Aquatic product water quality multi-parameter real-time on-line monitoring system

Abstract

The present application relates to the technical field of aquaculture monitoring and regulation, and discloses a water quality multi-parameter real-time online monitoring system. The system comprises a multi-source data acquisition module for obtaining and denoising water quality data; a time sequence feature analysis module for generating multi-dimensional feature clusters containing the correlation of various parameters; an anomaly identification module for dynamic deviation compensation, monitoring sub-domain demarcation and key parameter extraction; a collaborative regulation module for constructing a regulation decision unit to generate a multi-parameter collaborative regulation strategy; and an iterative correction module for correcting the regulation strategy to output real-time adjustment instructions. An early warning module is also provided for biological safety assessment and early warning and activation of the emergency intervention mechanism. The system can accurately monitor the multi-parameters of the water quality of aquaculture in real time, realize intelligent regulation and biological safety early warning, effectively improve the yield and quality of aquaculture, reduce the risk of aquaculture, and promote the sustainable development of the industry.

Description

Technical Field

[0001] This invention relates to the field of aquaculture monitoring and control technology, specifically to a real-time online monitoring system for multiple parameters of aquatic water quality. Background Technology

[0002] Against the backdrop of the booming development of the modern aquaculture industry, water quality management plays a decisive role in the success of aquaculture. The water quality in aquaculture environments is extremely complex, encompassing various parameters such as dissolved oxygen concentration, temperature, pH, and pollutant levels. These parameters are constantly changing and interacting with each other. Abnormal fluctuations in any of these parameters can disrupt the ecological balance of the aquaculture water, posing a serious threat to the healthy growth of aquatic organisms and even leading to large-scale disease outbreaks, resulting in significant economic losses for aquaculture farmers.

[0003] Traditional water quality monitoring methods primarily rely on manual sampling and laboratory testing, which have numerous drawbacks. Manual sampling is not only costly in terms of manpower, resources, and time, but also has a low sampling frequency, making it impossible to capture real-time changes in water quality. For example, in some large-scale aquaculture farms, manual sampling may take several days to complete a full cycle, during which time significant changes in water quality may have already occurred. Laboratory testing is complex, and the process from sampling to obtaining test results often takes a long time, preventing farmers from taking timely control measures based on the results. By the time water quality problems are detected, the optimal adjustment period may have already passed, leading to damage to aquaculture organisms.

[0004] With the development of sensor technology, some simple water quality monitoring devices have begun to be applied in aquaculture. However, most of these devices can only monitor a single or a few water quality parameters, failing to comprehensively reflect the overall water quality status. Furthermore, these devices have significant shortcomings in data processing and analysis capabilities, only providing raw monitoring data and unable to conduct in-depth data mining and analysis, making it difficult to help aquaculture farmers accurately judge water quality trends and potential risks. For example, a single dissolved oxygen monitoring device can only display the current dissolved oxygen level; farmers cannot know the trend of this value or its correlation with other water quality parameters, thus making it impossible to take targeted measures to optimize water quality.

[0005] In terms of water quality control, most fish farmers currently rely primarily on experience. For example, when they find the water turbid, they may blindly change the water without fully considering the impact of the water change on other factors such as the microbial community and pH level. This unscientific approach not only fails to achieve the desired water quality control results but may also lead to water waste and increased aquaculture costs. Furthermore, due to a lack of in-depth understanding of the complex relationships between water quality parameters, fish farmers are often at a loss when faced with multiple water quality problems simultaneously, and are unable to formulate effective comprehensive control strategies.

[0006] Furthermore, biosecurity issues are receiving increasing attention in aquaculture. Pathogens and harmful substances in the aquaculture water can cause diseases in farmed organisms, affecting the quality and safety of farmed products. However, existing monitoring and control technologies are severely inadequate in biosecurity assessment and early warning, failing to promptly identify potential biosecurity risks and implement effective preventative measures. Summary of the Invention

[0007] The purpose of this invention is to provide a real-time online monitoring system for multiple parameters of aquatic water quality to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a real-time online monitoring system for multiple parameters of aquatic water quality, the system comprising:

[0009] The multi-source data acquisition module is used to acquire a set of water quality parameters within a preset time span in the aquaculture environment, and to perform noise suppression processing on the set of water quality parameters according to a preset multimodal denoising strategy to generate denoised water quality data.

[0010] The time-series feature parsing module is used to embed the denoised water quality data into a dynamic time-series feature space and generate a multi-dimensional feature cluster based on a time-series segmented adaptive clustering algorithm. The multi-dimensional feature cluster contains the nonlinear correlation between dissolved oxygen concentration, temperature gradient and pollutant distribution.

[0011] The anomaly identification module is used to perform dynamic deviation compensation on the multidimensional feature clusters according to the preset fluctuation tolerance rules, delineate the real-time monitoring subdomains through the fuzzy logic decision tree algorithm, and extract the key parameter set in each subdomain.

