Intelligent pond ecological management method, system and device

Abstract

The present invention provides an intelligent pond ecological management method, system and device, which collects multi-dimensional data of the pond in real time through sensors arranged inside and outside the pond, performs data preprocessing on the multi-dimensional data of the pond, extracts the feature set of the pond, establishes an intelligent diagnosis system based on the feature set, performs real-time abnormal condition diagnosis on the pond ecosystem based on the intelligent diagnosis system, generates and displays abnormal diagnosis results and corresponding early warning information, generates ecological regulation strategies in real time based on the abnormal diagnosis results and early warning information, and controls each actuator to carry out actual regulation actions through a distributed control network to complete closed-loop ecological regulation. The present invention collects multi-dimensional pond data and performs data processing, performs abnormal diagnosis on the pond ecosystem through an intelligent diagnosis system, and finally performs automatic regulation to achieve automated management of the pond ecosystem.

Description

Technical Field

[0001] The present invention belongs to the technical field of ecological regulation, and in particular relates to an intelligent pond ecological management method, system and device. Background Art

[0002] The organisms in the pond need a suitable ecological environment to survive. Affected by many factors, the ecological environment of the pond is not always stable. For some ponds that need special attention, real-time monitoring and intelligent adjustment are required to ensure the good stability of the pond ecosystem. However, the existing intelligent adjustment schemes for ponds have the following problems:

[0003] (1) Most of them are control and management solutions for pond water quality or fish, and there are few solutions for automated control of the entire pond ecosystem;

[0004] (2) There is a lack of diversity in the data collection of ponds, the collected information is not comprehensive enough, and the factors considered when automating the regulation of pond ecosystems are not comprehensive enough;

[0005] (3) After extracting the pond ecological environment data, there is no specific and accurate data processing plan, and the abnormal diagnosis of the pond ecosystem is not accurate enough, which has little reference value in the pond ecological regulation scenario.

[0006] Therefore, we need to develop an intelligent pond ecological management method, system and device that can achieve comprehensive perception, timely diagnosis and optimized management of the pond ecosystem through multi-source data collection, intelligent analysis and precise regulation. Summary of the invention

[0007] The purpose of the present invention is to provide an intelligent pond ecological management method, system and device to solve the problems mentioned in the above background technology, such as incomplete data extraction, inaccurate anomaly detection and few factors considered in the intelligent control scheme of the existing pond ecosystem.

[0008] To achieve the above object, the present invention provides an intelligent pond ecological management method, the method is specifically as follows:

[0009] Data collection: sensors placed inside and outside the pond are used to collect multi-dimensional data of the pond in real time. The multi-dimensional data includes pond environmental data and pond biological data, including but not limited to water quality parameters, meteorological conditions, and biological population density;

[0010] Data processing: preprocessing the multi-dimensional data of the pond, converting the multi-dimensional data into a unified three-dimensional data cube X, and extracting the feature set F of the pond, the data preprocessing includes: processing outliers based on a threshold method based on domain knowledge and a statistical 3σ criterion; identifying missing values ​​in the multi-dimensional data and filling them; performing time alignment by interpolation or downsampling; and dedimensionalization;

[0011] Abnormal diagnosis: Based on the feature set F, an intelligent diagnosis system is established. Based on the intelligent diagnosis system, real-time abnormal condition diagnosis of the pond ecosystem is performed, and abnormal diagnosis results and corresponding early warning information are generated and displayed;

[0012] Ecological regulation: Based on the abnormal diagnosis results and early warning information, the ecological regulation strategy is generated in real time, and the various actuators are controlled through the distributed control network to carry out actual regulation actions to complete the closed-loop ecological regulation.

[0013] Based on the above solution, the data preprocessing is to convert the multi-dimensional data into a unified three-dimensional data cube X, specifically:

[0014] A method combining a threshold method based on domain knowledge and a statistical 3σ criterion is used to identify and process outliers in the multi-dimensional data;

[0015] Identify missing values ​​in the multi-dimensional data. For short-term random missing values, use linear interpolation to fill them. For long-term continuous missing values, use the mean of historical data for the same period to fill them. The historical data for the same period is preferably selected from the same time period within a week. The short time includes but is not limited to a time period of less than 15 minutes, and the long time includes but is not limited to a time period of more than 1 hour.

[0016] Using the minimum time granularity as a benchmark, align the multi-dimensional data collected by all sensors to a unified time point through interpolation or downsampling;

[0017] The Z-score standardization method is used to transform the data of different indicators in the multi-dimensional data into a standard normal distribution with a mean of 0 and a standard deviation of 1. The calculation formula is: X = (x ′ -μ) / σ, where x ′ is the original value, μ is the mean of the indicator, and σ is the standard deviation.

[0018] Based on the above scheme, the step of extracting the feature set F of the pond includes:

[0019] Extract from the three-dimensional data cube X, but not limited to time series features, spatial features and multi-dimensional features, to form a feature set F = {f1(x), f2(x), ..., f q (x)}, where each feature f i (x) are all abstract representations of the three-dimensional data cube X in a specific form;

[0020] The time series characteristics include: mean, variance, skewness, kurtosis, autocorrelation coefficient and partial autocorrelation coefficient, etc.;

[0021] The spatial features include local features and spatial autocorrelation index, and the calculation formula is: Where N is the number of monitoring points, W is the sum of the spatial weight matrix, and w ij is the spatial weight, x i and x j are the observed values ​​at different locations, is the mean.

