Soil pollutant detection and analysis system based on intelligent sensor

Abstract

The invention relates to the technical field of environment monitoring and intelligent sensors, and particularly discloses a soil pollutant detection and analysis system based on an intelligent sensor. The system comprises a sensor array module, a data preprocessing and fusion module, an ecological toxicity dynamic evaluation module and a decision output module. According to the invention, the biological response sensing unit capable of directly reflecting the ecological effect is integrated, and the ecological toxicity dynamic evaluation model fusing the concentration, the biological response and the environmental factors is innovatively constructed, so that direct and dynamic quantification from the pollutant content to the ecological risk is realized. According to the method, the problem that an existing assessment method is insufficient in biological correlation is solved, the risk assessment result is closer to the real ecological environment effect, and a reliable basis is provided for precise treatment.

Description

Technical Field

[0001] This invention belongs to the field of environmental monitoring and intelligent sensor technology, specifically relating to a soil pollutant detection and analysis system based on intelligent sensors. Background Technology

[0002] Soil environmental monitoring and pollution control is an important branch of environmental protection. It uses technological means to identify, assess, and remediate contaminated soil to ensure ecological security and human health. This field involves the detection, analysis, and risk assessment of various pollutants, such as heavy metals and organic compounds.

[0003] Soil pollutant detection technology based on smart sensors enables real-time, in-situ monitoring of specific pollutant concentrations through a sensor network deployed in the soil. The basic principle of this technology is to use sensors to convert the chemical or physical signals of pollutants into measurable electrical signals, thereby obtaining their concentration information.

[0004] Current technologies primarily rely on sensors to directly measure pollutant concentrations. However, this single concentration data has several limitations: it only reflects the quantity of pollutants present and cannot directly reveal their biotoxic effects on soil microorganisms, plants, and the entire food chain. Different pollutants may exhibit drastically different ecological risks at the same concentration, and key information such as the bioavailability of pollutants and their complex pollution effects cannot be obtained through concentration indicators. This leads to risk assessments based on existing detection data often deviating from the actual situation, making it difficult to accurately guide subsequent pollution control decisions and ecological restoration efforts. Therefore, this constitutes a pressing technical challenge for precise soil pollution management. Summary of the Invention

[0005] The purpose of this invention is to provide a soil pollutant detection and analysis system based on intelligent sensors to solve the technical contradiction in the prior art that relies solely on pollutant concentration data, which cannot accurately assess its biotoxic effects and ecological risks, leading to biased risk assessment and inaccurate governance decisions.

[0006] This invention provides a soil pollutant detection and analysis system based on intelligent sensors, comprising: The sensor array module is used to collect multi-dimensional physicochemical and biological response signals of the soil environment in situ. The sensor array module consists of a concentration detection unit, a biological response sensing unit, and an environmental parameter monitoring unit. The data preprocessing and fusion module is used to standardize and clean, extract features and align the original heterogeneous data collected by the sensor array module, and generate a fused feature vector. The data preprocessing and fusion module includes a data cleaning unit, a feature extraction unit and a multi-source data fusion unit. The ecotoxicity dynamic assessment module is used to dynamically quantify the comprehensive ecotoxicity equivalent of pollutants based on the fused feature vector output by the data preprocessing and fusion module. The ecotoxicity dynamic assessment module has a built-in ecotoxicity calculation engine and a dynamic weight adjustment engine. The decision output module is used to map the comprehensive ecotoxicity equivalent output by the ecotoxicity dynamic assessment module into specific risk assessment levels and governance priority recommendations. The decision output module includes a level mapping unit and a strategy generation unit.

[0007] Preferably, the concentration detection unit is equipped with an electrochemical sensor and a spectroscopic sensor for directly measuring the concentration of heavy metal ions and specific organic pollutants in the soil; The bio-response sensing unit integrates a microbial fuel cell sensor and a plant physiological electrical signal sensor to capture changes in the metabolic activity of soil microbial communities and to indicate stress electrophysiological signals of plant roots. The environmental parameter monitoring unit is equipped with a soil moisture sensor, a soil temperature sensor, and a soil pH sensor to simultaneously monitor key environmental factors that affect the migration, transformation, and bioavailability of pollutants.

[0008] Preferably, the data cleaning unit uses a sliding window-based outlier detection algorithm and an interpolation algorithm to process noise and missing values ​​in the original sensor signal. The feature extraction unit performs time-domain and frequency-domain analysis on the cleaned time-series signal to extract statistical and spectral features, including mean, variance, peak value, and main frequency component. The multi-source data fusion unit concatenates the feature vectors from the concentration detection unit, the bioresponse sensing unit, and the environmental parameter monitoring unit under the same spatiotemporal coordinates, and performs dimensionality reduction using principal component analysis to output a low-dimensional fused feature vector representing the overall state of the current soil location.