[0012] The collaborative regulation module is used to construct a regulation decision unit containing a dynamic weight allocation network, and to perform gradient optimization and update of the dynamic weight allocation network using the set of key parameters to generate a multi-parameter collaborative regulation strategy.

[0013] The iterative correction module is used to perform multi-scale data fusion correction on the multi-parameter collaborative control strategy according to the preset local-global coupling analysis model, and output the real-time adjustment command of the entire aquatic environment.

[0014] Preferably, the generation of multidimensional feature clusters based on the time series segmentation adaptive clustering algorithm includes:

[0015] Extract the timestamp, monitoring point number, and parameter type from the denoised water quality data to construct a time-series tensor matrix;

[0016] The time series tensor matrix is ​​segmented using an overlapping window truncation algorithm to generate a set of time series segments.

[0017] A density-based outlier detection model is used to remove outliers from the time series segment set, retaining valid data segments;

[0018] The effective data fragments are analyzed in multiple dimensions using a spectral clustering algorithm to generate feature clusters with spatiotemporal continuity.

[0019] Preferably, the dynamic deviation compensation includes:

[0020] Based on the distribution of historical water quality data in the real-time monitoring subdomains, the moving average and standard deviation of key parameters in each subdomain are calculated.

[0021] Based on the adaptive moving average algorithm, the moving average and standard deviation are corrected by Kalman filtering to generate a dynamic benchmark curve.

[0022] The dynamic baseline curve is extended with fuzzy boundaries using an interval membership function to generate an adaptive compensation interval.

[0023] Preferably, the fuzzy logic decision tree algorithm delineates the real-time monitoring subdomain, including:

[0024] The decision objective is defined as the balanced optimization of parameter fluctuations and equipment energy consumption within the subdomain;

[0025] The constraints are set as sensor distribution density, environmental carrying capacity threshold, and biological activity cycle range.

[0026] The decision objective is solved using a heuristic search algorithm, and the optimal monitoring subdomain partitioning scheme is output.

[0027] Preferably, the dynamic weight allocation network performs gradient optimization updates, including:

[0028] The set of key parameters is input to the input node of the dynamic weight allocation network, and the temporal dependency features are extracted using a bidirectional recurrent neural network.

[0029] The temporal dependency features are prioritized using a gating mechanism to generate a parameter weight sequence.

[0030] The network connection weights are optimized using a combination of stochastic gradient descent and particle swarm optimization, and the control parameters of the control decision unit are updated accordingly.

[0031] Preferably, the multimodal denoising strategy includes:

[0032] Identify abnormal sampling points in the set of water quality parameters and perform interpolation replacement based on the time continuity criterion;

[0033] High-frequency noise components are detected and suppressed using wavelet packet thresholding.

[0034] The processed data is normalized to generate standard water quality data with values ​​ranging from [0,1].

[0035] Preferably, the local-global coupling analysis model includes:

[0036] A multi-factor correlation analysis framework based on the Morris screening method was constructed to quantify the coupling influence of adjustment commands on various water quality indicators.

[0037] A parameter perturbation sample set is generated by Latin hypercube sampling, and the interaction effect coefficients of each index are calculated.

[0038] Indicators with interaction effect coefficients higher than a preset critical value were selected as core variables for adjustment and regulation.

[0039] Preferably, the multi-scale data fusion correction includes:

[0040] By integrating sensor measured data, hydrological model prediction data, and aquaculture feeding records, a heterogeneous data correlation matrix is ​​constructed.

[0041] The heterogeneous data association matrix is ​​subjected to feature extraction using a nonnegative matrix factorization algorithm to obtain low-dimensional feature basis vectors.

[0042] The low-dimensional feature basis vector is input into the iterative correction module to generate the corrected adjustment command.

[0043] Preferably, the system further includes:

[0044] An early warning module containing a biosafety assessment function is constructed, which matches and analyzes the real-time adjustment instructions with biosafety thresholds and outputs multi-level early warning indicators;

[0045] The preset emergency intervention mechanism is activated based on the multi-level early warning indicators, generating instructions for starting / stopping oxygenation or water exchange.

[0046] Preferably, the present invention further includes an electronic device, the device comprising:

[0047] At least one processor; and

[0048] A memory communicatively connected to the at least one processor; wherein,

[0049] The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the operation of the aforementioned real-time online monitoring system for multiple parameters of aquatic water quality.