[0022] Based on the above scheme, the establishment of an intelligent diagnosis system is specifically as follows:

[0023] An ecological status assessment model is constructed based on a training data set, wherein the training data set includes historically collected multi-dimensional pond data, a feature set F corresponding to the multi-dimensional pond data, and a pond ecological status assessment result annotated by aquaculture experts, wherein the ecological status is divided into three categories: "healthy", "sub-healthy" and "abnormal";

[0024] The feature set F is subjected to abnormality detection by the ecological status assessment model to obtain the abnormality diagnosis result, which includes the health metric score of the pond ecosystem. The calculation formula of the health metric score is as follows:

[0025] Health score =100*(1-max(P abnormal ,0.5*P subhealth ))

[0026] Among them, P abnormal and P subhealth are the probabilities of the pond being in an "abnormal" and "sub-healthy" state, respectively, and the health metric score ranges from 0 to 100.

[0027] Based on the above scheme, the anomaly detection is completed by the three main components of the anomaly detector, state classifier and cause analyzer of the ecological status assessment model, specifically including:

[0028] The anomaly detector adopts the isolation forest algorithm, and the anomaly score of a data sample η is calculated as follows: s(η,n)=2^(-E(h(η)) / c(n)), where E(h(η)) is the average path length of sample η, c(n) is the average path length when n samples are given, and the closer the anomaly score is to 1, the greater the probability that the data sample η is an anomaly point;

[0029] The state classifier uses an ensemble learning method, combined with random forest and gradient boosting decision tree, to perform state classification prediction on the feature set F;

[0030] The prediction results of random forest and gradient boosting decision tree for feature set F are fused by weighted average. The weights are dynamically adjusted according to their performance on the validation set. The final ecological status classification probability calculation formula of the pond ecosystem is as follows:

[0031] P(y|x)=w rf *P rf (y|x)+w gbdt *P gbdt (y|x)

[0032] Among them, w rf and w gbdt are the weights of random forest and gradient boosting decision tree, respectively, rf (y|x) and P gbdt (y|x) are the probabilities of classifying sample x as y for random forest and gradient boosting decision tree respectively;

[0033] The cause analyzer interprets the abnormal detection result of the ecological status assessment model based on the Shapley addition explanation value;

[0034] For feature i in feature set F, its Shapley addition explanation value is calculated as follows:

[0035]

[0036] Among them, φ i represents the Shapley addition interpretation value of feature i, F is the set of all features, S is the subset that does not contain feature i, and f x (S) is the anomaly detection result of the ecological status assessment model under the feature set S. By calculating the Shapley addition explanation value of each feature, it is determined which factors contribute most to the generation of abnormal status, providing a basis for subsequent regulatory strategies.

[0037] Based on the above scheme, the real-time generation of ecological control strategy specifically includes:

[0038] Construct a comprehensive regulatory decision space, abstract each possible regulatory measure for the pond ecosystem into a decision variable, and give it a corresponding numerical representation to form a high-level decision vector D;

[0039] The environmental regulation problem of the pond ecosystem is transformed into a multi-objective optimization function maximum problem;

[0040] Based on the standard non-dominated sorting genetic algorithm, adaptive crossover and mutation operations are introduced to solve the multi-objective optimization function;

[0041] Construct a dynamic model of the pond ecosystem, estimate and verify parameters based on the historical ecological data of the pond, and predict the ecological change trend of the pond in the short term;

[0042] Based on the multi-objective optimization function and the dynamic model, a group of Pareto solutions are output, each solution represents a possible regulation strategy, and a final ecological regulation strategy is selected from the group of Pareto solutions.

[0043] Based on the above scheme, the multi-objective optimization function considers the following objectives: ecological benefit E(D), which is used to evaluate the degree of improvement of the pond ecosystem by regulatory measures; economic benefit C(D), which is used to calculate the cost and expected benefits of regulatory measures; operational complexity O(D), which is used to measure the difficulty of implementing regulatory measures; the multi-objective optimization function is: F(D)=w1*E(D)-w2*C(D)-w3*O(D), where w1, w2, and w3 are the weight coefficients of each objective.

[0044] The present invention also provides an intelligent pond ecological management system, the system specifically comprising:

[0045] The data collection layer is used to collect multi-dimensional data of the pond in real time through sensors arranged inside and outside the pond. The multi-dimensional data includes pond environmental data and pond biological data, including but not limited to water quality parameters, meteorological conditions and biological population density;

[0046] The data processing layer is used to perform data preprocessing on the multi-dimensional data of the pond, convert the multi-dimensional data into a unified three-dimensional data cube X, and extract the feature set F of the pond. The data preprocessing includes: processing outliers by combining a threshold method based on domain knowledge and a statistical 3σ criterion; identifying missing values ​​in the multi-dimensional data and filling them; performing time alignment by interpolation or downsampling; and dedimensionalization.

