Intelligent water quality monitoring and early warning method and system for reservoir environment

Abstract

The application discloses a kind of intelligent water quality monitoring and early warning method and system for reservoir environment.Method includes: deployment multi-depth water temperature sensor array, adaptive collection vertical temperature profile data;Establish the nonlinear mapping of water temperature and water optical refractive index, according to temperature gradient, water is divided into optical property uniform sub-water layer, based on Snell's law layer by layer calculation light deflection and cumulative optical path, obtain effective optical path correction coefficient;Establish the light collection efficiency exponential decay model with water depth and calculate compensation coefficient, correct original spectrum;Establish the quantitative mapping model of corrected spectrum and water quality parameter, and retrieve water quality parameter;Using multi-parameter joint discrimination strategy containing false alarm suppression logic determines abnormal and generates early warning report.The application effectively eliminates optical distortion deviation caused by water temperature stratification, improves deep water area water quality monitoring accuracy and early warning reliability.

Description

Technical Field

[0001] This application belongs to the technical field of environmental monitoring and early warning, specifically relating to an intelligent water quality monitoring and early warning method and system for reservoir environments. Background Technology

[0002] In reservoir water quality monitoring, the water quality parameter inversion method based on spectral analysis achieves real-time monitoring of indicators such as dissolved oxygen, turbidity, and chlorophyll by acquiring water spectral data and establishing a mapping relationship with water quality parameters. However, in the deep water area of ​​the reservoir, the water temperature exhibits a significant hierarchical distribution characteristic along the vertical direction. There is an obvious optical refractive index gradient between the surface warm water, the thermocline, and the deep cold water, which causes distortion of the light propagation path in the water body, resulting in a systematic bias in the spectral measurement of the deep water area that is difficult to ignore.

[0003] Common water quality monitoring methods typically improve monitoring accuracy by enhancing sensor performance or optimizing spectral inversion algorithms. However, these methods usually treat water bodies as optically homogeneous media, neglecting the complex changes in optical propagation characteristics caused by water temperature stratification. They are also unable to effectively correct for the effective optical path offset and light collection efficiency attenuation caused by non-uniform refractive index distribution. This results in significant deviations between deep-water water quality monitoring data and actual water quality conditions, making it impossible to provide reliable evidence for reservoir water quality safety early warning. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the present invention provides an intelligent water quality monitoring and early warning method and system for reservoir environment, which can effectively solve the problems in the background technology mentioned above.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: In a first aspect, the present invention provides an intelligent water quality monitoring and early warning method for reservoir environments, the method comprising the following steps: Deploying a multi-depth water temperature sensor array to collect vertical temperature profile data of the reservoir; wherein, deploying the multi-depth water temperature sensor array includes: acquiring the vertical temperature profile of the reservoir and calculating the average water temperature gradient change rate in each depth interval, and adaptively configuring the sensor arrangement spacing in each water layer area according to the water temperature gradient change rate, reducing the sensor arrangement spacing in areas with a large water temperature gradient change rate, and increasing the sensor arrangement spacing in areas with a small water temperature gradient change rate. Based on the collected vertical temperature profile data, a nonlinear mapping function between water temperature values ​​and the optical refractive index of water was established, and the characteristics of the optical refractive index gradient distribution of different water temperature layers were analyzed. Based on the optical refractive index gradient distribution characteristics, the water body is divided into multiple optically homogeneous sub-water layers based on the temperature gradient change rate. The refraction deflection angle and the direction of the refracted light when the light passes through the interface of each sub-water layer are calculated layer by layer based on Snell's law of refraction. The cumulative optical path length is obtained by integrating the length of the straight-line propagation path of the light in all sub-water layers using the piecewise summation method. Based on the cumulative optical path length and the apparent optical path length assuming uniform water medium, an effective optical path correction coefficient is established, and an exponential decay curve model of light collection efficiency with water depth is established to calculate the compensation coefficient. After performing spectral intensity normalization, spectral waveform smoothing and baseline correction operations on the original spectral data, the processed spectral data is multiplied by the effective optical path correction coefficient and the compensation coefficient to generate corrected spectral data. A quantitative mapping model between the corrected spectral data and water quality parameters is established. The real-time collected raw spectral data is input into the quantitative mapping model to obtain the predicted values ​​of water quality parameters. A multi-parameter joint discrimination strategy is adopted to compare the predicted values ​​of the water quality parameters with the preset safety threshold range to determine whether there is an abnormality in the water quality and generate an early warning report. The multi-parameter joint discrimination strategy includes false alarm suppression logic: when only the predicted value of a single water quality parameter exceeds the safety threshold range and the extent of the exceedance is less than the preset tolerance value, the abnormality judgment is not triggered.

[0006] On the other hand, the present invention provides an intelligent water quality monitoring and early warning system for reservoir environments, comprising: The temperature profile acquisition module is used to deploy a multi-depth water temperature sensor array at different depths in the vertical direction of the reservoir to acquire vertical temperature profile data of the reservoir. The deployment of the multi-depth water temperature sensor array includes: acquiring the vertical temperature profile of the reservoir and calculating the average water temperature gradient change rate in each depth interval; adaptively configuring the sensor arrangement spacing in each water layer area according to the water temperature gradient change rate, reducing the sensor arrangement spacing in areas with a large water temperature gradient change rate and increasing the sensor arrangement spacing in areas with a small water temperature gradient change rate. The dynamic update module is used to periodically collect water temperature data at each depth point based on the multi-depth water temperature sensor array, and dynamically adjust the data collection update frequency according to the water temperature change range to construct a real-time updated reservoir vertical temperature profile dataset. The refractive index analysis module is used to establish a nonlinear mapping function between water temperature and water optical refractive index based on the vertical temperature profile dataset of the reservoir, and to calculate the optical refractive index values ​​at each depth point to generate a refractive index distribution. The physical reconstruction module is used to divide the water body into multiple optically uniform sub-water layers based on the refractive index distribution by comparing the temperature gradient change rate with a preset temperature gradient change rate threshold, and to calculate the light deflection layer by layer based on Snell's law of refraction and perform path integration to obtain the cumulative optical path length and cumulative path curvature angle, so as to construct a physical reconstruction model of water body optical distortion. The optical path correction module is used to establish an effective optical path correction coefficient based on the cumulative optical path length and the apparent optical path length when the water medium is assumed to be uniform, and to establish an exponential decay curve model of light collection efficiency with water depth to calculate the compensation coefficient. After performing spectral intensity normalization, spectral waveform smoothing and baseline correction operations on the original spectral data, the processed spectral data is multiplied by the effective optical path correction coefficient and the compensation coefficient to generate corrected spectral data. The parameter mapping module is used to extract spectral features from the corrected spectral data and use a machine learning model to establish a quantitative mapping relationship between the spectral features and standard water quality parameters in order to retrieve water quality parameters. The early warning output module is used to determine abnormal water quality conditions based on the inverted water quality parameters and a multi-parameter joint discrimination strategy including false alarm suppression logic. When an abnormality is determined, an early warning report containing information on the type, degree, location and time of the abnormality is generated and sent to the early warning terminal device through a wireless communication link.