[0009] Preferably, the ecotoxicity calculation engine receives a fusion feature vector from the data preprocessing and fusion module as input, queries the built-in pollutant basic toxicity database based on the concentration data to calculate the theoretical toxicity equivalent based on the concentration, further introduces a biological response correction factor and an environmental factor correction coefficient, and executes the following calculation logic: the comprehensive ecotoxicity equivalent is equal to the product of the theoretical toxicity equivalent and the biological response correction factor, and then multiplied by the environmental factor correction coefficient. The dynamic weight adjustment engine is used to adaptively adjust the contribution weights of concentration data, biological response data, and environmental data in the fused feature vector based on pollutant type and pollution history. The dynamic weight adjustment engine has a built-in rule base and a lightweight neural network.

[0010] Preferably, the level mapping unit compares the continuous comprehensive ecotoxicity equivalent values ​​with multiple preset toxicity threshold intervals and automatically classifies them into corresponding risk levels. The strategy generation unit associates with the governance measures knowledge base, and based on the risk level of the current location, pollutant identification results, and soil environmental parameters, matches and outputs the optimal governance priority number and preliminary technical solution suggestions from the knowledge base.

[0011] Preferably, the calculation process of the biological response correction factor is as follows: The inhibition rate of microbial metabolic activity and the intensity of plant stress signals collected by the bioresponse sensing unit are compared and mapped with a preset dose-response curve to obtain the bioresponse correction factor.

[0012] Preferably, the calculation process for the environmental factor correction coefficient is as follows: The measured values ​​of soil moisture, temperature, and pH are input into a multivariate nonlinear regression model trained based on a large amount of historical data. This multivariate nonlinear regression model describes the influence of environmental conditions on the bioavailability of pollutants. The environmental factor correction coefficients are calculated through the model. The training process of the multivariate nonlinear regression model is as follows: collect historical datasets covering different soil types, different pollutant concentration gradients, and different combinations of environmental conditions. The datasets include environmental parameters, final bioaccumulation of pollutants, or ecological effect endpoint data. Using the aforementioned environmental parameters as input features and the normalized values ​​of bioaccumulation or ecological effect endpoint data as output labels, a neural network with two hidden layers is trained using the backpropagation algorithm until the model prediction error is less than a preset threshold.

[0013] Preferably, the operation process of the lightweight neural network in the dynamic weight adjustment engine is as follows: Using the ecotoxicity evolution trend and the mutation characteristics of biological response signals from the previous assessment period as input, the output is a fine-tuning amount for the weights of each data source in the current period. This fine-tuning amount is weighted and synthesized with the base weights output by the rule base to form the dynamic weights for the final application; The training process of the lightweight neural network is as follows: a training sample set is constructed, the sample features are toxicity trend indicators and biological response mutation indicators extracted from historical time series data, and the sample labels are the ideal weight adjustment directions determined by expert experience; the sample set is used to supervise the training of the small convolutional neural network.

[0014] Preferably, the system is deployed in a distributed architecture, including edge computing terminals deployed at each monitoring point and a cloud central server; The edge computing terminal is a lightweight version of the integrated sensor array module, data preprocessing and fusion module, and ecotoxicity dynamic assessment module. The cloud-based central server aggregates data from all edge terminals, runs a complete dynamic assessment model for ecotoxicity, performs in-depth analysis and iterative model training, and centrally manages the decision output module and knowledge base. The cloud-based central server also includes a model self-evolution unit, which periodically collects evaluation results generated during system operation, subsequent actual governance measures, and post-governance effect verification data. This data is used as new training samples to incrementally learn and optimize the calculation model and weight adjustment model in the ecotoxicity dynamic assessment module.

[0015] Preferably, the specific configuration of the microbial fuel cell sensor is as follows: Its anode is made of carbon felt material and loaded with localized microbial communities taken from the target soil, and the cathode is made of carbon cloth material covered with a catalytic layer. The anode and cathode are connected by an external resistor and placed in a specially designed soil chamber. Microbial metabolic activity is quantified using the stable current density or coulombic efficiency output by the sensor.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention breaks through the limitations of traditional soil testing that only focuses on pollutant concentration. By integrating bio-responsive sensing units that directly reflect ecological effects, and innovatively constructing an ecotoxicological dynamic assessment model that integrates concentration, bio-responsiveness, and environmental factors, it achieves direct and dynamic quantification from "pollutant content" to "ecological risk." This technical approach fundamentally solves the problem of insufficient biorelevance in existing assessment methods, making risk assessment results closer to the real ecological and environmental effects and providing a reliable basis for remediation.

[0017] 2. This invention, through the design of a dynamic weight adjustment engine and a model self-evolution unit, enables the system to possess environmental adaptability and continuous learning capabilities. The system can intelligently adjust the confidence levels of different data sources based on pollutant type, exposure history, and sudden events, and optimize internal model parameters using continuously accumulated field data. This design ensures that the system maintains the accuracy and robustness of its assessments in complex and ever-changing real-world soil environments over the long term, avoiding the assessment failure caused by changes in environmental conditions in static models.

