to Figure 6 , the trend prediction submodule, based on the water quality diagnosis report, uses a long short-term memory network to construct a time series prediction model, captures long-term dependencies through an attention mechanism, and generates a trend prediction curve; Based on the water quality diagnosis report, in the river pollution prediction scenario, the long and short time memory network constructs the cycle unit structure containing the forgetting gate, the input gate and the output gate, the time sequence window is slid to extract the change mode of the dissolved oxygen, COD and other parameters, the attention mechanism dynamically allocates the weight of the historical data, when predicting the trend of ammonia nitrogen concentration, the recent mutation data is given higher attention, the sliding window cross validation strategy is adopted in the network training stage, for example, when processing the quarterly monitoring data of a basin, the first three months of data are used as the training set, and the fourth month of data is used to verify the prediction accuracy of the model, the information flow is adjusted through the gate mechanism, and finally the water quality parameter change curve in the next seven days is output.

[0039] The mutation detection submodule detects the mutation of the data distribution based on the trend prediction curve, identifies the risk event through the posterior probability threshold method, and generates a risk event list; After the trend prediction curve is input, the Bayesian change point detection constructs a probability model to evaluate the difference of data distribution. In the drainage monitoring of the chemical industry park, the posterior probability ratio of the mean value of the heavy metal concentration in the adjacent time window is calculated, and a dynamic threshold is set to trigger the mutation alarm. When the lead ion concentration sequence in a period appears continuous deviation, the mutation point probability is calculated by Markov chain Monte Carlo method, for example, if the cumulative probability of the concentration of hexavalent chromium in four hours exceeds the set standard, it is automatically marked as a high-risk event. After being verified for three consecutive monitoring periods, a risk event list containing time stamp, pollution type and change amplitude is generated.

[0040] The threshold determination submodule sets three-level alarm rules based on the risk event list by using a multi-level threshold determination method, manages the alarm life cycle through a state machine model, and generates a warning level signal; After the risk event list is input, the multi-level threshold determination sets yellow, orange and red three-level alarm rules. In the drinking water source monitoring, the trigger condition is set according to the combination of pollution type and concentration amplitude, the state machine defines four states of alarm activation, confirmation, upgrade and release, when the ammonia nitrogen concentration exceeds the baseline value for two hours and reaches the orange threshold, the system jumps from standby state to confirmation state, and after the on-duty personnel feedback, it enters the active state, for example, when the total phosphorus in a reservoir abnormally rises to trigger the red alarm, the emergency monitoring mode is automatically started and the alarm state is locked until manual intervention is released.

[0041] The instruction issuing submodule pushes the alarm information based on the warning level signal by using the publish-subscribe mode, realizes the linkage of Internet of Things devices through the MQTT protocol, and generates risk warning instructions.

[0042] After the early warning level signal is input, the publish-subscribe mode distributes the alarm information through the topic routing. In the application in the sewage treatment plant, the MQTT protocol defines three types of topics of device control, data collection and state feedback. When the red early warning of heavy metal exceeding the standard is triggered, the valve closing instruction is pushed to the "emergency device control" topic, and a short message early warning is sent to the "manager notification" topic. After the Internet of Things gateway receives the instruction, the water inlet valve is closed and the emergency pool pump is started. For example, when the chromium of the drainage outlet of a certain electroplating plant exceeds the standard, the system completes the whole process response from data analysis to device linkage within five seconds.

[0043] Referring to Figure 7 , a spatial modeling submodule, based on the risk early warning instruction, uses a WebGL rendering engine to build a three-dimensional river model, generates a water quality distribution surface through Gaussian process interpolation, and generates a water quality spatial model; Based on the risk early warning instruction, in the watershed pollution emergency scene, the WebGL rendering engine loads the digital elevation model to build the three-dimensional grid of the river, restores the real topography and geomorphology by combining satellite remote sensing data mapping, and uses radial basis function to calculate the spatial continuous distribution according to the dissolved oxygen concentration data of the discrete monitoring points, for example, in the processing of a benzene series leakage event of a tributary, the pollutant concentration isosurface is generated based on the data of ten monitoring stations upstream and downstream, the interpolation accuracy is optimized through Kriging variance, and the terrain fluctuation parameters and water flow velocity data jointly constrain the interpolation process, and finally a rotatable and scalable three-dimensional water quality spatial model is formed.

[0044] A path tracking submodule, based on the water quality spatial model, uses a dynamic Sankey diagram to visualize the pollutant migration, simulates the diffusion path through a particle tracking algorithm, and generates a migration path atlas; After the water quality spatial model is input, the dynamic Sankey diagram constructs the flow relationship network of the pollution source and the receiving water body, and the particle tracking algorithm simulates the pollutant migration trajectory in combination with the flow velocity field data. In the accident simulation of a chemical industry park, the virtual particle mass corresponding to the total amount of leaked substances is set, the diffusion direction parameters are adjusted according to the real-time wind speed and direction, for example, in the simulation of ammonia water tank leakage, a batch of particles representing the pollution group moving front is generated every five minutes, the concentration attenuation process is mapped through color gradient, and the trajectory data is smoothed to generate a visual path atlas containing migration direction and diffusion range.