[0007] In summary, this application includes at least one of the following beneficial technical effects: 1. This invention establishes a nonlinear mapping function between water temperature and the optical refractive index of water, and divides the water body into multiple optically homogeneous sub-water layers by combining the temperature gradient change rate. The light deflection and cumulative optical path are calculated layer by layer, which can effectively eliminate the distortion of light propagation path caused by vertical water temperature stratification in reservoirs. It solves the problem of traditional methods treating the water body as an optically homogeneous medium and ignoring the influence of refractive index gradient, and significantly improves the physical authenticity of spectral data in deep water areas.

[0008] 2. This invention employs an adaptive deployment strategy for multi-depth water temperature sensor arrays. In the thermocline region where the water temperature gradient change rate is large, the sensor spacing is reduced to improve the temperature profile resolution. In the deep water region where the water temperature gradient change rate is small, the sensor spacing is increased to reduce hardware costs. At the same time, combined with a dynamic adjustment mechanism for the acquisition cycle, it achieves accurate tracking of changes in the thermal structure of the reservoir and a reasonable balance between resource consumption, ensuring the accuracy and timeliness of the temperature input data required for optical distortion correction.

[0009] 3. This invention employs a quantitative mapping model between corrected spectral data and water quality parameters, and combines it with a multi-parameter joint discrimination strategy that includes false alarm suppression logic to determine anomalies. This effectively compensates for the systematic deviations caused by optical path distortion and light collection efficiency attenuation, while avoiding frequent false alarms caused by instantaneous fluctuations in a single water quality parameter. As a result, it provides reservoir managers with more reliable and stable water quality anomaly early warning results, and improves the practicality and credibility of reservoir water quality safety monitoring. Attached Figure Description

[0010] Figure 1 A schematic diagram of the overall technical solution for intelligent water quality monitoring and early warning methods for reservoir environments; Figure 2 A schematic diagram illustrating the core principle of the layered adaptive optics distortion correction method; Figure 3 A schematic diagram of the deployment scheme and data acquisition process for a multi-depth water temperature sensor array; Figure 4 A schematic diagram of the logical flow of the physical reconstruction model of optical distortion in water bodies; Figure 5 A schematic diagram of the logic flow of the effective optical path correction and light collection efficiency compensation algorithm; Figure 6 This is a schematic diagram illustrating the construction process of the mapping model between corrected spectral data and water quality parameters. Figure 7 A schematic diagram of the logical flow for generating water quality monitoring results and outputting early warnings. Detailed Implementation

[0011] To make the objectives, technical solutions, and advantages of this invention clearer, the following description is provided in conjunction with the appendix. Figure 1 To be continued Figure 7 Specific embodiments are provided to further illustrate the present invention in detail.

[0012] Firstly, this embodiment provides an intelligent water quality monitoring and early warning method for reservoir environments, which includes the following technical steps.

[0013] The first step, S1, involves deploying a multi-depth water temperature sensor array and collecting vertical temperature profile data. Step S1 forms the physical foundation of the entire water quality monitoring and early warning method, aiming to obtain vertical temperature distribution data of the reservoir to provide data support for subsequent optical distortion correction. Due to the stratification of water temperature in the deep water area of ​​the reservoir, there are differences in optical properties between the surface warm water, the thermocline, and the deep cold water. Accurately obtaining the vertical temperature profile is a prerequisite for effective optical distortion correction. Step S1 uses a layered, adaptively arranged multi-depth water temperature sensor array to measure the vertical temperature distribution, providing a data source for the subsequent construction of the water temperature-refractive index mapping model. The deployment scheme is explained in detail below through steps S101 to S102.

[0014] Step S101: Determine the sensor spacing and number of nodes for each water layer region, and deploy a multi-depth water temperature sensor array at different vertical depths in the reservoir. The multi-depth water temperature sensor array contains three types of temperature sensor components, which are used to collect temperature data from different vertical water layer regions in the reservoir.

[0015] Within the surface water temperature region of 0 to 5 meters below the surface of the reservoir, a first type of temperature sensor group is arranged at a first preset interval to collect temperature data of the surface water layer. As an example and not a limitation, the first preset interval is set to 1 meter. If the two ends of the interval are included, a total of 6 sensor nodes are arranged in the surface water temperature region.

[0016] Within the thermocline region, 5 to 20 meters below the reservoir surface, a second type of temperature sensor group is deployed at a second preset interval to collect temperature gradient data of the thermocline. The second preset interval is smaller than the first preset interval; this is for example only and not a limitation. The second preset interval is set to 0.5 meters to accommodate the drastic temperature changes with depth in the thermocline region. A total of 31 sensor nodes are deployed in the thermocline region according to the second preset interval.

[0017] In the deep water layer region, from 20 meters below the reservoir surface to the bottom, a third type of temperature sensor group is deployed at a third preset interval to collect temperature data of the deep, cold water layer. The third preset interval is larger than the first preset interval; this is for example only and not a limitation. The third preset interval is set to 2 meters to accommodate the relatively gradual temperature changes in the deep water layer. The number of sensor nodes in the deep water layer region is determined based on the actual water depth of the reservoir. When the reservoir bottom depth is 30 meters, a total of 6 sensor nodes are deployed in the deep water layer region.

[0018] The spatial spacing between temperature sensors is adaptively configured based on the rate of change of water temperature gradient in the corresponding depth region. Before deployment, a vertical temperature profile is obtained using a portable temperature depth meter or by retrieving historical monitoring data from the reservoir, and the average rate of change of water temperature gradient in each depth interval is calculated accordingly. In areas with a large rate of change of water temperature gradient, the sensor spacing is reduced; in areas with a small rate of change of water temperature gradient, the spacing is increased, thereby achieving adaptive matching of the sensor placement scheme to the different thermal structural characteristics of the reservoir.

[0019] The above spacing values ​​are exemplary configurations for conventional medium-sized reservoirs. In practical applications, the specific values ​​of the first, second, and third preset spacings are adaptively adjusted based on the measured results of the actual water depth, thermocline thickness, and water temperature gradient change rate of the reservoir. If the reservoir's hydrological conditions change, the temperature profile is remeasured during the maintenance cycle, and the sensor node positions are adjusted accordingly.

[0020] Step S102: Configure the hardware architecture of the sensor array. The hardware architecture of the multi-depth water temperature sensor array includes sensor nodes, signal conditioning circuits, data acquisition units, and communication interfaces.

[0021] The sensor nodes employ high-precision platinum resistance temperature sensors or thermistor sensors. The signal conditioning circuit amplifies, filters, and performs analog-to-digital conversion on the weak electrical signals output by the sensor nodes. The data acquisition unit polls and samples each sensor node according to a preset timing sequence and stores the sampled data in a local buffer. The communication interface uses waterproof cables or a wireless RF module to transmit the acquired temperature data to a host computer or cloud server.

[0022] In summary, step S1 constructs a dedicated sensor array for monitoring the vertical temperature profile of the reservoir using a hierarchical adaptive deployment strategy. Regarding the deployment strategy, sensor spacing is matched to the differences in thermal characteristics at different depths of the reservoir, and specific parameters are determined using measured or historical data before deployment. In terms of hardware implementation, a complete technical chain is formed from sensing elements, signal conditioning, data acquisition to remote communication. The multi-depth water temperature sensor array provides vertical temperature distribution data for subsequent water temperature-refractive index mapping analysis.