[0018] 3. This invention constructs a complete closed-loop system from detection and assessment to decision-making recommendations. The decision output module transforms abstract toxicity equivalents into clear risk levels and specific remediation priorities and technical solution recommendations, enhancing the practical value and action guidance of the detection data. This system is not only a monitoring tool but also a decision support system that can improve the scientific rigor, timeliness, and economy of contaminated site management, remediation engineering planning, and implementation. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention; Figure 2 This is a schematic diagram of the core principle framework of the ecotoxicity dynamic assessment module in this invention; Figure 3 This is a logical flowchart of the data preprocessing and fusion module in this invention; Figure 4 This is a schematic diagram of the multi-level interaction relationship and data flow between the edge computing terminal and the cloud central server in this invention; Figure 5 This is a logical framework diagram of how the decision output module in this invention maps the comprehensive ecotoxicity equivalent to risk level and governance strategy. Detailed Implementation

[0020] This invention provides a soil pollutant detection and analysis system based on intelligent sensors, the overall technical architecture of which is shown in the attached figure. Figure 1 As shown, the system comprises four core components: a sensor array module, a data preprocessing and fusion module, an ecotoxicity dynamic assessment module, and a decision output module. Through in-situ, real-time, and multi-dimensional data acquisition and fusion analysis, the system dynamically quantifies the comprehensive ecotoxicity equivalent of soil pollutants and further outputs risk levels and remediation strategy recommendations, forming a complete closed loop from perception to decision-making. The following will be combined with the attached... Figure 1 To be continued Figure 5 This section provides a detailed implementation description of each functional module of the system, expanding upon it layer by layer.

[0021] First, the sensor array module is described in detail. This module is deployed at the soil sampling points to simultaneously acquire three types of heterogeneous information: pollutant concentration, biological response signals, and key environmental parameters. Its internal structure includes a concentration detection unit, a biological response sensing unit, and an environmental parameter monitoring unit.

[0022] The concentration detection unit integrates two types of hardware devices: an electrochemical sensor and a spectral sensor.

[0023] The electrochemical sensor employs a three-electrode system. The working electrode is a gold or carbon-based material modified with specific recognition molecules, the reference electrode is a silver / silver chloride electrode, and the counter electrode is a platinum wire. The current response signals of heavy metal ions such as lead, cadmium, and mercury at specific redox potentials are obtained by scanning with cyclic voltammetry or differential pulse voltammetry, and then converted into the corresponding ion concentrations by calibration curves.

[0024] The spectral sensor uses near-infrared or ultraviolet-visible absorption spectroscopy. It introduces the excitation light source into the soil micro-region through an optical fiber probe and receives the reflected or transmitted spectral signals. It uses the intensity of characteristic absorption peaks to invert the concentration of organic pollutants such as polycyclic aromatic hydrocarbons and pesticide residues.

[0025] All concentration data are output in micrograms per kilogram (μg / kg) or milligrams per kilogram (mg / kg), with timestamps and spatial coordinates.

[0026] The bioresponse sensing unit consists of a microbial fuel cell sensor and a plant physiological electrical signal sensor. The specific structure of the microbial fuel cell sensor is as follows: Its anode uses carbon felt material with high specific surface area, which is sterilized and loaded with localized microbial communities taken from the target soil to ensure that its metabolic characteristics are highly consistent with the in-situ environment; the cathode uses carbon cloth material covered with platinum carbon catalyst layer and is placed in a buffer solution containing dissolved oxygen; the anode and cathode are connected by an external resistor with a fixed resistance of 1000 ohms and are completely encapsulated in a specially made soil chamber that is permeable to water vapor and ions but isolated from large soil particles.

[0027] When organic pollutants in the soil are degraded by anodic microorganisms as electron donors, the generated electrons flow through the external circuit to the cathode, forming a stable current. The stable current density (mA / m²) or coulombic efficiency (%) is recorded in real time as a quantitative indicator of microbial metabolic activity. The metabolic activity inhibition rate is calculated by comparing it with the historical baseline value of the uncontaminated control site. , Given the current current density, The historical average baseline current density.

[0028] The deployment method of the plant physiological electrophysiological signal sensor is as follows: herbaceous plants with known sensitivity to target pollutants (such as ryegrass for heavy metals and Arabidopsis thaliana for polycyclic aromatic hydrocarbons) are selected as indicator plants. During their stable growth period, tungsten microelectrodes with a diameter of 50 micrometers are non-destructively implanted into specific areas of the root epidermal tissue using micromanipulation techniques. The microelectrodes are connected to a high-input-impedance bioelectric amplifier, and the sampling frequency is set to 1000 Hz to continuously record the transmembrane resting potential (unit: millivolt, mV) and the trigger frequency (unit: times / minute) and amplitude (unit: millivolt) of action potentials.