[0047] The diagnosis and early warning layer is used to establish an intelligent diagnosis system based on the feature set F, and to perform real-time abnormal condition diagnosis on the pond ecosystem based on the intelligent diagnosis system, and to generate and display abnormal diagnosis results and corresponding early warning information;

[0048] The ecological control layer combines real-time diagnosis results and warning levels to generate the optimal control strategy. Based on the generated control strategy, it controls each executive agency to carry out actual control actions to complete closed-loop ecological regulation.

[0049] In addition, the present invention also provides an intelligent pond ecological management device, the device comprising:

[0050] A memory for storing an intelligent pond ecological management program;

[0051] A processor, wherein the processor is configured to execute the intelligent pond ecological management method in the above solution;

[0052] The processor and the memory are connected via a bus, which includes but is not limited to an I2C bus.

[0053] Compared with the prior art, the present invention has the following advantages and effects:

[0054] (1) The present invention collects and analyzes pond environmental data and pond biological data in real time, and establishes an intelligent diagnosis system for abnormal conditions to generate accurate pond ecological environment abnormal state information. Based on this information, the pond is intelligently and comprehensively automated for ecological regulation, and pond state information can be obtained in a timely and accurate manner to protect the pond ecological environment.

[0055] (2) By collecting multi-dimensional data of the pond, including but not limited to water quality parameters, meteorological conditions, and biological population density, multiple aspects such as ecological benefits, economic benefits, and operational complexity are considered during ecological regulation, so that the pond ecosystem can be well regulated while maximizing economic benefits;

[0056] (3) In the abnormal diagnosis of the pond ecological environment, a variety of machine learning algorithms such as isolation forest algorithm, random forest, gradient boosting decision tree, and additive explanatory value are used to make the abnormal diagnosis results more accurate, ensuring the accuracy of the abnormal diagnosis results of the pond ecosystem, which has great reference value in the pond ecological regulation scenario. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for the specific embodiments or the description of the prior art. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn according to the actual scale.

[0058] Figure 1 This is a flow chart of an intelligent pond ecological management method provided by an embodiment of the present invention;

[0059] Figure 2 is a flow chart of a method for abnormality diagnosis of a pond ecosystem provided by an embodiment of the present invention;

[0060] Figure 3 It is a structural schematic diagram of an intelligent pond ecological management system provided by an embodiment of the present invention;

[0061] Figure 4 It is a structural schematic diagram of an intelligent pond ecological management device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0062] In order to more clearly explain the purpose, technical solutions and advantages of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. The example implementation methods can be implemented in various forms and should not be understood as being limited to the examples described herein. On the contrary, these implementation methods are provided to make the present invention more comprehensive and complete, and to fully convey the concepts of the example implementation methods to those skilled in the art.

[0063] In addition, the described features, structures or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present invention. However, those skilled in the art will appreciate that the technical solution of the present invention can be practiced without one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known methods, devices, implementations or operations are not shown or described in detail to avoid blurring various aspects of the present invention.

[0064] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0065] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to actual conditions.

[0066] The present invention is described in detail below in conjunction with specific embodiments:

[0067] Example 1

[0068] As attached Figure 1 As shown, Embodiment 1 of the present invention provides an intelligent pond ecological management method, and the specific steps of the method are as follows:

[0069] Step S1, data collection: using sensors arranged inside and outside the pond, real-time multi-dimensional data of the pond is collected, wherein the multi-dimensional data includes pond environmental data and pond biological data, including but not limited to water quality parameters, meteorological conditions, and biological population density;

[0070] Step S2, data processing: preprocessing the multi-dimensional data of the pond, converting the multi-dimensional data into a unified three-dimensional data cube X, and extracting the feature set F of the pond, the data preprocessing includes: processing outliers based on a threshold method based on domain knowledge and a statistical 3σ criterion; identifying missing values ​​in the multi-dimensional data and filling them; performing time alignment by interpolation or downsampling; and dedimensionalization;

[0071] Step S3, abnormality diagnosis: based on the feature set F, an intelligent diagnosis system is established, and based on the intelligent diagnosis system, real-time abnormal condition diagnosis of the pond ecosystem is performed, and abnormal diagnosis results and corresponding warning information are generated and displayed;

[0072] Step S4, ecological regulation: Based on the abnormal diagnosis results and early warning information, an ecological regulation strategy is generated in real time, and each actuator is controlled through a distributed control network to carry out actual regulation actions to complete a closed-loop ecological regulation.

[0073] Preferably, the sensor in step S1 is specifically a set of multi-level and multi-dimensional sensor networks;

[0074] Specifically, the sensor network includes multiple types of sensors such as water quality sensors, meteorological sensors, image sensors, and acoustic sensors, each of which continuously monitors a specific ecological parameter;

[0075] Specifically, in order to ensure the spatial representativeness of the collected data, this embodiment adopts a layered point layout strategy to set up real-time sensor monitoring points: in the vertical direction, the sensors are arranged in the key water layers of the pond, including but not limited to the surface layer (water depth 0.5m), the middle layer (water depth 1.5m) and the bottom layer (water depth 3.0m); in the horizontal direction, a grid layout is adopted, and the points are densely arranged in special areas where water quality is prone to abnormalities, such as water inlets, outlets, and dead corners. The above layout method can fully capture the spatial heterogeneity of water bodies and provide rich spatial information for subsequent data analysis;