[0023] The next step, S2, involves constructing and dynamically updating the reservoir's vertical temperature profile dataset. Step S2 follows the deployment of the multi-depth water temperature sensor array in step S1, systematically managing the collected raw temperature data to provide real-time updated vertical temperature profile data for subsequent refractive index analysis. The core of this step lies in adaptively adjusting the acquisition frequency to cope with dynamic changes in water temperature, and ensuring data integrity and transmission efficiency through a reasonable data structure and storage strategy. Steps S201 to S202 will be described in detail below.

[0024] Step S201: Set the initial acquisition cycle and adaptively adjust it according to the water temperature change range. Based on the multi-depth water temperature sensor array deployed in step S101, this embodiment sets the initial acquisition cycle. During operation according to the initial acquisition cycle, the data acquisition unit continuously monitors the water temperature change range between two adjacent acquisition cycles.

[0025] When the water temperature change exceeds a preset threshold, the data acquisition unit automatically shortens the acquisition cycle from the initial cycle to a dynamic cycle. When the water temperature change is below the preset threshold for several consecutive acquisition cycles, the data acquisition unit automatically restores the acquisition cycle to the initial cycle.

[0026] Through the aforementioned adaptive adjustment mechanism, the sampling frequency is automatically increased during periods of drastic water temperature fluctuations caused by seasonal changes or diurnal variations, and automatically decreased when the water temperature tends to stabilize, thereby reducing system power consumption and communication bandwidth usage without losing key dynamic information.

[0027] Step S202: Construct a vertical temperature profile dataset and perform data storage and synchronization. After acquiring the temperature values ​​of all sensor nodes at each acquisition time, the data acquisition unit organizes them into a vertical temperature profile dataset according to a preset data structure. The dataset is organized using a hierarchical index format and includes a depth scale field, a timestamp field, and a temperature value field. The depth scale field records the depth position of the sensor node in the vertical direction of the reservoir, the timestamp field records the time of the acquisition, and the temperature value field records the temperature reading at that depth point at that time.

[0028] The dataset is stored using a combination of local and remote storage: the local storage medium is a local solid-state storage device deployed at the monitoring site, used to temporarily cache the collected data to prevent data loss in the event of communication interruption; the remote storage medium is a cloud server or host computer database, used for long-term data storage and subsequent analysis and retrieval.

[0029] The dataset is updated using an incremental synchronization strategy, meaning that the data acquisition unit only transmits newly generated temperature profile data slices since the last synchronization to the remote storage medium, and does not repeatedly transmit historical data slices that have already been synchronized, in order to reduce communication bandwidth usage.

[0030] In summary, step S2 establishes a complete data management mechanism, from data acquisition frequency control to data storage and synchronization. Regarding acquisition frequency control, the acquisition cycle is adaptively adjusted using the water temperature change amplitude as a feedback signal. For data management, a hierarchical indexed data structure, dual local and remote storage media, and an incremental synchronization strategy ensure the integrity, retrieval, and transmission efficiency of the temperature profile data. The real-time updated vertical temperature profile dataset output in step S2 will serve as the direct data source for subsequently establishing the water temperature-refractive index mapping relationship.

[0031] In step S3 of this embodiment, a model relating water temperature to the optical refractive index of the water body is established, and the refractive index gradient distribution characteristics are analyzed. Step S3 uses the vertical temperature profile dataset output in step S2 as input to establish a quantitative mapping relationship between water temperature and the optical refractive index of the water body, and analyzes the refractive index gradient distribution characteristics of different water temperature layers, providing basic data for subsequent physical reconstruction calculations of the light propagation path. The specific implementation follows these steps.

[0032] Step S301: Establish a nonlinear mapping function between water temperature values ​​and the optical refractive index of the water body. This step establishes the mapping relationship between water temperature and refractive index beforehand through experimental calibration. The experimental calibration process is carried out under standard conditions, covering the complete range of possible water temperature changes in the reservoir. Multiple temperature sampling points are selected within the range of changes, and the optical refractive index values ​​of the water body corresponding to each temperature sampling point are measured to obtain multiple sets of discrete temperature-refractive index data points.

[0033] After obtaining the aforementioned discrete data points, a continuous nonlinear mapping function between water temperature values ​​and the optical refractive index of the water body is established using either a polynomial fitting method or a Gaussian process regression method. The polynomial fitting method determines the polynomial coefficients through least-squares fitting; the Gaussian process regression method selects a covariance function and optimizes the hyperparameters using training data to obtain a probabilistic model that can predict the refractive index at any temperature. The nonlinear mapping function obtained after fitting serves as the basis for subsequent refractive index calculations.

[0034] Step S302: Calculate the optical refractive index values ​​at each depth point and generate the refractive index distribution. Using the vertical temperature profile dataset provided in step S2, substitute the water temperature value at each depth location into the nonlinear mapping function established in step S301 to calculate the optical refractive index value corresponding to that depth point.

[0035] The calculation process employs either a lookup table method or a real-time interpolation method. The lookup table method stores the temperature-refractive index mapping relationship as a discrete data table, and during calculation, the refractive index is obtained by looking up the table and interpolating based on the measured temperature. The real-time interpolation method dynamically calls a nonlinear mapping function to directly calculate the refractive index during calculation. The calculation results for all depth points collectively constitute the optical refractive index distribution in the vertical direction of the reservoir.

[0036] Step S303: Analyze the refractive index gradient at the interface of adjacent sub-water layers and establish a calculation model for the light deflection angle. Perform gradient analysis on the refractive index distribution obtained in step S302. For any spatial interval between two adjacent depth points in the vertical direction, calculate the refractive index gradient value within that interval, which is the difference in refractive index values ​​between the two depth points divided by the spatial distance between them. By calculating the refractive index gradient values ​​of all adjacent depth points, find the spatial interval where the refractive index gradient reaches its maximum. The depth range of this interval is the critical temperature jump region with the maximum refractive index gradient.

[0037] Based on this, a calculation model for the refraction and deflection angle of light rays passing through different water temperature layers is established using Snell's law of refraction. The expression for Snell's law of refraction is:

[0038] in, The angle of incidence is the angle between the incident ray and the interface normal. The angle of refraction is the angle between the refracted ray and the interface normal. The optical refractive index of the sub-water layer where the incident light ray is located; is the optical refractive index of the sub-water layer where the refracted light ray is located.

[0039] In the calculation, given the incident ray direction vector and the interface normal direction vector, the angle of incidence is calculated first. Then, calculate the angle of refraction according to Snell's law of refraction. Then, the direction vector of the refracted light is determined by combining the interface normal direction vector. This calculation model can quantitatively describe the deflection direction and angle of light as it passes through each water temperature layer interface.

[0040] In summary, step S3 establishes a nonlinear mapping function between water temperature and refractive index based on experimental calibration data, calculates the refractive index values ​​at each depth point using the vertical temperature profile dataset, and generates a refractive index distribution. Based on this, the critical temperature jump region is located through gradient analysis of the refractive index distribution, and a calculation model for the refraction deflection angle of light passing through different water temperature layers is established based on Snell's law of refraction. The refractive index distribution data and the refraction deflection angle calculation model output in step S3 will serve as direct input conditions for constructing the physical reconstruction model of water optical distortion in step S4.