[0029] When pollutant stress leads to changes in cell membrane permeability or ion channel dysfunction, electrical signal characteristics shift significantly. The system extracts signal mutation features, such as a decrease in the absolute value of resting potential greater than 20% or an abnormal increase in action potential frequency more than 2 times, using a sliding window statistical method as a quantitative basis for plant stress intensity.

[0030] The environmental parameter monitoring unit is equipped with soil moisture sensors, soil temperature sensors, and soil pH sensors. The soil moisture sensor uses the frequency domain reflectance (FDR) principle, emitting 100 MHz high-frequency electromagnetic waves and inverting volumetric water content based on changes in dielectric constant, with an accuracy of ±2%. The soil temperature sensor is a PT1000 platinum resistance thermometer, with a measurement range of -20 to 80 degrees Celsius and a resolution of 0.1 degrees Celsius. The soil pH sensor uses a composite glass electrode with a built-in reference electrolyte, directly inserted into the soil slurry to measure pH values, with a range of 3.0 to 10.0 and an accuracy of ±0.1 pH units. All three environmental parameters are sampled synchronously at 10-second intervals and share the same spatiotemporal coordinate system with concentration and biological response data.

[0031] The above three types of raw data streams are transmitted uniformly to the data preprocessing and fusion module via the communication bus. Please refer to the appendix. Figure 3 The data preprocessing and fusion module executes three main sub-processes: standardized cleaning, feature extraction, and temporal adjustment. The data cleaning unit first aligns the timestamps of the sensor signals and employs a sliding window-based outlier detection algorithm: setting the window length to 60 sampling points and calculating the median of the data within the window. Absolute deviation from the median If a certain point satisfy If a value is found to be out of place, it is identified as an outlier and removed; missing values ​​are filled using cubic spline interpolation to ensure temporal continuity.

[0032] The cleaned signals are then fed into the feature extraction unit. For concentration signals, the moving average, standard deviation, maximum value, minimum value, and rate of change are extracted. For microbial current signals, in addition to the mean and variance, the main peak frequency of the power spectral density and the energy proportion in the low-frequency band (0.01-0.1 Hz) are extracted. For plant electrical signals, the action potential event density, resting potential fluctuation amplitude, and spectral entropy are extracted. For environmental parameters, the moving average and first derivative are extracted. All features are normalized to the [0,1] interval to eliminate dimensional differences.

[0033] The multi-source data fusion unit receives feature vectors from three sub-units and concatenates them according to the same spatiotemporal coordinates to form an initial high-dimensional feature vector, typically with 15 to 25 dimensions. Subsequently, principal component analysis (PCA) is used to reduce the dimensionality of this initial high-dimensional feature vector. The covariance matrix of PCA is constructed from historical training datasets, retaining principal components with a cumulative contribution rate greater than 95%, ultimately outputting a low-dimensional fusion feature vector of 5 to 8 dimensions. This low-dimensional fusion feature vector serves as the core input characterizing the current comprehensive state of soil monitoring points and is passed to the ecotoxicity dynamic assessment module.

[0034] The ecotoxicity dynamic assessment module is the core intelligent engine of the system, and its internal logic is as follows: Figure 2 As shown, it comprises two main subsystems: an ecotoxicity calculation engine and a dynamic weight adjustment engine. The ecotoxicity calculation engine first parses the concentration data of various pollutants from the fused feature vector and then queries the built-in pollutant basic toxicity database. This pollutant basic toxicity database stores toxicity benchmark values ​​such as reference dose (RfD), carcinogenic slope factor (CSF), or half-maximum effective concentration (EC50) for hundreds of common pollutants, with units uniformly converted to toxicity equivalent factor (TEF). For example, the TEF for cadmium is 1.0, for lead it is 0.5, and for benzo[a]pyrene it is 5.0.

[0035] Theoretical toxicity equivalent Calculated by weighted summation: , For the first The concentration of the pollutants, The corresponding toxicity equivalent factor.

[0036] Subsequently, the engine introduced a biological response correction factor. Environmental factor correction coefficient .

[0037] Biological response correction factor It is synthesized from two weighted components: the inhibition rate of microbial metabolic activity. With plant stress signal intensity . It can be directly calculated from the aforementioned inhibition rate formula; This is obtained by fuzzy membership mapping of plant electrical signal characteristics. For example, three fuzzy sets are defined: “mild stress”, “moderate stress”, and “severe stress”. The membership degree is calculated based on the degree to which the frequency and amplitude of the action potential deviate from the baseline, and then a weighted average is taken to obtain a continuous value between 0 and 1. , and The preset weights are all initially set to 0.5, and can be adjusted by the dynamic weight adjustment engine.