[0045] A situation deduction submodule, based on the migration path atlas, uses virtual reality technology to build an immersive scene, predicts the pollution diffusion trend through a fluid dynamics model, and generates a situation deduction animation; After the migration path map is input, the virtual reality system constructs a panoramic three-dimensional scene of the river basin, a fluid dynamics model solves the Navier-Stokes equation to simulate water flow movement, in the prediction of thermal plume diffusion of the nuclear power plant, the temperature diffusion coefficient and the heat capacity parameter of the water body are set, and the time stepping algorithm is used to calculate the diffusion form of the thermal plume, for example, when simulating thermal pollution at a tidal estuary, the boundary conditions are dynamically adjusted according to the rising and falling tide cycle data, new flow field data is generated every half hour to drive particle movement, the immersion observation of the advancing speed and the influence range of the pollution front is realized through the head-mounted device, and the deduction animation containing the evolution process in the next twelve hours is output.

[0046] The dashboard generation submodule integrates multiple source views based on the situation deduction animation using data dashboard technology, adapts to different terminal devices through responsive layout, and generates decision support dashboards.

[0047] After the situation deduction animation is input, the data dashboard integrates multiple view elements such as three-dimensional models, trend curves, and early warning maps, and the responsive layout defines the component arrangement rules under different screen sizes. In the mobile monitoring platform, the map component automatically switches to full-screen mode, and the curve chart is converted to vertical sliding view. For example, when handling a cross-administrative district river basin pollution event, the dashboard synchronously displays provincial boundary section data comparison, emergency resource distribution, real-time video monitoring, etc. Through CSS media query, the terminal such as mobile phone, tablet, command large screen, etc. is adapted, and finally the adaptive decision-making dashboard that maintains information integrity is generated.

[0048] Referring to Figure 8 , the deployment optimization submodule is based on the decision support dashboard and uses deep reinforcement learning to build an optimization model. The optimal sensor layout is calculated through the Q-learning algorithm, and the optimized deployment scheme is generated. Based on the decision support dashboard, in the optimization of the monitoring network in the chemical industry park, the state space is constructed by deep reinforcement learning, including sensor coverage, energy consumption cost, data transmission quality, etc. The Q-learning algorithm defines the deployment position adjustment as the action set, and the reward function integrates the monitoring blind area reduction rate and the deployment cost change amount. In the river section monitoring scene, the agent learns the optimal layout strategy through iterative exploration, for example, in the deployment optimization of a certain river basin, ten sensors are randomly arranged initially, and after one hundred iterations, a topology structure that considers main road coverage and branch monitoring points is formed. The monitoring efficiency improvement rate and the operation and maintenance cost change rate of the new layout are calculated after each adjustment, and finally the sensor coordinate set that meets the multi-objective constraints is output.

[0049] The fault prediction submodule is based on the optimized deployment scheme and uses digital twin technology to construct a device mirror. The remaining useful life is predicted through the Weibull distribution model, and a fault prediction report is generated. After optimizing the deployment scheme input, the digital twin model synchronizes the working state parameters of the physical sensor, and the Weibull distribution model fits the shape parameter and characteristic life parameter according to the historical failure data. In the cooling water monitoring system of the nuclear power plant, the degradation characteristics such as the vibration amplitude of the pressure sensor and the temperature drift are collected, and the cumulative failure probability curve is calculated. For example, in the life prediction of a certain pH electrode, based on the calibration deviation data of the past three hundred days, the failure probability distribution of the accelerated aging stage is fitted. When the predicted remaining life is lower than the safety threshold, a preventive maintenance reminder is triggered, and a failure warning document containing the component number, predicted failure time and recommended measures is generated.

[0050] The self-repairing sub-module identifies the failure mode based on the failure prediction report using self-organizing mapping, generates a repair strategy set by fault tree analysis; After the failure prediction report is input, the self-organizing mapping network projects the failure features onto a two-dimensional diagnostic plane. In the monitoring of the pump station in the sewage treatment plant, a failure feature vector containing parameters such as current fluctuation, bearing temperature, and vibration spectrum is constructed, and a failure mode cluster is formed through competitive learning. The fault tree analysis establishes the logical relationship between the leaf node failure and the root cause. For example, in the case of an overload failure of a sludge pump, it is traced back to multiple factors such as mechanical seal wear and motor insulation aging, and a repair strategy combination containing spare parts replacement, parameter adjustment, and cleaning and maintenance is generated. The strategy priority is dynamically sorted according to the failure impact degree and repair cost.