[0041] The next step, S4, involves constructing a physical reconstruction model of optical distortion in the water body. Step S4 builds upon the refractive index distribution data and refraction deflection angle calculation model output from step S3, establishing a physical reconstruction model describing the actual propagation path of light in the water body. This provides a quantitative basis for subsequent effective optical path correction. Steps S401 to S403 are described in detail below.

[0042] Step S401: Based on the temperature gradient change rate, the water body is divided into multiple optically homogeneous sub-water layers. Using the vertical temperature profile data output in step S2 as a basis, the water body is vertically stratified. For any two adjacent depth points, the difference in temperature values ​​between the two depth points is divided by the difference in depth values ​​to obtain the temperature gradient change rate for that depth interval.

[0043] in, Water temperature, in degrees Celsius; Depth, in meters; and The index of two adjacent depth points is a positive integer; For the first The rate of change of temperature gradient in a depth range, in degrees Celsius per meter.

[0044] The calculated rate of change of temperature gradient With preset threshold Comparison. Among them, A preset temperature gradient change rate threshold is used to determine the water stratification boundary, expressed in degrees Celsius per meter; this threshold is preset based on the statistical characteristics of the thermocline gradient in historical water temperature profile data of the reservoir. Below When this occurs, it indicates that the water temperature change is gradual within this range, and adjacent depth points are merged into the same sub-water layer, which corresponds to a constant optical refractive index value. Higher than When the temperature changes drastically within a given range, a new sub-water layer boundary is demarcated at that location, dividing the upper and lower parts into different sub-water layers. After the above processing, the entire vertical water body is divided into multiple sub-water layers, with the optical refractive index within each sub-water layer considered a constant value.

[0045] Step S402 involves calculating the light deflection layer by layer based on Snell's law of refraction and performing path integration. After completing the water stratification, the refraction deflection angle calculation model established in step S3 is used to calculate the refraction deflection angle and the direction of the refracted light as it passes through each sub-water layer interface, based on Snell's law of refraction. The expression for Snell's law of refraction has been given in step S3, and the meaning of each letter has been explained in step S3.

[0046] For light incident on the first The water layer and the first Given the interface between the two water layers, the incident ray direction vector and the interface normal direction vector, calculate the incident angle. Then, calculate the angle of refraction according to Snell's law of refraction. Then, by combining the interface normal direction vector, the direction vector of the refracted light is determined through vector operations.

[0047] The integral calculation is performed along the actual propagation path of the light using a piecewise summation method. For each sub-water layer the light passes through, the length of the straight-line propagation path of the light within that sub-water layer is calculated. The cumulative optical path length is obtained by summing the straight-line propagation path lengths of light within all sub-water layers:

[0048] in This represents the total number of sub-water layers through which the light passes. For the light in the first The length of the linear propagation path within a sub-water layer This represents the cumulative optical path length.

[0049] The angle of refraction and deflection of light at each sub-water layer interface By stacking the values ​​one by one, the cumulative path curvature angle can be obtained:

[0050] in For the light in the first The angle of refraction at each interface. This represents the cumulative path curvature angle.

[0051] Step S403: Establish a correction mapping relationship between the effective optical path and the apparent optical path, based on the cumulative optical path length calculated in step S402. and cumulative path bending angle A corrective mapping relationship between the effective optical path and the apparent optical path is established. The effective optical path refers to the actual geometric path length of light propagating in a non-uniform medium such as water, i.e., the cumulative optical path length. The apparent path length refers to the geometric path length of light rays traveling in a straight line when the water medium is assumed to be homogeneous; that is, the straight-line distance from the light source to the detector. .

[0052] The calibration mapping relationship was established using a nonlinear regression method: for multiple different water depth conditions, the effective optical path value was calculated for each water depth condition. , sightseeing tour value and cumulative path bending angle , constitute the training dataset ,in The total number of samples.

[0053] Using apparent optical path length and cumulative path curvature angle as inputs and effective optical path length as outputs, a nonlinear regression fitting is performed to find the mapping function. Make the sum of squared prediction errors Minimize. The output parameters of the nonlinear regression fit are the effective optical path correction coefficients. :

[0054] This coefficient characterizes the degree of deviation of the actual propagation path from the ideal straight path, and serves as the basis for spectral data correction in subsequent steps.

[0055] In summary, step S4 stratifies the water body using a temperature gradient threshold, calculates the light deflection layer by layer, and accumulates the results to obtain the actual optical path and path curvature. Finally, a nonlinear regression mapping relationship between the effective optical path and the apparent optical path is established. The correction mapping relationship and effective optical path correction coefficient output by this model will serve as the direct basis for performing effective optical path correction in step S5.

[0056] In step S5 of the method in this embodiment, effective optical path correction and light collection efficiency compensation are performed. Step S5 takes the effective optical path correction coefficient and the physical reconstruction model of water optical distortion output in step S4, and performs physical correction on the original spectral data to eliminate the systematic bias introduced by optical path distortion and light collection efficiency attenuation. This is carried out according to the following steps.

[0057] Step S501: Obtain the effective optical path correction coefficient under the current water depth conditions. Based on the correction mapping relationship between the effective optical path and the apparent optical path established by the physical reconstruction model of water optical distortion in step S4, obtain the effective optical path correction coefficient under the current water depth conditions. The calculation formula for this coefficient has been given in step S4 as the effective optical path divided by the apparent optical path.

[0058] Step S502: Establish a model of the attenuation curve of light collection efficiency as a function of water depth and calculate the compensation coefficient. Due to factors such as the absorption and scattering of light by the water body and the limitation of the receiving angle of the light collection device, the light collection efficiency exhibits attenuation characteristics with increasing water depth. This step establishes a mathematical model describing this attenuation characteristic and calculates the compensation coefficient based on the model.

[0059] The change in light collection efficiency with water depth is described by an exponential decay model:

[0060] in, Water depth, measured in meters; water depth The light collection efficiency at the location is dimensionless. The water surface is The theoretical maximum light collection efficiency at that time is dimensionless. This is the light collection efficiency attenuation coefficient, expressed per meter.

[0061] Attenuation coefficient The acquisition method is as follows: In a standard water body with known clear water quality and no suspended particulate matter, standard light sources and photodetectors are placed at multiple different water depths, and the actual light collection efficiency values ​​at each water depth are measured to obtain multiple sets of discrete data points of water depth-light collection efficiency; then, an exponential decay model is used... To determine the attenuation coefficient, the least squares fitting method is used to fit the discrete data points. The optimal estimate is obtained. This optimal estimate comprehensively reflects the combined effects of water absorption, scattering, and device receiving angle limitations on light collection efficiency.

[0062] The compensation coefficient is calculated using an inverse function mapping, which restores the attenuated light collection efficiency to its theoretical maximum value. :

[0063] in, The compensation coefficient is dimensionless. This is the theoretical maximum light collection efficiency at the water surface, which is dimensionless. water depth The light collection efficiency at the location is dimensionless. This is the light collection efficiency attenuation coefficient, expressed per meter. The current water depth is in meters.

[0064] Step S503 involves applying the correction coefficient and compensation coefficient to the preprocessing stage of the original spectral data. The effective optical path correction coefficient obtained in step S501 and the compensation coefficient calculated in step S502 are applied to the preprocessing stage of the original spectral data. The preprocessing stage includes spectral intensity normalization, spectral waveform smoothing, and baseline correction operations.