[0076] Exemplarily, the water quality sensor is mainly used to measure the physical and chemical indicators of the water body, which include but are not limited to dissolved oxygen, pH value, temperature, turbidity, ammonia nitrogen, nitrate, nitrite, total phosphorus, etc. These parameters can directly reflect the ecological status and pollution level of the water body; the meteorological sensor is used to monitor the environmental conditions around the pond, which include but are not limited to light intensity, air temperature, humidity, wind speed, rainfall, etc. These factors directly or indirectly affect the physical and chemical properties and biological activities of the water body; the image sensor is used to perform visible light imaging analysis of water bodies and organisms. In this embodiment, a high-definition camera is installed on the surface of the pond to monitor water color changes, floating objects, fish activities, etc. in real time through image recognition technology; the acoustic sensor is used to monitor changes in the underwater sound spectrum and judge the state of biological activity. Different aquatic organisms will produce sounds in a specific frequency range. By analyzing these sounds, biomass, population structure and even certain physiological states can be inferred.

[0077] Preferably, it should be noted that step S2 aims to improve the quality of multi-dimensional data, eliminate noise and outliers, and extract key features that can effectively characterize the ecological status of the pond from massive raw data;

[0078] Preferably, the data preprocessing in step S2 includes but is not limited to outlier processing, missing value filling, time alignment and dedimensionalization, etc., which are specifically as follows:

[0079] Step S201, identifying and processing abnormal values ​​in the multi-dimensional data by combining a threshold method based on domain knowledge and a statistical 3σ criterion;

[0080] Exemplarily, this embodiment preprocesses the multi-dimensional data collected by the sensor network, and the preprocessing process includes operations such as outlier processing, missing value filling, time alignment and dedimensionalization. This embodiment adopts a method combining a threshold method based on domain knowledge and a statistical 3σ criterion to identify and process outliers in multi-dimensional data; taking dissolved oxygen as an example, according to the experience of aquaculture, its normal range is usually between 0-20 mg / L, so for values ​​outside this range, the system will automatically mark them as abnormal; for values ​​within this range but deviating from the mean by more than 3 standard deviations, they will be regarded as potential outliers. For the identified outliers, this embodiment adopts the method of adjacent value replacement for processing, that is, replacing the outliers with the average value of the time points before and after the outliers. This method can effectively remove the impact of abnormal fluctuations while retaining the temporal continuity of the data.

[0081] Step S202: Identify missing values ​​in the multi-dimensional data. For short-term random missing values, use linear interpolation to fill them. For long-term continuous missing values, use the mean of historical data for the same period to fill them. The historical data for the same period is preferably selected from the same time period within a week. The short time includes but is not limited to a time period of less than 15 minutes, and the long time includes but is not limited to a time period of more than 1 hour.

[0082] For example, in this embodiment, the missing values ​​in the multi-dimensional data are filled by linear interpolation for random missing values ​​within a short period of time, such as within 15 minutes, and the mean of historical data for the same period is used for filling for continuous missing values ​​for a longer period of time, such as more than 1 hour. For example, if the dissolved oxygen data from 10:00 to 11:00 on August 26 is missing in the multi-dimensional data, the system will search for dissolved oxygen data in the same period of the past 7 days and calculate its mean as the filling value to fill the missing value.

[0083] Step S203: Using the minimum time granularity as a benchmark, align the multi-dimensional data collected by all sensors to a unified time point by interpolation or downsampling;

[0084] For example, in this implementation, due to the different sampling frequencies of different sensors, the original data, i.e., multi-dimensional data, often has the problem of time misalignment. This embodiment uses the minimum time granularity as a benchmark to align all data to a unified time point by interpolation or downsampling. For example, if the minimum sampling interval of the system is 5 minutes, all data will be adjusted to a data point every 5 minutes. For sensors with higher sampling frequencies, such as dissolved oxygen sensors that sample once per minute, the average value of the data within 5 minutes is used as the aligned value; for sensors with lower sampling frequencies, such as ammonia nitrogen sensors that sample once every 30 minutes, the values ​​of intermediate time points are generated by linear interpolation.

[0085] Step S204: Use the Z-score standardization method to convert the data of different indicators in the multi-dimensional data into a standard normal distribution with a mean of 0 and a standard deviation of 1. The calculation formula is: X = (x ′ -μ) / σ, where x ′ is the original value, μ is the mean of the indicator, and σ is the standard deviation;

[0086] Specifically, the de-dimensionalization in step S204 is to eliminate the influence of the dimension difference between different indicators on the subsequent analysis. The Z-score standardization method adopted in this embodiment can convert the data of different indicators into a standard normal distribution with a mean of 0 and a standard deviation of 1, which is convenient for comparison and comprehensive analysis between different indicators.

[0087] It should be noted that after the above data preprocessing steps, the original multi-dimensional data collected by the sensor network is converted into a unified three-dimensional data cube X = (x ijk ), where i represents the time dimension, with a value range of 1 to T, where T is the total number of time steps; j represents the space dimension, with a value range of 1 to N, where N is the total number of sensors; k represents the indicator dimension, with a value range of 1 to M, where M is the total number of monitoring indicators. Each element x ijk Represents the normalized value of the kth indicator at the jth sensor position at the ith time step.