[0050] Compared with the prior art, the beneficial effects of the present invention are:

[0051] From a data acquisition and processing perspective, the multi-source data acquisition module can obtain a set of water quality parameters within a preset time span and effectively suppress noise through a multi-modal denoising strategy. This denoising strategy can accurately identify abnormal sampling points, perform interpolation replacement based on the time continuity criterion, and simultaneously use wavelet packet thresholding to suppress high-frequency noise components. Finally, it performs normalization transformation to generate standard water quality data. This makes the acquired data more accurate and reliable, providing a solid foundation for subsequent analysis and decision-making. Compared with traditional monitoring methods, this system can collect data in real time and at high frequency, avoiding the delays and errors of manual sampling, allowing aquaculture farmers to promptly grasp the dynamic changes in water quality.

[0052] The temporal feature analysis module embeds the denoised water quality data into a dynamic temporal feature space, and uses a time-series segmented adaptive clustering algorithm to generate multi-dimensional feature clusters, including nonlinear correlations between dissolved oxygen concentration, temperature gradient, and pollutant distribution. This process is achieved through constructing a time-series tensor matrix, piecewise processing, outlier removal, and spectral clustering algorithms, uncovering the complex relationships hidden between water quality parameters and contributing to a deeper understanding of the intrinsic patterns of water quality changes. Aquaculture farmers can use these patterns to predict water quality trends in advance and take corresponding preventative measures to reduce aquaculture risks.

[0053] In terms of anomaly identification, the anomaly identification module dynamically compensates for deviations in multi-dimensional feature clusters based on preset fluctuation tolerance rules. It then uses a fuzzy logic decision tree algorithm to delineate real-time monitoring subdomains and extract key parameter sets. Dynamic deviation compensation generates adaptive compensation intervals by calculating moving averages and standard deviations, Kalman filtering corrections, and fuzzy boundary expansion, enabling more accurate identification of water quality anomalies. The fuzzy logic decision tree algorithm aims to balance parameter fluctuations within the subdomain with equipment energy consumption, combining constraints such as sensor distribution density, environmental carrying capacity thresholds, and biological activity cycle ranges to solve for the optimal monitoring subdomain partitioning scheme. This approach not only enables timely detection of water quality anomalies but also reasonably reduces equipment energy consumption and improves monitoring efficiency while ensuring monitoring accuracy.

[0054] The collaborative regulation module constructs a regulation decision unit containing a dynamic weight allocation network. It utilizes a set of key parameters to perform gradient optimization and updates on the network, generating a multi-parameter collaborative regulation strategy. Temporal-dependent features are extracted using a bidirectional recurrent neural network, priority is ranked using a gating mechanism, and network connection weights are optimized using a combination of stochastic gradient descent and particle swarm optimization, achieving collaborative regulation of multiple water quality parameters. Unlike traditional experience-based regulation methods, this system can scientifically and rationally adjust water quality parameters based on real-time monitoring data, ensuring the aquaculture water body is in optimal condition, reducing diseases and mortality in aquaculture organisms caused by water quality issues, and improving both yield and quality.

[0055] The iterative correction module, based on a pre-defined local-global coupling analysis model, performs multi-scale data fusion correction on the multi-parameter collaborative control strategy, outputting real-time regulation commands for the entire aquatic environment. By constructing a multi-factor correlation analysis framework based on the Morris screening method, generating a parameter perturbation sample set through Latin hypercube sampling, and calculating interaction effect coefficients, the core variables for correction and regulation are selected, achieving precise optimization of the control strategy. The multi-scale data fusion correction integrates sensor measured data, hydrological model prediction data, and aquaculture feeding records, using a non-negative matrix factorization algorithm to extract low-dimensional feature basis vectors, generating more accurate regulation commands, further improving the accuracy and effectiveness of regulation.

[0056] The system also includes an early warning module, which constructs a biosafety assessment function to match and analyze real-time adjustment commands with biosafety thresholds, outputs multi-level early warning indicators, and activates a preset emergency intervention mechanism based on the early warning, generating instructions to start / stop oxygenation or water exchange. This function greatly enhances the biosafety assurance capabilities during the aquaculture process, enabling timely detection of potential biosafety risks, taking effective countermeasures, ensuring the healthy growth of farmed organisms, and improving the safety of farmed products.

[0057] Overall, the system organically combines data collection, analysis, control, and early warning, forming a complete closed-loop management system. By improving the accuracy and timeliness of water quality monitoring, optimizing water quality control strategies, and strengthening biosafety early warning and emergency response capabilities, it effectively reduces the risks of aquaculture, improves aquaculture efficiency and economic benefits, and promotes the sustainable development of the aquaculture industry. At the same time, the application of this system also helps to drive the aquaculture industry towards intelligent and scientific management, enhancing the competitiveness of the entire industry. Attached Figure Description

[0058] Figure 1 This is a schematic diagram illustrating the working principle of the real-time online monitoring system for multiple parameters of aquatic water quality described in this invention.