[0065] The specific implementation of spectral intensity normalization is as follows: select a reference wavelength in the original spectral data that is not affected by changes in water quality parameters, use the spectral intensity value at the reference wavelength as the normalization benchmark, divide the spectral intensity value at each wavelength position by the benchmark value, and eliminate the influence of light source intensity fluctuations and differences in optical path coupling efficiency.

[0066] The smoothing operation of the spectral waveform is implemented using the moving average method. Specifically, for each data point in the spectral data sequence, adjacent data points within a preset window width are taken with that point as the center, and the arithmetic mean of these data points is calculated as the smoothed value of that point. This process is repeated for all data points to complete the smoothing process, which is used to eliminate high-frequency random noise.

[0067] The baseline correction operation is achieved by polynomial fitting baseline subtraction method. Specifically, the wavelength ranges in the spectral data that are not occupied by the absorption peaks of the analyte are identified, and low-order polynomial fitting is performed on the data points in these ranges to obtain the baseline curve. Then, the baseline curve is subtracted from the original spectral data to eliminate baseline drift caused by factors such as water scattering and instrument dark current.

[0068] After the above three preprocessing operations are completed, the processed spectral data is multiplied by the effective optical path correction factor and the compensation factor to generate corrected spectral data, which is used as input for the subsequent water quality parameter inversion step.

[0069] In summary, steps S501 to S503 sequentially complete the acquisition of the effective optical path correction coefficient, the establishment of the light collection efficiency attenuation curve model and the calculation of the compensation coefficient, and the application of the correction parameter in spectral preprocessing, forming a complete link from coefficient acquisition to correction execution.

[0070] The next step, S6, establishes a quantitative mapping model between the corrected spectral data and water quality parameters. Step S6 takes the corrected spectral data output from step S5 and establishes a quantitative mapping relationship between the spectral data and water quality parameters such as dissolved oxygen concentration, turbidity, and chlorophyll content, providing a data inversion basis for subsequent water quality monitoring and early warning. Steps S601 to S604 are described in detail below.

[0071] Step S601: Collect water quality samples and obtain corresponding spectral data. Collect water quality samples from reservoirs with multiple known water quality parameter standard values. Water quality parameter standard values ​​refer to the accurate values ​​of water quality indicators such as dissolved oxygen concentration, turbidity, and chlorophyll content obtained through standard chemical analysis methods or standard instrument measurement methods, serving as true values ​​for subsequent model training. For each water quality sample, a spectrometer is used to collect spectral data in the visible to near-infrared band to obtain the raw spectral data corresponding to the water quality parameter standard values.

[0072] Step S602: Preprocess the spectral data and extract spectral features. The original spectral data corresponding to each water quality sample obtained in step S601 are processed according to the effective optical path correction and light collection efficiency compensation method described in step S5 to obtain the corrected spectral data.

[0073] Spectral features are extracted from the corrected spectral data. These features include: spectral peak positions (the wavelengths corresponding to the local maximum intensity values ​​in the spectral curve); spectral integral area (the area enclosed by the spectral curve and the wavelength axis within a specific wavelength range); spectral derivative features (the first or second derivative of the spectral curve with respect to wavelength, used to characterize the rate of change of the spectral curve); and spectral principal component scores (the principal component projection values ​​obtained after dimensionality reduction of the spectral data through principal component analysis). These spectral features collectively constitute a feature vector describing the spectral characteristics of each water quality sample.

[0074] Step S603: Construct a quantitative mapping model between the corrected spectral data features and water quality parameters. Use regression analysis algorithm or neural network algorithm to establish a quantitative mapping model between the corrected spectral data features and dissolved oxygen concentration, turbidity value, and chlorophyll content.

[0075] For regression analysis algorithms, there are partial least squares regression, support vector regression, and random forest regression. Taking partial least squares regression as an example, its modeling process is as follows: the spectral feature vector extracted in step S602 is used as the independent variable matrix, and the corresponding water quality parameter standard values ​​are used as the dependent variable vector. By extracting the direction of maximum covariance between the independent variable matrix and the dependent variable vector, a linear regression model in a low-dimensional latent variable space is established.

[0076] Neural network algorithms include multilayer perceptrons, convolutional neural networks, and long short-term memory networks. We will use a multilayer perceptron as an example. A multilayer perceptron consists of an input layer, hidden layers, and an output layer. The number of nodes in the input layer is equal to the dimension of the spectral feature vector in step S602. The number of nodes in the output layer is equal to the number of water quality parameter types to be predicted. The hidden layer contains at least one layer, and each layer contains several nodes. Nodes in adjacent layers are fully connected through weight coefficients.

[0077] The training process of the multilayer perceptron is as follows: the spectral feature vector extracted in step S602 is used as the model input data, and the standard values ​​of water quality parameters obtained in step S601 are used as the model output targets, thus forming a training sample set. The loss function of the multilayer perceptron adopts the mean squared error loss function, the expression of which is:

[0078] in, The value is the loss function value, which is dimensionless. The total number of training samples is a positive integer. The sequence number of the training sample; The total number of water quality parameters, which is a positive integer; This refers to the type number of the water quality parameter; For the first In the training samples, the th Standard values ​​for various water quality parameters, in units of measurement for each parameter; For multilayer perceptrons targeting the first In the training samples, the th The predicted values ​​of the water quality parameters are output, in units of measurement corresponding to each parameter. The training objective is to continuously adjust the weight coefficients and bias parameters of each layer in the multilayer perceptron using the backpropagation algorithm, so that the loss function... Minimize the value of , even if the model predicts . Approximate the true value as closely as possible .

[0079] After the above training process, the multilayer perceptron establishes a mapping relationship from the input spectral feature vector to the output water quality parameter prediction value.

[0080] Step S604: The prediction accuracy of the mapping model is evaluated using a validation dataset, independent of the training samples. The samples in the validation dataset also contain spectral feature vectors and their corresponding standard values ​​of water quality parameters, but these samples were not involved in the model training process in step S603. The spectral feature vectors from the validation dataset are input into the mapping model to obtain the predicted values ​​of water quality parameters output by the model. The predicted values ​​are compared with the standard values ​​of water quality parameters in the validation dataset to calculate the prediction error. The model parameters are adjusted based on the validation results until the prediction accuracy meets the preset requirements.

[0081] Step S6 involves collecting reservoir water samples with known water quality parameter standard values, obtaining corrected spectral data, extracting spectral features, and establishing a quantitative mapping model between spectral features and water quality parameters using regression analysis or neural network algorithms. The model's accuracy is then evaluated and parameters are adjusted using an independent validation dataset. The trained mapping model output from Step S6 will serve as the core calculation for real-time water quality monitoring data inversion and early warning output in Step S7.

[0082] Finally, in step S7, water quality monitoring results are generated and early warning judgments and outputs are performed. The real-time collected spectral data is converted into predicted values ​​of water quality parameters, and anomaly judgments and early warning outputs are made based on the prediction results, thereby realizing the real-time monitoring function of reservoir water quality. The following steps S701 to S704 are described in detail.

[0083] Step S701: Input the real-time collected raw spectral data into the mapping model to obtain predicted water quality parameters. Input the real-time collected raw spectral data into the mapping model trained and validated in step S6. The mapping model outputs the corresponding predicted water quality parameters. The predicted water quality parameters include predicted dissolved oxygen concentration, predicted turbidity, and predicted chlorophyll content.