[0088] Preferably, extracting the feature set F of the pond in step S2 includes:

[0089] Step S205: extracting features including but not limited to time series features, spatial features and multi-dimensional features from the three-dimensional data cube X to form a feature set F = {f1(x), f2(x), ..., f q (x)}, where each feature f i (x) are all abstract representations of the three-dimensional data cube X in a specific form;

[0090] It should be noted that, in this embodiment, feature extraction is performed based on the three-dimensional data cube X. The purpose of feature extraction is to extract key information that can effectively characterize the ecological status of the pond from the original data. This embodiment designs a multi-dimensional and multi-scale feature extraction scheme, including time series features, spatial features and multi-dimensional features.

[0091] Preferably, the time series characteristics in step S205 include: mean, variance, skewness, kurtosis, autocorrelation coefficient and partial autocorrelation coefficient, etc.;

[0092] Specifically, in this embodiment, the time series features that need to be calculated include: mean, variance, skewness, kurtosis, etc., which can reflect the overall distribution of the data, specifically including first-order differences, second-order differences, moving averages, etc., which can reflect the changing trends and fluctuations of the data; the time series features also include autocorrelation coefficients, partial autocorrelation coefficients, etc., which are used to detect possible periodic patterns in the data. Taking dissolved oxygen as an example, assuming that the dissolved oxygen data of a certain monitoring point in the past 24 hours is DO = [5.8, 6.2, 6.5, 6.8, 7.0, 7.1, 7.2, 7.1, 7.0, 6.8, 6.5, 6.2, 5.9, 5.7, 5.6, 5.8, 6.1, 6.4, 6.7, 6.9, 7.0, 7.1, 7.0, 6.9], then the calculation of this part of the time series features is as follows: mean = 6.5125; variance = 0.2801; first-order Difference = [0.4, 0.3, 0.3, 0.2, 0.1, 0.1, -0.1, -0.1, -0.2, -0.3, -0.3, -0.3, -0.2, -0.1, 0.2, 0.3, 0.3, 0.3, 0.2, 0.1, 0.1, -0.1, -0.1]; autocorrelation coefficient of 24-hour period = 0.9256. These characteristics comprehensively show the diurnal variation pattern of dissolved oxygen and reflect the influence of photosynthesis and respiration on dissolved oxygen in water.

[0093] Preferably, the spatial features in step S205 include local features and spatial autocorrelation index;

[0094] Specifically, in this embodiment, the spatial features are used to characterize the relationship between different monitoring points and the overall spatial distribution pattern. In this embodiment, the following spatial features are calculated: local features and spatial autocorrelation index, such as Moran's I, which reflects the overall spatial correlation. The calculation formula is:

[0095]

[0096] Where N is the number of monitoring points, i and j are different monitoring points, W is the sum of the spatial weight matrix, and w ij is a spatial weight and is usually defined based on distance, x i and x j are the observed values ​​at different locations, is the mean;

[0097] Preferably, the multidimensional feature extraction method in step S205 includes but is not limited to correlation analysis, independent component analysis, and principal component analysis;

[0098] Specifically, the multidimensional features are intended to capture the relationships between different indicators. This embodiment uses methods including but not limited to correlation analysis, independent component analysis, principal component analysis, etc. to extract the multidimensional features.

[0099] Specifically, after the above feature extraction process, the original three-dimensional data cube X is transformed into a highly generalized feature set F = {f1(x), f2(x), ..., f q (x)}. Each feature f i (x) are all abstract representations of the data cube in a certain form. For example, f1(x) may represent the average level of dissolved oxygen in the whole pond within 24 hours, f2(x) may represent the spatial autocorrelation of pH value, and f3(x) may represent the correlation coefficient between dissolved oxygen and water temperature.

[0100] Preferably, the establishment of the intelligent diagnosis system in step S3 specifically includes:

[0101] Step S301, constructing an ecological status assessment model based on a training data set, wherein the training data set includes historically collected multi-dimensional pond data, a feature set F corresponding to the multi-dimensional pond data, and a pond ecological status assessment result annotated by aquaculture experts, wherein the ecological status is divided into three categories: "healthy", "sub-healthy" and "abnormal" status;

[0102] Exemplarily, this embodiment constructs a comprehensive training data set, which includes historically collected pond data, a corresponding feature set F, and ecological status assessment results annotated by aquaculture experts. The ecological status is divided into three categories: "healthy", "sub-healthy" and "abnormal". In order to ensure the quality and reliability of the training data set, the training data set is subjected to data enhancement processing, including: ensuring the balance of the number of samples in each category by oversampling the minority category and undersampling the majority category method to avoid bias in the ecological status assessment model; using K-fold cross-validation to test the generalization ability of the ecological status assessment model, wherein K is preferably 5, and stratified sampling is used to ensure that the proportion of each category in each fold is consistent with the original data set; using a time series segmentation method to allocate the training data set, using data from an early time period a as a training set, and data from a late time period b as a test set to simulate actual application scenarios. Exemplarily, a can select the first 50% time period of the training data set, and b can select the last 30% time period of the training data set.