[0038] Environmental factor correction coefficient The result was calculated using a multivariate nonlinear regression model. This multivariate nonlinear regression model uses the aforementioned soil moisture... ,temperature pH The input variable is the pollutant bioavailability correction coefficient. The multivariate nonlinear regression model is in the form of a neural network with two hidden layers, each with 10 and 5 nodes respectively, and the activation function is ReLU.

[0039] The training process is as follows: Data was collected covering various soil types, including sandy soil, loam, and clay, with pollutant concentration gradients ranging from background levels to 10 times the standard limit, and various environmental conditions were combined. , , Historical datasets; , , Using the bioaccumulation factor (BCF) of pollutants measured under the same conditions in earthworms or plants as input features, the normalized value is used as the output label. The backpropagation algorithm is employed to optimize the network weights, with the mean squared error as the loss function. Training continues until the validation set error is less than 0.05. During runtime, actual measurements will be used... , , Substituting into the multivariate nonlinear regression model, the output is... This reflects the enhancing or inhibiting effect of environmental conditions on toxicity expression.

[0040] Ultimately, the comprehensive ecotoxicity equivalent The calculation logic is as follows: The formula reflects the coupling effect of concentration, biological response and environment, making the toxicity assessment results closer to real ecological scenarios.

[0041] The dynamic weight adjustment engine is responsible for optimizing the contribution weights of each data source during the data fusion process. Internally, it includes a rule base and a lightweight neural network. The rule base predefines initial weight allocation strategies for different pollutant categories: for heavy metal pollution, concentration data is weighted at 0.6, biological response at 0.3, and environmental parameters at 0.1; for persistent organic pollutants, the weights are 0.4, 0.4, and 0.2, respectively. The lightweight neural network is used to fine-tune these weights in real time.

[0042] This lightweight neural network is a small convolutional neural network, and its input is a toxicity trend indicator extracted over the past 7 days (such as...). The daily average rate of change, second derivative) and biological response mutation indicators (such as Week-on-week increase The number of mutation points is calculated, and the output is the incremental adjustment of the three weights. The training sample set is constructed from historical running data, and the labels are the ideal weight adjustment directions determined by domain experts based on subsequent validation data.

[0043] During system operation, the network analyzes the latest data window every 24 hours, outputs incremental adjustment suggestions, and performs a weighted synthesis with the base weights of the rule base: , Based on weights, For incremental adjustment, The learning rate is set to 0.1 to ensure smooth and stable adjustment. The synthesized dynamic weights are fed back to the principal component analysis stage of the data preprocessing and fusion module, affecting the projection direction of the feature vectors, thereby achieving adaptive optimization of the evaluation model.

[0044] Decision output module receives The value is then converted into actionable management instructions. Please refer to the appendix. Figure 5 The decision output module includes a grading mapping unit and a strategy generation unit. The grading mapping unit presets five toxicity threshold ranges: For security level, It is classified as low-risk. It is classified as medium risk. It is classified as high-risk. It is classified as extremely high risk. The system automatically categorizes the data through numerical comparison and generates corresponding color codes (green, yellow, orange, red, and purple) for visual display.

[0045] The strategy generation unit is associated with a structured knowledge base of governance measures. This knowledge base stores thousands of governance rules, each of which includes prerequisites (risk level, dominant pollutant type, soil texture, pH range) and recommended solutions (governance technology, engineering parameters, expected cycle).

[0046] For example, when the soil is classified as "high-risk", "cadmium is the dominant pollutant", and "the soil is acidic clay", the optimal solution matched by the system is: "use limestone powder for chemical passivation, with an application rate of 2000 kg per hectare and a mixing depth of 20 cm, which is expected to reduce bioavailability by more than 60%; or plant Sedum sarmentosum for plant extraction, with a planting density of 25 plants per square meter and a harvesting cycle of 90 days."

[0047] Based on the current location's risk level, pollutant identification results (output by the concentration detection unit), and soil environmental parameters (from the environmental parameter monitoring unit), the system retrieves the Top-3 matching solutions from the knowledge base, sorts them by treatment cost, cycle, and feasibility, and outputs a treatment priority number (1 being the highest priority) and detailed technical recommendations.

[0048] The entire system is deployed in a distributed architecture, as shown in the attached diagram. Figure 4 As shown, each monitoring point is equipped with a lightweight version of an edge computing terminal, which integrates a sensor array module, a data preprocessing and fusion module, and an ecotoxicity dynamic assessment module.

[0049] The edge terminals utilize ARM Cortex-A72 processors and run an embedded Linux system, featuring local data caching, real-time evaluation, and 4G / 5G communication capabilities. Their lightweight model prunes and quantizes principal component analysis and neural networks, keeping inference latency below 500 milliseconds to meet rapid on-site early warning requirements. All edge terminals upload raw data and evaluation results to the cloud central server via a secure, encrypted channel.