[0059] Figure 2The flowchart for dynamic deviation compensation;

[0060] Figure 3 A flowchart for multi-scale data fusion correction;

[0061] Figure 4 This is a flowchart for early warning and emergency intervention. Detailed Implementation

[0062] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0063] Please see Figures 1-4 This invention provides a real-time online monitoring system for multiple parameters of aquatic water quality, and its specific implementation is described in detail below.

[0064] Multi-source data acquisition module: In aquaculture environments, this module collects various water quality parameters over a preset time span, forming a water quality parameter set. These parameters cover crucial indicators for aquaculture, such as water temperature, dissolved oxygen concentration, pH value, ammonia nitrogen content, and pollutant distribution. Since the actual collected data may be affected by environmental noise, sensor errors, and other factors, the module performs noise suppression processing on the collected water quality parameter set according to a preset multimodal denoising strategy to ensure data accuracy. After denoising processing, the final denoised water quality data is generated, providing a reliable data foundation for subsequent analysis.

[0065] The temporal feature analysis module embeds the denoised water quality data generated by the multi-source data acquisition module into a dynamic temporal feature space. Within this space, a time-series segmented adaptive clustering algorithm is used for in-depth data analysis. This algorithm generates multi-dimensional feature clusters that not only contain information on key parameters such as dissolved oxygen concentration, temperature gradient, and pollutant distribution, but also uncover complex nonlinear correlations between them. These correlations are crucial for accurately grasping the trends and patterns of water quality changes.

[0066] Anomaly Identification Module: Based on preset fluctuation tolerance rules, this module dynamically compensates for deviations in the multi-dimensional feature clusters generated by the time-series feature analysis module. Specifically, it adjusts and corrects potential deviations based on changes in real-time monitoring data. Simultaneously, a fuzzy logic decision tree algorithm is used to delineate real-time monitoring sub-domains, dividing the entire monitoring area into multiple sub-regions according to different characteristics and needs. Within each sub-domain, a set of key parameters is extracted. These key parameters represent the main characteristics of water quality within that sub-domain, providing a core basis for subsequent control measures.

[0067] The collaborative regulation module constructs a regulation decision-making unit containing a dynamic weight allocation network. Using the key parameter set extracted by the anomaly detection module, the dynamic weight allocation network is updated through gradient optimization. During this process, the weights of each parameter in the network are dynamically adjusted based on their importance and real-time changes, thereby generating a multi-parameter collaborative regulation strategy. This strategy comprehensively considers changes in multiple water quality parameters, achieving collaborative regulation of water quality in the aquaculture environment and ensuring that water quality remains within a suitable range for aquaculture growth.

[0068] Iterative Correction Module: Based on a pre-defined local-global coupling analysis model, this module performs multi-scale data fusion correction on the multi-parameter collaborative control strategy generated by the collaborative control module. It comprehensively considers local and global water quality information, as well as data changes at different time and spatial scales, to optimize and improve the control strategy. Ultimately, it outputs real-time, comprehensive aquatic environment control commands. These commands can be directly used to control aeration equipment, water exchange equipment, feeding equipment, etc., achieving precise control of the aquaculture environment.

[0069] Example 1:

[0070] This embodiment details the process of generating multidimensional feature clusters based on a time-series segmented adaptive clustering algorithm. This process plays a crucial role in the entire aquatic water quality multi-parameter real-time online monitoring system for in-depth mining of water quality data characteristics and understanding of water quality change patterns.

[0071] Aquaculture environments are complex and variable, and a multi-source data acquisition module continuously collects water quality data from different monitoring points. After processing using a multimodal denoising strategy, denoised water quality data is obtained. When generating multidimensional feature clusters, the first step is to extract key information from the denoised water quality data and construct a time-series tensor matrix. It is assumed that the denoised water quality data contains three types of key information: timestamp, monitoring point number, and parameter type. Using the timestamp as... It represents the discrete time points from the start of data collection to the current moment, precisely recording the order in which data was collected, for example... This represents the time when the data was collected at 9:00 AM. This represents the data collection time at 9:10 AM; the monitoring point number is... Used to distinguish different monitoring locations within the aquaculture area. This likely refers to the monitoring point in the northeast corner of the pond. This represents the monitoring point at the center of the pond; the parameter type is... It covers various water quality parameters such as dissolved oxygen concentration, temperature, pH value, and ammonia nitrogen content. Corresponding dissolved oxygen concentration, Corresponding temperature. A time-series tensor matrix constructed based on this information. Its elements It can comprehensively and systematically store various water quality parameter data at each time point and each monitoring point.