[0103] Step S302: The ecological status assessment model performs anomaly detection on the feature set F through a machine learning algorithm to obtain the anomaly diagnosis result, which includes a health metric score of the pond ecosystem. The calculation formula of the health metric score is as follows:

[0104] Health score =100*(1-max(P abnormal ,0.5*P subhealth ))

[0105] Among them, P abnormal and P subhealth are the probabilities of the pond being in an "abnormal" and "sub-healthy" state, respectively, and the health metric scores range from 0 (extremely unhealthy) to 100 (very healthy);

[0106] Preferably, the anomaly detection in step S302 is completed by three main components of the anomaly detector, state classifier and cause analyzer of the ecological state assessment model, specifically:

[0107] Step S3021, the anomaly detector adopts the isolation forest algorithm, first randomly selects a feature A from the feature set F, randomly selects a split point within the value range of feature A, divides the feature set F into two parts based on the split point, and continues to divide each divided subset until each sample is individually isolated or reaches a preset tree height limit, and calculates an anomaly score for each sample;

[0108] Specifically, in this embodiment, the anomaly detector uses the Isolation Forest algorithm. First, a feature is selected from the random feature set F, and then a split point is randomly selected within the value range of the feature, and the data set is divided into two parts accordingly. Finally, this process is repeated for each divided subset until each data sample is isolated individually or the pre-set tree height limit is reached. This process constructs a recursive splitting structure based on random splitting. For a data sample η, its anomaly score is calculated as follows:

[0109] s(η,n)=2^(-E(h(η)) / c(n))

[0110] Among them, E(h(η)) is the average path length of sample η, c(n) is the average path length when n samples are given, and the closer the anomaly score is to 1, the greater the probability that the data sample η is an anomaly point.

[0111] Step S3022: The state classifier uses an ensemble learning method, combined with random forest and gradient boosting decision tree, to perform state classification prediction on the feature set F;

[0112] Specifically, the state classifier uses an ensemble learning method, combining Random Forest and Gradient Boosting Decision Tree (GBDT). Among them, the main parameter settings of Random Forest are the number of trees: 100; maximum depth: 10; minimum number of leaf node samples: 5; feature selection: √n, n is the total number of features. The main parameter settings of GBDT are the number of trees: 100; learning rate: 0.1; maximum depth: 5; minimum number of leaf node samples: 10.

[0113] Furthermore, the prediction results of the random forest and gradient boosting decision tree for the feature set F are fused by weighted averaging. The weights are dynamically adjusted according to their performance on the validation set. The final ecological state classification probability calculation formula of the pond ecosystem is as follows:

[0114] P(y|x)=w rf *P rf (y|x)+w gbdt *P gbdt (y|x)

[0115] Among them, w rf and w gbdt are the weights of random forest and gradient boosting decision tree, respectively, rf (y|x) and P gbdt (y|x) is the probability that the random forest and gradient boosting decision trees classify the sample x as y.

[0116] Step S3023: Based on the Shapley addition explanation value, the cause analyzer explains the abnormal detection result of the ecological status assessment model;

[0117] Specifically, the cause analyzer explains the abnormal detection results of the ecological status assessment model for the feature set F based on the Shapley Additive exPlanations (SHAP) value. The SHAP value is based on the concept of Shapley value in game theory and is used to quantify the contribution of each feature to the prediction result. For feature i in feature set F, its SHAP value is calculated as follows:

[0118]

[0119] Among them, φ i represents the SHAP value of feature i, F is the set of all features, S is the subset that does not contain feature i, and f x(S) is the anomaly detection result of the ecological status assessment model under the feature set S. By calculating the SHAP value of each feature, it is determined which factors contribute most to the generation of abnormal states, providing a basis for subsequent regulatory strategies.

[0120] Preferably, based on the three main components of the anomaly detector, state classifier and cause analyzer of the ecological state assessment model in the intelligent diagnosis system, this embodiment provides an abnormality diagnosis method for a pond ecosystem, as shown in the attached Figure 2 As shown, the specific process is as follows:

[0121] Step S31: Calculate the input feature set F = {f1(x), f2(x), ..., f q (x)}, if the anomaly score exceeds the first preset threshold, an anomaly warning is directly triggered;

[0122] Specifically, in this embodiment, the first preset threshold is preferably 0.8;

[0123] Step S32, inputting the feature set F into the state classifier for state evaluation, obtaining the abnormal state classification probability, and outputting the probability of the pond being in the three states of "healthy", "sub-healthy" and "abnormal"; if the probability of the "abnormal" state exceeds the second preset threshold, the "abnormal" state warning will be triggered, and if the probability of the "sub-healthy" state exceeds the third preset threshold, the "sub-healthy" state warning will be triggered;

[0124] Specifically, in this embodiment, the second preset threshold is preferably 0.6, and the third preset threshold is preferably 0.8;

[0125] Step S33: When an "abnormal" state warning or a "sub-healthy" state warning is triggered, the cause analyzer calculates the SHAP value of each feature in the feature set F, finds B features with the largest SHAP value, that is, the features with the largest contribution to the abnormality, as candidates for the abnormal cause, and obtains the abnormal cause analysis result;

[0126] Specifically, B needs to be selected according to the specific situation of feature set F. In this embodiment, 3 to 5 are preferably selected;

[0127] Step S34: generating structured warning information according to the abnormality score, abnormal state classification probability and abnormality cause analysis result;

[0128] Preferably, the warning information includes abnormality type, warning level, key indicators, possible causes and suggested measures;