[0084] The prediction process is executed using either batch processing or real-time streaming. Batch processing involves inputting multiple accumulated spectral data samples within the current time window into the mapping model for batch prediction at preset time intervals, such as every 10 minutes or every 30 minutes. Real-time streaming, on the other hand, inputs a new spectral data sample into the mapping model immediately upon acquisition, performing real-time prediction and outputting the predicted water quality parameters for the current moment.

[0085] Step S702 involves comparing the predicted water quality parameters obtained in step S701 with the preset safety threshold range. The safety threshold range is preset based on factors such as the reservoir's function and purpose, water quality standards, and historical monitoring data. The safety threshold range includes the lower limit threshold for dissolved oxygen concentration, the upper limit threshold for turbidity, and the warning threshold for chlorophyll content.

[0086] The lower limit threshold for dissolved oxygen concentration refers to the minimum dissolved oxygen concentration required to maintain the ecological health of a reservoir. A concentration below this value indicates a risk of oxygen deficiency. The upper limit threshold for turbidity refers to the maximum permissible turbidity level in a reservoir. Exceeding this value indicates a risk of excessive suspended solids. The warning threshold for chlorophyll content refers to the maximum permissible chlorophyll content in a reservoir. Exceeding this value indicates a risk of excessive algal blooms.

[0087] Step S703: Determine whether there is an abnormal water quality condition based on a multi-parameter joint discrimination strategy. The comparison analysis results from step S702 are comprehensively judged according to the multi-parameter joint discrimination strategy. The specific judgment logic of the multi-parameter joint discrimination strategy is as follows: when the predicted dissolved oxygen concentration is lower than the lower limit threshold, it is judged as a first-type anomaly; when the predicted turbidity value exceeds the upper limit threshold, it is judged as a second-type anomaly; when the predicted chlorophyll content exceeds the warning threshold, it is judged as a third-type anomaly; when any one or more of the above three types of anomalies are triggered, it is determined that there is an abnormal water quality condition in the reservoir.

[0088] The multi-parameter joint discrimination strategy also includes false alarm suppression logic: when only a single water quality parameter's predicted value exceeds the safety threshold range, and the magnitude of the exceedance is less than a preset tolerance value, anomaly detection is not triggered to prevent frequent false alarms caused by instantaneous fluctuations of a single parameter. When the predicted values ​​of any two or more water quality parameters simultaneously exceed their respective safety threshold ranges, anomaly detection is triggered directly, and the tolerance value is no longer considered. The specific value of the tolerance value is preset based on the fluctuation statistical characteristics of the actual monitoring data of the reservoir.

[0089] Step S704: Generate an early warning report and send it to the early warning terminal device via a wireless communication link. An early warning report is generated when step S703 determines that there is an abnormal condition in the reservoir water quality. The early warning report includes information on the type of abnormality, the degree of abnormality, the location of occurrence, and the time. The type of abnormality refers to the type of water quality parameter that triggered the abnormality determination, including abnormal dissolved oxygen, abnormal turbidity, and abnormal chlorophyll. The degree of abnormality refers to the magnitude of the exceedance of the safety threshold range, divided into two levels: slight exceedance and severe exceedance. The location of occurrence refers to the location information of the reservoir area where the sensor that detected the water quality abnormality is located. The time information refers to the data collection time corresponding to this abnormality determination.

[0090] The data format of the early warning report conforms to the unified alarm protocol standard, which specifies the name, data type, value range and encoding rules of each field in the early warning report.

[0091] After an early warning report is generated, it is sent to the early warning terminal equipment at the monitoring center via a wireless communication link. The wireless communication link uses a 4G, 5G, or LoRa communication module. The early warning terminal equipment includes a monitoring center server, mobile terminals for on-duty personnel, and on-site alarms. Early warning reports can be sent via SMS, mobile application push notifications, and email notifications.

[0092] In summary, step S7 obtains predicted water quality parameters by inputting the real-time collected spectral data into the trained mapping model. The predicted values ​​are then compared with preset safety thresholds, and a multi-parameter joint discrimination strategy is used to determine anomalies. Upon determining an anomaly, a warning report containing information on the anomaly type, severity, location, and time is generated and transmitted to the warning terminal equipment at the monitoring center via a wireless communication link. The warning report output in step S7 provides direct evidence for reservoir management personnel's emergency response decisions, thus completing a full monitoring and warning closed loop from data acquisition, optical correction, parameter inversion to warning output.

[0093] Based on the above method, this embodiment also provides an intelligent water quality monitoring and early warning system for reservoir environments. The system includes a temperature profile acquisition module, a dynamic update module, a refractive index analysis module, a physical reconstruction module, an optical path correction module, a parameter mapping module, and an early warning output module.

[0094] The temperature profile acquisition module is used to deploy multi-depth water temperature sensor arrays at different depths in the vertical direction of the reservoir to acquire vertical temperature profile data of the reservoir.

[0095] The dynamic update module is used to periodically collect water temperature data at various depths based on a multi-depth water temperature sensor array, construct a real-time updated vertical temperature profile dataset of the reservoir, and dynamically adjust the data collection update frequency according to the seasonal water temperature change rate.

[0096] The refractive index analysis module is used to establish a model of the correspondence between water temperature and optical refractive index of water body based on the vertical temperature profile data of reservoir, analyze the distribution characteristics of optical refractive index gradient of different water temperature layers, and quantify the nonlinear mathematical relationship between refractive index and water temperature.

[0097] The physical reconstruction module is used to construct a physical reconstruction model of optical distortion of water body based on the correspondence model between water temperature and optical refractive index of water body and the optical refractive index gradient distribution characteristics of different water temperature layers, so as to realize the simulation and calculation of optical path distortion and path bending.

[0098] The optical path correction module is used to calculate the effective optical path correction coefficient of each water layer based on the physical reconstruction model of optical distortion in water, to compensate for the light attenuation caused by optical path distortion, and to correct the physical attenuation characteristic that the light collection efficiency decreases with increasing water depth.

[0099] The parameter mapping module is used to establish a quantitative mapping relationship between the optically distorted spectral data and standard water quality parameters, and to use machine learning models or physical empirical models to retrieve water quality parameters such as dissolved oxygen concentration, turbidity, and chlorophyll content.

[0100] The early warning output module is used to generate a water quality monitoring report based on the corrected spectral inversion results. When the water quality parameters exceed the preset safety threshold, an early warning signal is triggered, realizing real-time monitoring and early warning of abnormal water quality in the reservoir.

[0101] To facilitate understanding, the technical solution of this embodiment is described in detail below with reference to a specific application example. The intelligent water quality monitoring and early warning system of this invention is deployed in the deep-water monitoring area of ​​a large reservoir. The multi-depth water temperature sensor array includes a first type of temperature sensor group arranged in the surface area, a second type of temperature sensor group in the thermocline area, and a third type of temperature sensor group in the deep-water area. Each sensor group is distributed according to a preset depth spacing, with a total of multiple sensor nodes, covering a complete vertical profile from the water surface to the bottom of the reservoir. During system operation, the dynamic update module adaptively adjusts the data acquisition frequency based on the real-time monitored water temperature changes. Under strong sunlight conditions in summer, the surface water temperature of the reservoir rises rapidly, and the depth of the thermocline changes significantly. At this time, the dynamic update module automatically shortens the data acquisition cycle and increases the update frequency of the temperature profile data to capture the dynamic migration of the thermocline position.