[0072] The temporal tensor matrix is ​​pieced together using an overlapping window truncation algorithm. The size of the overlapping window is set to [value missing]. Step size is The overlapping window design aims to better preserve the temporal continuity of data and avoid losing important information due to window partitioning. The step size controls the interval between window movements, affecting the number of partitions and the data coverage. For example, if... (This indicates that the window contains data from 10 time points.) (This means the window moves 5 time points each time), from the time series tensor matrix Starting from the initial position, based on the window size Extract data segments, and move the step size each time. This generates a set of time-series segments. Each time-series segment contains data on various water quality parameters from multiple monitoring points within a specific time period. For example, a segment might contain data on dissolved oxygen concentration, temperature, and ammonia nitrogen content from three different monitoring points in the pond between 9:00 AM and 9:50 AM.

[0073] A density-based outlier detection model is used to analyze a collection of time-series segments. Outlier removal is performed. The model identifies outliers based on the density of data points within each time segment. Under normal circumstances, data points are distributed around a central trend, exhibiting a certain density. Outliers are those data points that differ significantly from their surrounding data points and have noticeably lower density. The model calculates the density around each data point and compares it to a set threshold. If the density of a data point is below the threshold, it is identified as an outlier and removed. This step removes abnormal data caused by sensor malfunctions, sudden environmental interference, etc., retaining valid data segments and resulting in a set of valid data segments. For example, in a set of data on dissolved oxygen concentration, most data fluctuate within a reasonable range, but one data point deviates significantly from this range. After being judged by a density-based outlier detection model, this data point will be identified as an outlier and removed.

[0074] Finally, the effective data fragment set is processed using the spectral clustering algorithm. Perform multi-dimensional correlation analysis to generate feature clusters with spatiotemporal continuity. Spectral clustering algorithms are based on graph theory, treating data points as nodes in a graph, with the similarity between nodes represented by edge weights. For a valid set of data segments... For each data point, its similarity to other data points is calculated to construct a similarity matrix. This similarity matrix is ​​then processed to obtain a Laplace matrix. By performing eigenvalue decomposition on the Laplace matrix, the data points can be divided into different clusters. In this embodiment, these clusters reflect complex relationships between different times, different monitoring points, and different water quality parameters. For example, within a certain time period, the dissolved oxygen concentration, temperature, and ammonia nitrogen content near a specific monitoring point may exhibit a certain pattern of coordinated change, which will be reflected in the characteristic clusters. By analyzing these characteristic clusters, aquaculture personnel can discover patterns in water quality changes and potential problems. For instance, if it is found that within a certain characteristic cluster, the dissolved oxygen concentration at several monitoring points continuously decreases over several consecutive time points, while the temperature and ammonia nitrogen content show an upward trend, this may indicate water quality deterioration in the area, requiring timely control measures, such as increasing the operating time of aeration equipment and adjusting the water exchange frequency, to ensure a healthy growth environment for aquaculture organisms.

[0075] Example 2:

[0076] This embodiment focuses on the specific implementation process of dynamic deviation compensation. In real-time monitoring of the aquaculture environment, after the anomaly identification module obtains relevant data from the real-time monitoring subdomain, it performs dynamic deviation compensation.

[0077] Based on the historical water quality data distribution of the real-time monitoring sub-domains, the moving average and standard deviation of key parameters within each sub-domain are calculated. Let the historical data sequence of a certain key parameter (such as dissolved oxygen concentration) within the real-time monitoring sub-domain be (The original text is incomplete and requires further context). The moving average window size is Then the moving average of this key parameter The calculation formula is:

[0078]

[0079] in, This indicates the current time point in the calculation of the moving average. Moving averages can reflect the average level of key parameters over a certain period of time.

[0080] At the same time, calculate the standard deviation of this key parameter. The calculation formula is:

[0081]

[0082] Standard deviation is used to measure the dispersion of key parameter data and reflects the fluctuation of the data.

[0083] Based on the adaptive moving average algorithm, Kalman filtering is applied to correct the moving average and standard deviation. Kalman filtering is an optimal linear recursive filtering algorithm that continuously adjusts the estimates based on the system's observed and predicted data, resulting in more accurate estimates. Through Kalman filtering correction, a dynamic benchmark curve that better reflects the actual situation can be obtained. In this process, the parameters of the Kalman filter are continuously updated according to changes in real-time monitoring data, thereby dynamically adjusting the moving average and standard deviation.

[0084] An adaptive compensation interval is generated by extending the dynamic baseline curve with fuzzy boundaries using an interval membership function. The interval membership function determines whether a key parameter deviates from the dynamic baseline curve and falls within the normal, warning, or abnormal range. By setting different membership thresholds, the dynamic baseline curve is extended upwards and downwards to form the adaptive compensation interval. When a key parameter monitored in real time exceeds this adaptive compensation interval, the system will issue a timely warning, prompting aquaculture personnel to take appropriate measures to ensure that the water quality is within the suitable range for aquaculture. For example, if the dissolved oxygen concentration exceeds the lower limit of the adaptive compensation interval, it may indicate hypoxia in the water, requiring immediate activation of aeration equipment; if it exceeds the upper limit, it may indicate over-aeration or other abnormalities, requiring further investigation and adjustment of aquaculture operations.