[0129] Specifically, the warning level is divided into yellow warning, orange warning and red warning according to the severity of the abnormality; the abnormality types include but are not limited to water quality deterioration, algal bloom risk, hypoxia risk, etc.; the key indicators include the main indicators causing the abnormality and their current values, such as dissolved oxygen: 3.2 mg / L (low); the possible causes are analyzed based on the SHAP value, such as continuous high temperature leading to a decrease in dissolved oxygen; the recommended measures are given based on the possible causes of the abnormality type, such as starting emergency oxygen enrichment equipment;

[0130] Step S35: Based on the probabilities of "abnormal" and "sub-healthy" states, calculate the health metric score of the pond represented by the feature set F, with the score ranging from 0 (extremely unhealthy) to 100 (very healthy). The calculation formula is as follows:

[0131] Health score =100*(1-max(P abnormal ,0.5*P subhealth ))

[0132] Among them, P abnormal and P subhealth are the probabilities of the pond being in an "abnormal" and "sub-healthy" state respectively.

[0133] It should be noted that the intelligent diagnostic system proposed in this embodiment can monitor the ecological status of the pond in real time, discover potential problems in a timely manner, and provide detailed diagnostic information. The system can not only identify known abnormal patterns, but also capture new and unprecedented abnormal situations through anomaly detectors. By combining the training data set and some of the machine learning algorithms mentioned above, the system can continuously learn and adapt to different breeding environments and conditions, thereby improving the accuracy and reliability of diagnosis.

[0134] Preferably, the real-time generation of the ecological regulation strategy in step S4 specifically includes:

[0135] Step S401, constructing a comprehensive control decision space, abstracting each possible control measure for the pond ecosystem into a decision variable, and assigning a corresponding numerical representation to form a high-dimensional decision vector D, wherein the decision variables include but are not limited to the dosage of the nitrogen-fixing bacteria preparation, the operating time of the aeration equipment, the water change ratio, etc.;

[0136] Exemplarily, in this embodiment, a comprehensive control decision space is first constructed, and each possible control measure is abstracted as a decision variable and given an appropriate numerical representation; for example, the dosage of nitrogen-fixing bacteria preparation can be represented by an integer of 0 to 100, representing the number of milliliters per cubic meter of water; the operating time of the aeration equipment can be represented by a floating point number of 0 to 24, representing the number of hours of operation per day; the water change ratio can be represented by a floating point number of 0 to 1, representing the proportion of water replaced per day to the total water volume. The above decision variables constitute a high-dimensional decision vector D = [d1, d2, ..., d n ], recorded as the control measure vector D, where each component d i Represents a specific regulatory measure;

[0137] Step S402, converting the environmental regulation problem of the pond ecosystem into a multi-objective optimization function maximum problem;

[0138] Exemplarily, this embodiment designs a multi-objective optimization function, which comprehensively considers the following objectives: ecological benefit E(D), which is used to evaluate the degree of improvement of the control measures on the pond ecosystem; economic benefit C(D), which is used to calculate the cost and expected benefits of the control measures; operation complexity O(D), which is used to measure the difficulty of implementing the control measures. The optimization problem of the pond ecological environment can be formally expressed as:

[0139] Maximize F(D) = w1*E(D)-w2*C(D)-w3*O(D)

[0140] Satisfy the constraints:

[0141] g i (D) <= 0, i = 1, 2, ..., m (inequality constraint)

[0142] h j (D)=0,j=1,2,...,n(equality constraint)

[0143] Among them, w1, w2, w3 are the weight coefficients of each target, which can be adjusted according to the specific situation; g i (D) and h j (D) represent inequality constraints and equality constraints, respectively, which are used to ensure that the control strategy meets various practical constraints.

[0144] Exemplarily, the calculation of the ecological benefit E(D) is based on the health metric score and the warning level, and an ecological status improvement function ΔE(D) is defined, which can predict the expected improvement in the health metric score after taking the corresponding regulatory measures of the regulatory measure vector D; the function construction of the ecological benefit E(D) is based on historical data and expert knowledge, and is trained using machine learning methods such as random forest regression; the calculation of the economic benefit C(D) includes the direct costs of regulatory measures such as drugs and electricity consumption and the expected output gains such as the benefits brought by increased fish growth rate; the calculation of the operation complexity O(D) takes into account factors such as manpower requirements, time consumption and technical difficulty; the constraints include two categories: strong constraints and weak constraints. Strong constraints are mainly aimed at the value range and regulatory restrictions of regulatory measures. For example, the concentration of drugs released into the pond must not exceed the safety threshold specified by the state, which can be expressed as g1(D): d1≤MAX DOSE , where d1 represents the drug dose, MAX DOSE is the maximum allowable dose; weak constraints are for the intermediate performance of ecological processes, for example, requiring the diurnal fluctuation of dissolved oxygen to be controlled within a certain range, which can be expressed as g2(D): |DO max -DO min |≤Δ DO , where DO max and DO min are the predicted maximum and minimum values ​​of dissolved oxygen, Δ DO is the allowed fluctuation range.