[0050] The cloud-based central server aggregates data from across the network, runs a complete dynamic assessment model for ecotoxicity, and performs cross-regional trend analysis, model parameter calibration, and knowledge base updates. In addition, the system includes a model self-evolution unit that collects three types of feedback data periodically (weekly): First, the output of the ecotoxicity dynamic assessment module value; Second, records of the governance measures actually adopted by users; Thirdly, the pollutant concentration and biological response data were retested after treatment.

[0051] These data, labeled as "input-action-outcome" triplets, serve as new training samples for incremental learning of the multivariate nonlinear regression model in the ecotoxicity calculation engine and the lightweight neural network in the dynamic weight adjustment engine. An online learning algorithm is employed, fine-tuning model parameters using only new samples each time to avoid catastrophic forgetting. Through this mechanism, the system's evaluation accuracy continuously improves over time, forming a positive evolutionary closed loop of "perception-decision-action-learning."

[0052] In summary, this embodiment, through precise sensor array design, rigorous data fusion process, innovative ecotoxicity dynamic assessment model, and closed-loop decision support mechanism, achieves accurate, dynamic, and operable assessment of soil pollution ecological risks, providing unprecedented technical support for soil environmental management.

[0053] Based on the aforementioned embodiments, this embodiment further refines the specific implementation of the system in emergency response scenarios for sudden pollution events. When the bioresponse sensing unit at a certain monitoring point detects a sudden increase of more than 50% in the inhibition rate of microbial metabolic activity within a short period of time (e.g., within 1 hour), or an abnormal surge of more than 3 times in the frequency of plant action potentials, the system immediately triggers a level-one early warning mechanism.

[0054] At this point, the edge computing terminal automatically switches to a high-frequency sampling mode: the sampling interval of the electrochemical sensor in the concentration detection unit is shortened from 10 minutes to 1 minute, and the spectral sensor starts continuous scanning; the microelectrode sampling rate of the bioresponse sensing unit is increased to 2000 Hz to capture transient electrical signals; and the environmental parameter monitoring unit simultaneously increases its sampling frequency to once every 30 seconds. All high-frequency data is temporarily stored in a local cache and uploaded to the cloud in real time via a priority queue.

[0055] Upon receiving the early warning signal, the cloud-based central server immediately activates the emergency analysis submodule. This submodule retrieves historical data from the same period to construct a dynamic baseline model. For example, if the current season is spring, the system automatically filters data from the same period over the past three years, using the same soil type and similar meteorological conditions, and calculates the normal fluctuation range of each parameter. Deviation analysis is performed between the real-time data and the dynamic baseline. If the concentration data deviation exceeds three times the standard deviation, and the biological response signal is synchronously abnormal, it is determined to be a sudden pollution event.

[0056] Subsequently, the system initiated auxiliary analysis for pollutant source tracing. Using data from multiple monitoring points deployed within the region, a pollutant diffusion inversion model was constructed. This model, based on soil hydrodynamic equations and combined with measured humidity and temperature fields, estimated the migration velocity and direction of pollutants. Simultaneously, by comparing pollutant characteristic spectra (such as heavy metal element proportions and organic matter fingerprints) from different locations, the type and possible location of the pollution source were preliminarily identified.

[0057] In the ecotoxicity dynamic assessment module, the dynamic weight adjustment engine temporarily modifies the weighting strategy: the weight of biological response data is temporarily increased to 0.7, concentration data is decreased to 0.2, and environmental parameters are maintained at 0.1, to highlight the immediate warning effect of biological effects. (Comprehensive ecotoxicity equivalent) The calculation frequency has been increased to once every 5 minutes, and a toxicity evolution heat map is generated to intuitively show the pollution spread trend.

[0058] The decision-making output module adjusts its output strategy accordingly. In addition to the standard risk level, it generates "emergency response recommendations," including: immediately sealing off suspected contaminated areas, initiating intensive sampling of groundwater monitoring wells, deploying mobile remediation equipment on standby, and notifying surrounding sensitive ecological protection areas to take protective measures. The remediation priority number is forcibly set to 1 and pushed to the emergency command platform of the relevant management department.

[0059] After the incident is handled, the model's self-evolutionary unit packages the entire process data of this incident—from early warning triggering, high-frequency monitoring, source tracing analysis to the handling effect—into typical cases and adds them to the training set. In particular, the system analyzes the response timing and sensitivity of each sensor in this incident to optimize future early warning thresholds. For example, if an abnormality is detected in the plant electrical signal 15 minutes earlier than the microbial current when a leak of a certain type of organic solvent is detected, a "plant response priority" flag is added to the pollutant category in the rule base to guide subsequent weight allocation.

[0060] Through the above mechanism, this embodiment enables the system to not only have accurate assessment capabilities under normal circumstances, but also to demonstrate rapid perception, intelligent judgment and efficient collaborative emergency response capabilities in sudden pollution events, significantly improving the resilience and agility of the soil environmental risk prevention and control system.