[0085] Example 3:

[0086] This embodiment details the specific implementation of the fuzzy logic decision tree algorithm for delineating real-time monitoring subdomains. In an aquatic water quality monitoring system, the anomaly identification module utilizes the fuzzy logic decision tree algorithm to divide the monitoring area.

[0087] The decision objective is defined as the balanced optimization of parameter fluctuations and equipment energy consumption within a subdomain. In aquaculture, it is crucial to ensure stable water quality parameters and minimize the impact of fluctuations on aquatic organisms. Simultaneously, it is essential to rationally control the operation of equipment such as aeration and water exchange systems to reduce energy consumption and improve aquaculture efficiency. Therefore, balancing these two objectives as the decision objective allows for the rational utilization of resources while ensuring water quality.

[0088] The constraints are sensor distribution density, environmental carrying capacity threshold, and biological activity cycle range. Sensor distribution density determines the accuracy and coverage of the acquired data. When dividing the monitoring subdomains, the distribution of sensors needs to be considered to ensure that there is sufficient data to support analysis in each subdomain. The environmental carrying capacity threshold refers to the range of water quality parameter variations that the aquaculture environment can tolerate; exceeding this range may harm aquaculture organisms. The biological activity cycle range reflects the different water quality requirements of aquaculture organisms at different growth stages. For example, during the fish breeding season, the requirements for dissolved oxygen concentration and pH are more stringent; while in the later stages of growth, the tolerance for ammonia nitrogen content may change. These constraints provide important limitations and guidance for the fuzzy logic decision tree algorithm.

[0089] A heuristic search algorithm is used to solve the decision objective and output the optimal monitoring subdomain partitioning scheme. A heuristic search algorithm is an algorithm that uses heuristic information from the problem to guide the search direction, enabling it to quickly find a near-optimal solution within the search space. In this embodiment, the heuristic search algorithm searches and evaluates different subdomain partitioning schemes based on the defined decision objective and set constraints. By continuously adjusting the boundaries and sizes of the subdomains, calculating the parameter fluctuations and equipment energy consumption under each scheme, and filtering based on the constraints, the optimal monitoring subdomain partitioning scheme is finally obtained. For example, in a certain aquaculture pond, based on the distribution of sensors, the pond is divided into multiple subdomains of different sizes. The heuristic search algorithm calculates the water quality parameter fluctuations and equipment energy consumption under each subdomain partitioning scheme. Considering the water flow conditions and biological distribution characteristics in different areas of the pond, the optimal subdomain partitioning scheme that ensures water quality stability while reducing equipment energy consumption is finally determined. Such a partitioning scheme helps to more accurately monitor and regulate the water quality in different areas, improving the overall efficiency of aquaculture.

[0090] Example 4:

[0091] This embodiment details the specific process of gradient optimization and updating of the dynamic weight allocation network. In the collaborative control module, the gradient optimization and updating of the dynamic weight allocation network is performed using a set of key parameters.

[0092] The set of key parameters extracted by the anomaly detection module is input into the input node of the dynamic weight allocation network. Let the set of key parameters be... These parameters contain crucial information for water quality control, such as dissolved oxygen concentration, temperature, and ammonia nitrogen content. A bidirectional recurrent neural network (Bi-RNN) is used to process the input set of key parameters to extract time-dependent features. The Bi-RNN can process sequential data simultaneously from both forward and backward directions, thus better capturing the time dependencies within the data. When processing the set of key parameters, the Bi-RNN learns the dynamic changes between parameters based on their chronological order in the time series, generating feature vectors rich in time-series information.

[0093] A gating mechanism is used to prioritize the extracted temporal dependency features, generating a parameter weight sequence. The gating mechanism can filter and weight different features based on their importance and relevance. In this embodiment, the gating mechanism calculates the importance score of each feature based on the temporal dependency features output by the bidirectional recurrent neural network, then sorts the features according to their scores and assigns corresponding weights to each key parameter. Let the generated parameter weight sequence be... ,in Indicates key parameters The greater the weight, the more important the parameter is in the regulation process.

[0094] This paper employs a combination of stochastic gradient descent (SGD) and particle swarm optimization (PSO) to optimize network connection weights and update the control parameters of the control decision unit. SGD is a commonly used optimization algorithm that updates parameters by calculating the gradient of a small batch of samples in each iteration, thus avoiding getting trapped in local optima to some extent. PSO simulates the foraging behavior of a flock of birds, finding the optimal solution through information sharing and cooperation among particles. In this embodiment, the two algorithms are combined. First, SGD is used to initially optimize the network connection weights, and then PSO is used to further search for better weight values. Let the network connection weights be... Through continuous iterative optimization, the network can better generate reasonable control strategies based on the set of key parameters. For example, in actual aquaculture, when a decrease in dissolved oxygen concentration and an increase in ammonia nitrogen content are detected, after optimization and updating of the dynamic weight allocation network, the system will reasonably adjust the operating parameters of aeration equipment and water exchange equipment according to the weights of these two key parameters, thereby achieving effective water quality control and ensuring the survival environment of aquatic organisms.