[0102] The raw temperature data acquired by the temperature profile acquisition module is preprocessed by a signal conditioning circuit, including zero-point correction, gain calibration, and temperature drift compensation. The preprocessed temperature data is stored in a standard data format and synchronously transmitted to a cloud server. Upon receiving the temperature profile data, the refractive index analysis module calculates the optical refractive index values ​​at each depth point according to a pre-built nonlinear mapping function and generates a refractive index distribution curve. The physical reconstruction module divides the water body into multiple sub-water layers based on the refractive index distribution curve and calculates the propagation path and refraction deflection angle of light within each sub-water layer. Assuming that during a monitoring process, the apparent optical path length at a specific depth below the water surface is a fixed value, while the actual effective optical path length is a different correction value due to the optical refractive index gradient distribution caused by water temperature stratification, the physical reconstruction model calculates the effective optical path length correction coefficient under this depth condition.

[0103] After obtaining the effective optical path correction coefficient, the optical path correction module calculates the compensation coefficient by combining it with the light collection efficiency attenuation curve model. The corrected spectral data eliminates the systematic bias caused by optical path distortion and decreased light collection efficiency. The parameter mapping module inputs the corrected spectral data features into a trained neural network mapping model, outputting predicted values ​​for dissolved oxygen concentration, turbidity, and chlorophyll content. Suppose that in a monitoring session, the predicted dissolved oxygen concentration output by the mapping model is lower than the preset safety lower limit threshold. At this time, the early warning output module determines that the water quality is abnormal, generates an early warning report, and sends it to the early warning terminal at the monitoring center via a wireless communication link. Upon receiving the early warning information, monitoring personnel can take appropriate measures to handle the situation.

[0104] On the other hand, this embodiment provides an intelligent water quality monitoring and early warning system for reservoir environments, including: The temperature profile acquisition module is used to deploy a multi-depth water temperature sensor array at different depths in the vertical direction of the reservoir to acquire vertical temperature profile data of the reservoir. The deployment of the multi-depth water temperature sensor array includes: acquiring the vertical temperature profile of the reservoir and calculating the average water temperature gradient change rate in each depth range; adaptively configuring the sensor arrangement spacing in each water layer area according to the water temperature gradient change rate, reducing the sensor arrangement spacing in areas with a large water temperature gradient change rate and increasing the sensor arrangement spacing in areas with a small water temperature gradient change rate. The dynamic update module is used to periodically collect water temperature data at various depth points based on a multi-depth water temperature sensor array, and dynamically adjust the data collection update frequency according to the water temperature change range to build a real-time updated reservoir vertical temperature profile dataset. The refractive index analysis module is used to establish a nonlinear mapping function between water temperature and water optical refractive index based on the reservoir vertical temperature profile dataset, and to calculate the optical refractive index values ​​at each depth point to generate a refractive index distribution. The physical reconstruction module is used to divide the water body into multiple optically homogeneous sub-water layers based on the refractive index distribution by comparing the rate of change of temperature gradient with a preset temperature gradient rate of change threshold. Based on Snell's law of refraction, it calculates the light deflection layer by layer and performs path integration to obtain the cumulative optical path length and cumulative path curvature angle, so as to construct a physical reconstruction model of water body optical distortion. The optical path correction module is used to establish an effective optical path correction coefficient based on the cumulative optical path length and the apparent optical path length under the assumption of homogeneous water medium, and to establish an exponential decay curve model of light collection efficiency with water depth to calculate the compensation coefficient. After performing spectral intensity normalization, spectral waveform smoothing and baseline correction operations on the original spectral data, the processed spectral data is multiplied by the effective optical path correction coefficient and the compensation coefficient to generate the corrected spectral data. The parameter mapping module is used to extract spectral features from the calibrated spectral data and to establish a quantitative mapping relationship between spectral features and standard water quality parameters using a machine learning model, so as to retrieve water quality parameters. The early warning output module is used to determine abnormal water quality conditions based on the inverted water quality parameters and a multi-parameter joint discrimination strategy including false alarm suppression logic. When an abnormality is determined, an early warning report containing information on the type, degree, location and time of the abnormality is generated and sent to the early warning terminal device through a wireless communication link.

[0105] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects.

[0106] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A smart water quality monitoring and early warning method for reservoir environments, characterized in that, include: Deploying a multi-depth water temperature sensor array to collect vertical temperature profile data of the reservoir; wherein, deploying the multi-depth water temperature sensor array includes: acquiring the vertical temperature profile of the reservoir and calculating the average water temperature gradient change rate in each depth interval, and adaptively configuring the sensor arrangement spacing in each water layer area according to the water temperature gradient change rate, reducing the sensor arrangement spacing in areas with a large water temperature gradient change rate, and increasing the sensor arrangement spacing in areas with a small water temperature gradient change rate. Based on the collected vertical temperature profile data, a nonlinear mapping function between water temperature values ​​and the optical refractive index of water was established, and the characteristics of the optical refractive index gradient distribution of different water temperature layers were analyzed. Based on the optical refractive index gradient distribution characteristics, the water body is divided into multiple optically homogeneous sub-water layers based on the temperature gradient change rate. The refraction deflection angle and the direction of the refracted light when the light passes through the interface of each sub-water layer are calculated layer by layer based on Snell's law of refraction. The cumulative optical path length is obtained by integrating the length of the straight-line propagation path of the light in all sub-water layers using the piecewise summation method. Based on the cumulative optical path length and the apparent optical path length assuming uniform water medium, an effective optical path correction coefficient is established, and an exponential decay curve model of light collection efficiency with water depth is established to calculate the compensation coefficient. After performing spectral intensity normalization, spectral waveform smoothing and baseline correction operations on the original spectral data, the processed spectral data is multiplied by the effective optical path correction coefficient and the compensation coefficient to generate corrected spectral data. A quantitative mapping model between the corrected spectral data and water quality parameters is established. The real-time collected raw spectral data is input into the quantitative mapping model to obtain the predicted values ​​of water quality parameters. A multi-parameter joint discrimination strategy is adopted to compare the predicted values ​​of the water quality parameters with the preset safety threshold range to determine whether there is an abnormality in the water quality and generate an early warning report. The multi-parameter joint discrimination strategy includes false alarm suppression logic: when only the predicted value of a single water quality parameter exceeds the safety threshold range and the extent of the exceedance is less than the preset tolerance value, the abnormality judgment is not triggered.

2. The intelligent water quality monitoring and early warning method for reservoir environments according to claim 1, characterized in that, When deploying a multi-depth water temperature sensor array, in the surface water temperature region 0 to 5 meters below the reservoir surface, a first type of temperature sensor group is arranged at a first preset interval of 1 meter; in the thermocline region 5 to 20 meters below the reservoir surface, a second type of temperature sensor group is arranged at a second preset interval of 0.5 meters, where the second preset interval is smaller than the first preset interval. In the deep water layer area from 20 meters below the surface of the reservoir to the bottom of the reservoir, a third type of temperature sensor group is arranged at a third preset interval of 2 meters, wherein the third preset interval is greater than the first preset interval.