[0051] The archive updating sub-module records maintenance logs based on the repair strategy set using blockchain technology, and updates the device lifecycle data through a time series database to generate a system health archive.

[0052] After the repair strategy set is input, the blockchain network uses a smart contract to record the timestamp, performer, and operation content of the maintenance operation. The time series database stores multi-dimensional data such as sensor readings, device status codes, and environmental parameters with a minute granularity. In the monitoring system of the oil refinery, each valve maintenance record generates a blockchain transaction containing the operation video hash value, spare parts batch number, and detection report. The device lifecycle data is stored in chronological order, supporting multi-dimensional retrieval by device number, time range, and failure type. Finally, a system health archive library is formed that is tamper-proof and supports efficient querying.

[0053] The above are preferred embodiments of the present application, which do not limit the protection scope of the present application. Therefore, any equivalent changes made in accordance with the structure, shape, and principle of the present application should be covered within the protection scope of the present application.

Claims

1. A real-time water quality monitoring system, characterized in that... Includes the following modules: The multi-source sensing module, based on a multi-parameter sensing array, adopts an adaptive sampling strategy and uses a state estimation algorithm to achieve sensor timing synchronization. Combined with a microfluidic chip, it completes the acquisition of multi-dimensional parameters of water bodies and generates a multi-modal sensing dataset. The multi-source sensing module includes a multi-source sensing submodule, a signal conditioning submodule, a timing synchronization submodule, and an anomaly capture submodule; The data cleansing module, based on a multimodal sensor dataset, uses a sliding window polynomial fitting to eliminate baseline drift, suppresses high-frequency interference through a wavelet threshold denoising algorithm, removes outliers using Mahalanobis distance detection, and generates a cleansing feature matrix. The data cleansing module includes a signal enhancement submodule, a noise suppression submodule, an anomaly cleaning submodule, and a data alignment submodule; The feature parsing module, based on the purification feature matrix, uses improved principal component analysis for dimensionality compression, achieves multi-source feature alignment through dynamic time warping, and extracts key water quality indicators by combining mutual information entropy algorithm to generate feature fingerprint spectrum; The feature parsing module includes a dimensionality compression submodule, a temporal alignment submodule, a feature optimization submodule, and a graph construction submodule; The intelligent diagnostic module uses a feature fingerprint spectrum, a deep belief network to identify pollutants, a support vector machine to classify water quality, and a fuzzy inference system to complete a comprehensive health assessment and generate a water quality diagnostic report. The intelligent diagnostic module includes a pollutant identification submodule, a level determination submodule, a health assessment submodule, and a report generation submodule; The dynamic early warning module, based on water quality diagnostic reports, uses a long short-term memory network for trend prediction, identifies abrupt change signals through Bayesian change point detection, and triggers graded alarms by combining a multi-level threshold judgment method to generate risk warning instructions. The dynamic early warning module includes a trend prediction submodule, a mutation detection submodule, a threshold determination submodule, and an instruction issuance submodule; The visualization module, based on risk warning instructions, uses the WebGL rendering engine to build a 3D water quality map, displays the migration path of pollutants through a dynamic Sankey diagram, and combines virtual reality technology to achieve immersive monitoring and generate decision support dashboards. The visualization module includes a spatial modeling submodule, a path tracing submodule, a situational simulation submodule, and a dashboard generation submodule; The self-maintenance module, based on a decision support dashboard, uses reinforcement learning algorithms to optimize sensor deployment, predicts equipment health through a digital twin model, and combines self-organizing mapping to achieve fault self-diagnosis and generate system health records. The self-maintenance module includes a deployment optimization submodule, a fault prediction submodule, a self-repair submodule, and a file update submodule.

2. The real-time water quality monitoring system according to claim 1, characterized in that: The multi-source sensing submodule, based on a microelectrode array, uses adaptive pulse width modulation to acquire pH / ORP parameters and eliminates polarization effects through a three-electrode differential detection method to generate the original electrochemical signal set. The signal conditioning submodule, based on the original electrochemical signal set, uses an instrumentation amplifier circuit to amplify the signal, suppresses high-frequency noise through a Butterworth low-pass filter, and generates a pre-processed analog signal. The timing synchronization submodule, based on preprocessed analog signals, uses a state estimation algorithm to achieve multi-sensor clock synchronization, combines a sliding time window to complete data alignment, and generates timing synchronization data packets. The anomaly capture submodule, based on time-series synchronous data packets, uses the 3σ criterion to detect sudden anomalies, identifies gradual drifts using the sliding t-test, and generates a multimodal sensing dataset.