[0095] Example 5:

[0096] This embodiment mainly elaborates on multimodal denoising strategies and related content of local-global coupling analysis models.

[0097] Regarding the multimodal denoising strategy, after acquiring the water quality parameter set, the multi-source data acquisition module first identifies abnormal sampling points. Abnormal sampling points may be data errors caused by sensor malfunctions, environmental interference, etc. Interpolation replacement is performed based on the time continuity criterion, assuming the time point of the abnormal sampling point is... The adjacent normal sampling points before and after it are respectively and The corresponding data value is and Then, linear interpolation is used to replace the outlier sampling points, and the replaced values ​​are... The calculation formula is:

[0098]

[0099] In this way, information from adjacent normal data points is used to repair abnormal sampling points, ensuring the continuity and accuracy of the data.

[0100] To detect high-frequency noise components, wavelet packet thresholding decomposition was employed to suppress them. Wavelet packet thresholding decomposition decomposes the signal into different frequency sub-bands. By setting an appropriate threshold, the sub-band containing high-frequency noise is processed to remove the noise. In practice, based on the characteristics of the water quality data and the frequency range of the noise, a suitable wavelet basis function and decomposition level were selected to perform wavelet packet decomposition on the data. Then, the coefficients of the high-frequency sub-bands were thresholded; coefficients below the threshold were set to 0, and coefficients above the threshold were reduced. After processing with wavelet packet thresholding decomposition, the impact of high-frequency noise on the water quality data was effectively reduced.

[0101] The processed data is normalized to generate standard water quality data with values ​​ranging from [0,1]. Normalization eliminates dimensional differences between different parameters, facilitating subsequent data processing and analysis. Let the processed data be... The result after normalization The calculation formula is:

[0102]

[0103] in, This represents the minimum value in the processed data. This indicates the maximum value in the processed data.

[0104] Regarding the local-global coupling analysis model, the iterative correction module first constructs a multi-factor correlation analysis framework based on the Morris screening method to quantify the coupling influence of adjustment commands on various water quality indicators. The Morris screening method is a global sensitivity analysis method that performs a small number of elementary effect calculations on the model input parameters to screen out parameters that have a significant impact on the output results. In this embodiment, this method is used to analyze the degree of influence of different adjustment commands (such as oxygenation commands, water exchange commands, etc.) on various water quality indicators (such as dissolved oxygen concentration, ammonia nitrogen content, etc.). Let the adjustment command be... Water quality indicators are The coupling influence degree is The Morris screening method can be used to obtain the coupling influence matrix of each regulation command on different water quality indicators.

[0105] A parameter perturbation sample set is generated using Latin hypercube sampling, and the interaction coefficients of each indicator are calculated. Latin hypercube sampling is a stratified sampling method that can more uniformly cover the parameter space given a certain number of samples. After generating the parameter perturbation sample set, these samples are used to calculate the interaction coefficients between various water quality indicators. Let the interaction coefficients be... ,in and Different water quality indicators can be represented by interaction coefficients, which can be used to understand the degree of mutual influence between them.

[0106] Indicators with interaction effect coefficients exceeding a preset critical value were selected as core variables for corrective regulation. The preset critical value was determined based on practical aquaculture experience and extensive experimental data. Regarding the interaction effect coefficient... Greater than the preset threshold The indicators for Relevant indicators are used as core variables for corrective regulation. These core variables play a crucial role in subsequent multi-scale data fusion and correction processes, enabling more targeted optimization of multi-parameter synergistic regulation strategies and improving the accuracy and effectiveness of regulation. For example, when the interaction coefficient between dissolved oxygen concentration and ammonia nitrogen content is found to be high and exceeds the preset critical value, the changes in these two indicators will be the focus when correcting the regulation strategy. By comprehensively considering the mutual influence between the two, more reasonable regulatory instructions will be formulated to ensure the stability and health of the aquaculture environment.

[0107] 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.

[0108] 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 related limitations and their equivalents.