3. The intelligent water quality monitoring and early warning method for reservoir environments according to claim 1, characterized in that, When collecting vertical temperature profile data of the reservoir, an initial collection period is set, and the water temperature change amplitude between two adjacent collection periods is continuously monitored. When the water temperature change amplitude exceeds a preset change threshold, the collection period is automatically shortened from the initial collection period to a dynamic collection period. When the water temperature change amplitude is lower than the preset change threshold in multiple consecutive collection periods, the collection period is automatically restored to the initial collection period.

4. The intelligent water quality monitoring and early warning method for reservoir environments according to claim 1, characterized in that, The water body is divided into multiple optically homogeneous sub-water layers based on the temperature gradient change rate. Specifically, this includes: dividing the temperature difference between any two adjacent depth points by the depth difference to obtain the temperature gradient change rate; when the temperature gradient change rate is lower than a preset temperature gradient change rate threshold, merging adjacent depth points into the same sub-water layer, with the same sub-water layer corresponding to a constant optical refractive index value; when the temperature gradient change rate is higher than the preset temperature gradient change rate threshold, delineating a new sub-water layer boundary at the location where the temperature gradient change rate is higher than the preset temperature gradient change rate threshold, thus dividing the upper and lower parts into different sub-water layers.

5. The intelligent water quality monitoring and early warning method for reservoir environments according to claim 4, characterized in that, Establishing a nonlinear mapping function between water temperature values ​​and the optical refractive index of the water body is performed before dividing the water body into multiple optically homogeneous sub-water layers. Specifically, this includes: within the range of possible water temperature changes in the reservoir, experimentally calibrating and measuring the optical refractive index values ​​of the water body corresponding to multiple temperature sampling points to obtain multiple sets of temperature-refractive index discrete data points. The multiple sets of temperature-refractive index discrete data points are fitted using a polynomial fitting method or a Gaussian process regression method to obtain a continuous nonlinear mapping function between water temperature values ​​and the optical refractive index of the water body, which is used to calculate the constant optical refractive index values ​​corresponding to each of the sub-water layers.

6. The intelligent water quality monitoring and early warning method for reservoir environments according to claim 1, characterized in that, Establish an effective optical path correction coefficient, specifically including: for multiple different water depth conditions, calculate the effective optical path value, apparent optical path value and cumulative path curvature angle under each water depth condition to form a training dataset; Using apparent optical path length and cumulative path curvature angle as inputs and effective optical path length as output, a nonlinear regression method is used for fitting to obtain the effective optical path length correction coefficient; wherein, the cumulative path curvature angle is obtained by superimposing the refraction deflection angle of light at each sub-water layer interface one by one.

7. The intelligent water quality monitoring and early warning method for reservoir environments according to claim 1, characterized in that, To establish an exponential decay curve model of light collection efficiency as a function of water depth and to calculate the compensation coefficient, the following steps were taken: in a standard water body with known clear water quality and no suspended particulate matter, the actual light collection efficiency values ​​were measured at multiple different water depths to obtain multiple sets of discrete data points of water depth and light collection efficiency. Using the exponential decay model To determine the light collection efficiency attenuation coefficient, the least squares fitting method was used to fit the discrete data points of water depth and light collection efficiency to the multiple sets of data. The optimal estimate, where water depth Light collection efficiency at the location, This represents the theoretical maximum light collection efficiency at the water surface. Because of the water depth, The light collection efficiency attenuation coefficient is used; and the compensation coefficient is calculated using inverse function mapping. The calculation formula is: .

8. The intelligent water quality monitoring and early warning method for reservoir environments according to claim 1, characterized in that, Establishing a quantitative mapping model between the corrected spectral data and water quality parameters specifically includes: extracting spectral features from the corrected spectral data, wherein the spectral features include spectral peak position, spectral integral area, spectral derivative features, and spectral principal component scores, and the spectral peak position, spectral integral area, spectral derivative features, and spectral principal component scores together constitute a feature vector describing the spectral characteristics; using a multilayer perceptron neural network, with the feature vector as the model input, and standard water quality parameter values ​​including dissolved oxygen concentration, turbidity, and chlorophyll content as the model output targets, the quantitative mapping model is trained to obtain the model.

9. The intelligent water quality monitoring and early warning method for reservoir environments according to claim 1, characterized in that, The generated early warning report specifically includes: when it is determined that there is an abnormal water quality, the generated early warning report includes the type of abnormality determined according to the multi-parameter joint discrimination strategy, the degree of abnormality corresponding to the amplitude level exceeding the safety threshold range, the location information of the reservoir area where the sensor that detected the water quality abnormality is located, and the collection time corresponding to the abnormality determination. The data format of the early warning report conforms to the unified alarm protocol standard and is sent to the early warning terminal equipment in the monitoring center via a wireless communication link.

10. An intelligent water quality monitoring and early warning system for reservoir environments, applied to the method described in any one of claims 1-9, characterized in that, include: The temperature profile acquisition module is used to deploy a multi-depth water temperature sensor array at different depths in the vertical direction of the reservoir to acquire vertical temperature profile data of the reservoir. The deployment of the multi-depth water temperature sensor array includes: acquiring the vertical temperature profile of the reservoir and calculating the average water temperature gradient change rate in each depth interval; adaptively configuring the sensor arrangement spacing in each water layer area according to the water temperature gradient change rate, reducing the sensor arrangement spacing in areas with a large water temperature gradient change rate and increasing the sensor arrangement spacing in areas with a small water temperature gradient change rate. The dynamic update module is used to periodically collect water temperature data at each depth point based on the multi-depth water temperature sensor array, and dynamically adjust the data collection update frequency according to the water temperature change range to construct a real-time updated reservoir vertical temperature profile dataset. The refractive index analysis module is used to establish a nonlinear mapping function between water temperature and water optical refractive index based on the vertical temperature profile dataset of the reservoir, and to calculate the optical refractive index values ​​at each depth point to generate a refractive index distribution. The physical reconstruction module is used to divide the water body into multiple optically uniform sub-water layers based on the refractive index distribution by comparing the temperature gradient change rate with a preset temperature gradient change rate threshold, and to calculate the light deflection layer by layer based on Snell's law of refraction and perform path integration to obtain the cumulative optical path length and cumulative path curvature angle, so as to construct a physical reconstruction model of water body optical distortion. The optical path correction module is used to establish an effective optical path correction coefficient based on the cumulative optical path length and the apparent optical path length when the water medium is assumed to be uniform, and to establish an exponential decay curve model of light collection efficiency with water depth to calculate the compensation coefficient. After performing spectral intensity normalization, spectral waveform smoothing and baseline correction operations on the original spectral data, the processed spectral data is multiplied by the effective optical path correction coefficient and the compensation coefficient to generate corrected spectral data. The parameter mapping module is used to extract spectral features from the corrected spectral data and use a machine learning model to establish a quantitative mapping relationship between the spectral features and standard water quality parameters in order to retrieve water quality parameters. The early warning output module is used to determine abnormal water quality conditions based on the inverted water quality parameters and a multi-parameter joint discrimination strategy including false alarm suppression logic. When an abnormality is determined, an early warning report containing information on the type, degree, location and time of the abnormality is generated and sent to the early warning terminal device through a wireless communication link.

Images