[0061] This embodiment focuses on the in-depth application of the system in long-term chronically contaminated sites, with particular emphasis on the continuous optimization of assessment accuracy by the model's self-evolving units. In a historically polluted heavy metal site, the system operated continuously for 18 months, accumulating over 500,000 valid data records.

[0062] Initially, due to a lack of localized parameters, the multivariate nonlinear regression model in the ecotoxicity calculation engine used general parameters, leading to an overestimation of cadmium bioavailability under acidic red soil conditions by approximately 30%. However, as the model's self-evolutionary units continuously received validation data after remediation—for example, after applying phosphate passivating agents, although the total soil cadmium concentration remained unchanged, the cadmium content in plant roots decreased by 65%, and earthworm survival rate recovered from 40% to 85%—the system gradually corrected the environmental factor correction coefficients. The computational logic.

[0063] Specifically, the self-evolutionary unit uses paired "environmental parameter-biological effect" data before and after each treatment as new samples to retrain the multivariate nonlinear regression model. After six iterations, the model's prediction error under local red soil conditions decreased from the initial 0.25 to 0.08. Meanwhile, the dynamic weight adjustment engine discovered that, within the site, plant physiological electrical signals responded more stably to cadmium stress than microbial currents and were less affected by seasonal temperature fluctuations. Therefore, the lightweight neural network gradually increased the weight of plant response data from 0.3 to 0.45, while correspondingly decreasing the weight of microbial data.

[0064] In addition, the knowledge base of the decision output module has also evolved in tandem. Initially, general-purpose lime was recommended as the passivating agent, but repeated practice showed that it was easily leached in red soil and its effect was not lasting. The system recorded the good results after switching to hydroxyapatite (passivation efficiency increased by 40% and the effective period was extended to 2 years), automatically updated the knowledge base rules, replaced the preferred passivating agent in the "acidic red soil + cadmium pollution" scenario with hydroxyapatite, and adjusted the recommended dosage to 1500 kg per hectare.

[0065] During this process, the system also identified a novel interaction effect: when the soil organic matter content is greater than 3%, the biological response signal remains weak even with high total pollutant concentrations. Therefore, the model's self-evolutionary unit virtually adds an "organic matter content" feature to the environmental parameter monitoring unit, indirectly estimating parameters by fusing them with remote sensing data or periodic laboratory test results, and incorporating this information into the model. Model input. This improvement increased the consistency between the assessment results and the field ecological survey from 75% to 92%.

[0066] This embodiment fully demonstrates that through continuous data feedback and model iteration, the system can deeply adapt to the complex environment of a specific site, evolving from a "general assessment tool" to a "dedicated intelligent advisor," providing increasingly accurate decision support for long-term pollution control.

[0067] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0068] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart sensor based soil pollutant detection analysis system characterized in that, The application relates to a soil pollution risk assessment and management system, which comprises the following modules: a sensor array module for in-situ acquisition of multi-dimensional physical, chemical and biological response signals of a soil environment, wherein the sensor array module is composed of a concentration detection unit, a biological response sensing unit and an environmental parameter monitoring unit; a data preprocessing and fusion module for standardizing and cleaning original heterogeneous data collected by the sensor array module, extracting features and performing space-time alignment, and generating a fusion feature vector, wherein the data preprocessing and fusion module comprises a data cleaning unit, a feature extraction unit and a multi-source data fusion unit; an ecological toxicity dynamic evaluation module for dynamically quantifying the comprehensive ecological toxicity equivalent of a pollutant based on the fusion feature vector output by the data preprocessing and fusion module, wherein the ecological toxicity dynamic evaluation module is internally provided with an ecological toxicity calculation engine and a dynamic weight adjustment engine; a decision output module for mapping the comprehensive ecological toxicity equivalent output by the ecological toxicity dynamic evaluation module into a specific risk assessment level and management priority suggestion, wherein the decision output module comprises a level mapping unit and a strategy generation unit.

2. A smart sensor based soil pollutant detection and analysis system as claimed in claim 1 wherein, The concentration detection unit is provided with electrochemical sensors and spectral sensors for directly measuring the concentrations of heavy metal ions and specific organic pollutants in the soil. The biological response sensing unit is integrated with microbial fuel cell type sensors and plant physiological electrical signal sensors for capturing the metabolic activity changes of soil microbial communities and the stress electro-physiological signals of plant roots. The environmental parameter monitoring unit is provided with soil humidity sensors, soil temperature sensors and soil pH sensors for synchronously monitoring key environmental factors that affect the migration, transformation and bioavailability of pollutants.