Claims

1. A real-time online monitoring system for multiple parameters of aquatic water quality, characterized in that, The system includes: The multi-source data acquisition module is used to acquire a set of water quality parameters within a preset time span in the aquaculture environment, and to perform noise suppression processing on the set of water quality parameters according to a preset multimodal denoising strategy to generate denoised water quality data. The time-series feature parsing module is used to embed the denoised water quality data into a dynamic time-series feature space and generate a multi-dimensional feature cluster based on a time-series segmented adaptive clustering algorithm. The multi-dimensional feature cluster contains the nonlinear correlation between dissolved oxygen concentration, temperature gradient and pollutant distribution. The anomaly identification module is used to perform dynamic deviation compensation on the multidimensional feature clusters according to the preset fluctuation tolerance rules, delineate the real-time monitoring subdomains through the fuzzy logic decision tree algorithm, and extract the key parameter set in each subdomain. The collaborative regulation module is used to construct a regulation decision unit containing a dynamic weight allocation network, and to perform gradient optimization and update of the dynamic weight allocation network using the set of key parameters to generate a multi-parameter collaborative regulation strategy. The iterative correction module is used to perform multi-scale data fusion correction on the multi-parameter collaborative control strategy according to the preset local-global coupling analysis model, and output the real-time adjustment command of the entire aquatic environment. The fuzzy logic decision tree algorithm delineates the real-time monitoring subdomain, including: The decision objective is defined as the balanced optimization of parameter fluctuations and equipment energy consumption within the subdomain; The constraints are set as sensor distribution density, environmental carrying capacity threshold, and biological activity cycle range. The decision objective is solved using a heuristic search algorithm, and the optimal monitoring subdomain partitioning scheme is output.

2. The real-time online monitoring system for multiple parameters of aquatic water quality as described in claim 1, characterized in that, The time-series segmented adaptive clustering algorithm generates multidimensional feature clusters, including: Extract the timestamp, monitoring point number, and parameter type from the denoised water quality data to construct a time-series tensor matrix; The time series tensor matrix is ​​segmented using an overlapping window truncation algorithm to generate a set of time series segments. A density-based outlier detection model is used to remove outliers from the time series segment set, retaining valid data segments; The effective data fragments are analyzed in multiple dimensions using a spectral clustering algorithm to generate feature clusters with spatiotemporal continuity.

3. The aquatic water quality multi-parameter real-time online monitoring system as described in claim 2, characterized in that, The dynamic deviation compensation includes: Based on the distribution of historical water quality data in the real-time monitoring subdomains, the moving average and standard deviation of key parameters in each subdomain are calculated. Based on the adaptive moving average algorithm, the moving average and standard deviation are corrected by Kalman filtering to generate a dynamic benchmark curve. The dynamic baseline curve is extended with fuzzy boundaries using an interval membership function to generate an adaptive compensation interval.

4. The aquatic water quality multi-parameter real-time online monitoring system as described in claim 1, characterized in that, The dynamic weight allocation network performs gradient optimization updates, including: The set of key parameters is input to the input node of the dynamic weight allocation network, and the temporal dependency features are extracted using a bidirectional recurrent neural network. The temporal dependency features are prioritized using a gating mechanism to generate a parameter weight sequence. The network connection weights are optimized using a combination of stochastic gradient descent and particle swarm optimization, and the control parameters of the control decision unit are updated accordingly.

5. The real-time online monitoring system for multiple parameters of aquatic water quality as described in claim 1, characterized in that, The multimodal denoising strategy includes: Identify abnormal sampling points in the set of water quality parameters and perform interpolation replacement based on the time continuity criterion; High-frequency noise components are detected and suppressed using wavelet packet thresholding. The processed data is normalized to generate standard water quality data with values ​​ranging from [0,1].

6. The real-time online monitoring system for multiple parameters of aquatic water quality as described in claim 1, characterized in that, The local-global coupling analysis model includes: A multi-factor correlation analysis framework based on the Morris screening method was constructed to quantify the coupling influence of adjustment commands on various water quality indicators. A parameter perturbation sample set is generated by Latin hypercube sampling, and the interaction effect coefficients of each index are calculated. Indicators with interaction effect coefficients higher than a preset critical value were selected as core variables for adjustment and regulation.

7. The real-time online monitoring system for multiple parameters of aquatic water quality as described in claim 1, characterized in that, The multi-scale data fusion correction includes: By integrating sensor measured data, hydrological model prediction data, and aquaculture feeding records, a heterogeneous data correlation matrix is ​​constructed. The heterogeneous data association matrix is ​​subjected to feature extraction using a nonnegative matrix factorization algorithm to obtain low-dimensional feature basis vectors. The low-dimensional feature basis vector is input into the iterative correction module to generate the corrected adjustment command.

8. The real-time online monitoring system for multiple parameters of aquatic water quality as described in any one of claims 1 to 7, characterized in that, The system also includes: An early warning module containing a biosafety assessment function is constructed, which matches and analyzes the real-time adjustment instructions with biosafety thresholds and outputs multi-level early warning indicators; The preset emergency intervention mechanism is activated based on the multi-level early warning indicators, generating instructions for starting / stopping oxygenation or water exchange.

9. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the operation of the real-time online monitoring system for multiple parameters of aquatic water quality as described in any one of claims 1 to 8.

Images