3. The real-time water quality monitoring system according to claim 1, characterized in that: The signal enhancement submodule, based on a multimodal sensor dataset, reconstructs the signal baseline using a sliding window polynomial fitting, improves the signal-to-noise ratio through adaptive gain control, and generates a baseline correction signal. The noise suppression submodule, based on the baseline correction signal, uses a wavelet threshold denoising algorithm to decompose high-frequency components, retains effective features through an improved threshold function, and generates a denoised feature vector; The anomaly cleaning submodule identifies outliers based on denoised feature vectors, uses Mahalanobis distance detection to identify outliers, repairs missing data using the KNN interpolation algorithm, and generates a cleaned dataset. The data alignment submodule, based on the cleaned dataset, uses dynamic time warping to align multi-source time series, combines Pearson correlation coefficient to verify data consistency, and generates a cleaned feature matrix.

4. The real-time water quality monitoring system according to claim 1, characterized in that: The dimension compression submodule, based on the purified feature matrix, uses improved principal component analysis to perform feature dimensionality reduction, and retains feature information through the variance contribution rate threshold method to generate a dimensionality-reduced feature set; The temporal alignment submodule, based on a dimensionality-reduced feature set, uses dynamic time warping to correct the phase difference of multi-source data, combines cosine similarity to verify the alignment effect, and generates a temporal alignment matrix. The feature optimization submodule calculates feature importance based on the time alignment matrix and the mutual information entropy algorithm. It then uses a forward search strategy to filter the top 10 key indicators and generate the optimal feature vector. The map construction submodule constructs a three-dimensional feature space based on the optimized feature vector and uses radial basis function interpolation. It then generates a visual map and a feature fingerprint map through the isosurface extraction algorithm.

5. The real-time water quality monitoring system according to claim 1, characterized in that: The pollutant identification submodule, based on the feature fingerprint spectrum, uses a deep belief network for feature learning, and achieves pollutant classification through stacking restricted Boltzmann machines to generate pollutant classification results; The grade determination submodule, based on the pollutant classification results, uses a support vector machine to build a multi-classification model, and implements a five-level water quality classification through a one-to-one strategy to generate water quality grade labels; The health assessment submodule, based on water quality grade labels, uses a fuzzy inference system to build an assessment rule base, calculates a comprehensive health index through a defuzzification algorithm, and generates a health index. The report generation submodule, based on the health index, uses template engine technology to integrate multi-dimensional data and automatically generates diagnostic conclusions through natural language processing, producing a water quality diagnostic report.

6. The real-time water quality monitoring system according to claim 1, characterized in that: The trend prediction submodule, based on the water quality diagnostic report, uses a long short-term memory network to build a time-series prediction model, captures long-term dependencies through an attention mechanism, and generates a trend prediction curve. The mutation detection submodule, based on the trend prediction curve, uses Bayesian change point detection and analysis to identify mutations in data distribution, identifies risk events through the posterior probability threshold method, and generates a list of risk events. The threshold determination submodule, based on the risk event list, uses a multi-level threshold determination method to set three levels of alarm rules, manages the alarm lifecycle through a state machine model, and generates early warning level signals. The instruction issuance submodule pushes alarm information based on the warning level signal using a publish-subscribe model, and generates risk warning instructions by linking IoT devices through the MQTT protocol.

7. The real-time water quality monitoring system according to claim 1, characterized in that: The spatial modeling submodule, based on risk warning instructions, uses the WebGL rendering engine to construct a 3D river model and generates a water quality distribution surface through Gaussian process interpolation to generate a water quality spatial model. The path tracing submodule, based on the water quality spatial model, uses a dynamic Sankey diagram to visualize pollutant migration and simulates diffusion paths using a particle tracing algorithm to generate migration path maps. The situation simulation submodule, based on the migration path map, uses virtual reality technology to construct an immersive scene, predicts the pollution diffusion trend through a fluid dynamics model, and generates situation simulation animations. The Kanban generation submodule, based on situational simulation animation, integrates multi-source views using data Kanban technology and generates decision support Kanban boards by adapting to different terminal devices through responsive layout.

8. The real-time water quality monitoring system according to claim 1, characterized in that: The deployment optimization submodule, based on the decision support dashboard, uses deep reinforcement learning to build an optimization model and calculates the optimal sensor layout using the Q-learning algorithm to generate an optimized deployment scheme. The fault prediction submodule, based on the optimized deployment scheme, uses digital twin technology to build a device image, predicts the remaining service life through the Weibull distribution model, and generates a fault prediction report; The self-healing submodule, based on the fault prediction report, uses self-organizing mapping for fault mode identification, generates repair strategies through fault tree analysis, and generates a set of repair strategies. The file update submodule, based on the repair strategy set, uses blockchain technology to record maintenance logs, updates the device's full lifecycle data through a time-series database, and generates a system health file.

Images