3. The smart sensor based soil pollutant detection and analysis system as claimed in claim 2, wherein, The data cleaning unit adopts a sliding window-based outlier detection algorithm and an interpolation algorithm to process noise and missing values in the original sensor signals. The feature extraction unit performs time domain and frequency domain analysis on the cleaned time series signals to extract statistical features and spectral features including mean, variance, peak value and main frequency component. The multi-source data fusion unit splices the feature vectors from the concentration detection unit, the biological response sensing unit and the environmental parameter monitoring unit under the same space-time coordinates, and performs dimension reduction processing by using a principal component analysis method to output a low-dimensional fusion feature vector representing the comprehensive state of the current soil point.

4. The smart sensor based soil pollutant detection and analysis system as claimed in claim 3, wherein, The ecological toxicity calculation engine receives the fusion feature vector from the data preprocessing and fusion module as input, queries an internal pollutant basic toxicity database according to the concentration data to calculate a concentration-based theoretical toxicity equivalent, further introduces a biological response correction factor and an environmental factor correction coefficient, and performs the following calculation logic: the comprehensive ecological toxicity equivalent is equal to the product of the theoretical toxicity equivalent and the biological response correction factor, and then multiplied by the environmental factor correction coefficient. The dynamic weight adjustment engine is used for adaptively adjusting the contribution weights of concentration data, biological response data and environmental data in the fusion feature vector according to the type and pollution history of the pollutant, and the dynamic weight adjustment engine is internally provided with a rule base and a lightweight neural network.

5. The smart sensor based soil pollutant detection and analysis system as claimed in claim 4, wherein, The grade mapping unit compares the continuous integrated ecological toxicity equivalent value with preset toxicity threshold intervals, and automatically classifies the corresponding risk grade; The strategy generation unit is associated with a governance measure knowledge base, and matches and outputs the optimal governance priority number and preliminary technical scheme suggestion from the knowledge base according to the risk grade of the current point, the pollutant identification result and the soil environment parameter.

6. The smart sensor based soil pollutant detection and analysis system as claimed in claim 5, wherein, The calculation process of the biological response correction factor is as follows: The biological response sensing unit collects the microbial metabolic activity inhibition rate and the plant stress signal intensity, and compares and maps them with a preset dose-response relationship curve to obtain the biological response correction factor.

7. A smart sensor based soil pollutant detection and analysis system as claimed in claim 6, wherein, The calculation process of the environmental factor correction coefficient is as follows: The measured values of soil humidity, temperature, and pH are input into a multivariate nonlinear regression model trained based on a large amount of historical data, which describes the influence law of environmental conditions on the bioavailability of pollutants. The environmental factor correction coefficient is calculated by the model; The training process of the multivariate nonlinear regression model is as follows: collect historical data sets covering different soil types, different pollutant concentration gradients, and different environmental condition combinations. The data set contains environmental parameters, final biological enrichment of pollutants, or ecological effect endpoint data. Take the aforementioned environmental parameters as input features, and take the normalized value of the biological enrichment or ecological effect endpoint data as output labels. Train a neural network with 2 hidden layers using the back propagation algorithm until the model prediction error is less than the preset threshold.

8. The smart sensor based soil pollutant detection and analysis system as claimed in claim 7, wherein, The running process of the lightweight neural network in the dynamic weight adjustment engine is as follows: The ecological toxicity evolution trend and the mutation characteristics of the biological response signal in the previous evaluation period are input, and the adjustment amount of the weight of each data source in the current period is output; The adjustment amount is weighted and synthesized with the base weight output by the rule base to form the final dynamic weight applied; The training process of the lightweight neural network is as follows: construct a training sample set, the sample features are the toxicity trend indicators and biological response mutation indicators extracted from the historical time series data, and the sample labels are the ideal weight adjustment directions labeled by experts; use the sample set to supervise the training of the small convolutional neural network.

9. The smart sensor based soil pollutant detection and analysis system as claimed in claim 8, wherein, The system is deployed in a distributed architecture, including edge computing terminals deployed at each monitoring point and a cloud center server; The edge computing terminal integrates a sensor array module, a data preprocessing and fusion module, and a lightweight version of the ecological toxicity dynamic evaluation module; The cloud center server aggregates data from all edge terminals, runs the complete ecological toxicity dynamic evaluation model for deep analysis and model iterative training, and centrally manages the decision output module and knowledge base; The cloud center server also includes a model self-evolution unit that periodically collects evaluation results, subsequent actual governance measures, and post-governance effect verification data generated during system operation, and uses these data as new training samples to perform incremental learning and parameter optimization on the calculation model and weight adjustment model in the ecological toxicity dynamic evaluation module.

10. The smart sensor based soil pollutant detection and analysis system as claimed in claim 9, wherein, The specific structure of the microbial fuel cell type sensor is as follows: The anode is made of carbon felt material and loaded with localized microbial community from the target soil, the cathode is made of carbon cloth material covered with catalytic layer, the anode and the cathode are connected through an external resistor and placed in a specially designed soil cabin; The metabolic activity of the microorganisms is quantified as a stable current density or coulombic efficiency output by the sensor.

Images