[0145] Step S403, introducing adaptive crossover and mutation operations on the basis of the standard non-dominated sorting genetic algorithm to solve the multi-objective optimization function;

[0146] For example, in order to solve complex multi-objective optimization problems, this embodiment adopts an improved non-dominated sorting genetic algorithm (NSGA-III). The main steps of the algorithm include: initializing the population, non-dominated sorting, crowding calculation, selection, crossover, mutation and elite retention. Specifically, based on the standard NSGA-III, this embodiment introduces adaptive crossover and mutation operations to improve the convergence speed of the algorithm and the diversity of solutions. The crossover probability and mutation probability It will be dynamically adjusted according to the diversity of the population, and the calculation formula is as follows:

[0147]

[0148] in, and is the basic crossover and mutation probability, f′ is the better fitness value of the two individuals involved in the crossover, f is the fitness value of the current individual, and f max and fmin are the maximum and minimum fitness values ​​in the population, respectively, and α and β are adjustment parameters.

[0149] Step S403, constructing a dynamic model of the pond ecosystem, performing parameter estimation and verification based on the historical ecological data of the pond, and predicting the ecological change trend of the pond in the short term, wherein the historical ecological data includes the feature set F;

[0150] For example, in order to predict the future ecological state, a dynamic model of a pond ecosystem is constructed, which describes the time evolution of key ecological factors such as dissolved oxygen, ammonia nitrogen, phytoplankton, etc. based on a set of ordinary differential equations. The general form of the dynamic model is: dβ / dt=f(β,D,t), where β is the state vector of the ecological factor, D is the control measure vector, t is time, and f is a nonlinear function describing the system dynamics; the dynamic model estimates and verifies parameters through historical data of the pond, and can accurately predict the trend of ecological environment changes in the pond in the short term, such as within 7 days.

[0151] Step S404: Based on the multi-objective optimization function and the dynamic model, a group of Pareto optimal solutions is output, each solution represents a possible regulation strategy, and a final ecological regulation strategy is selected from the group of Pareto optimal solutions;

[0152] Exemplarily, according to the specific conditions of the pond ecosystem, such as the warning level, operating conditions, etc., the most suitable one is selected from the group of Pareto optimal solutions as the final control strategy, and the selection process adopts a multi-criteria decision-making method based on fuzzy logic, which comprehensively considers various indicators of the ecological control strategy and the current decision-making preference of the pond ecosystem;

[0153] Specifically, the generated control strategy includes not only specific operation instructions such as increasing the aerator operation time by 3 hours, adding 50 mL / m 3 etc., and also includes the expected effects and precautions after the implementation of each operation instruction.

[0154] Step S405, according to the ecological regulation strategy, controlling each execution mechanism to carry out actual regulation actions to complete closed-loop ecological regulation;

[0155] Exemplarily, this embodiment converts abstract ecological control strategies into specific equipment control instructions and ensures that these instructions are executed accurately and timely. The core of the execution is a distributed control network, including a central controller, multiple field control units and various execution devices, which can accurately convert complex control strategies into actual management actions and realize the refined control of the pond ecosystem. The adaptability and robustness of the system ensure that the ecological balance of the pond can be maintained in the face of various unexpected situations, greatly improving the stability and output efficiency of aquaculture.

[0156] In this embodiment, multi-dimensional data of the pond is collected in real time, the feature set F of the pond is extracted, the abnormal condition of the pond ecosystem is diagnosed in real time through the intelligent diagnosis system, the abnormal diagnosis results and corresponding early warning information are generated and displayed, and based on the abnormal diagnosis results and early warning information, an ecological regulation strategy is generated in real time, and each actuator is controlled through a distributed control network to carry out actual regulation actions to complete a closed-loop ecological regulation. This embodiment provides a pond ecosystem regulation scheme with accurate abnormal diagnosis and a high degree of automation, which can achieve comprehensive perception, timely diagnosis and optimized management of the pond ecosystem.

[0157] Embodiment 2:

[0158] As attached Figure 3 As shown, Embodiment 2 of the present invention provides an intelligent pond ecological management system 200 as follows:

[0159] The data collection layer 201 is used to collect multi-dimensional data of the pond in real time through sensors arranged inside and outside the pond;

[0160] The data processing layer 202 is used to perform data preprocessing on the multi-dimensional data of the pond to extract the feature set F of the pond, wherein the data preprocessing includes but is not limited to operations such as outlier processing, missing value filling, time alignment and dedimensionalization;

[0161] The diagnosis and early warning layer 203 is used to establish an intelligent diagnosis system based on the feature set F, and to perform real-time abnormal condition diagnosis on the pond ecosystem based on the intelligent diagnosis system, and to generate and display abnormal diagnosis results and corresponding early warning information;

[0162] The ecological control layer 204 generates the optimal control strategy based on the real-time diagnosis results and the warning level. According to the generated control strategy, it controls each executive agency to carry out actual control actions to complete the closed-loop ecological regulation.

[0163] In this embodiment, by setting up a data collection layer, a data processing layer, a diagnosis and early warning layer, and an ecological regulation layer, multi-dimensional pond data is collected and processed, abnormal diagnosis of the pond ecosystem is performed, and finally automatic adjustment is performed to achieve high-precision automated regulation and management of the pond ecosystem.

[0164] Embodiment 3:

[0165] As attached Figure 4 As shown, Embodiment 3 of the present invention provides an intelligent pond ecological management device 400, including:

[0166] Memory 401, used for storing intelligent pond ecological management program;

[0167] A processor 402, wherein the processor 402 is configured to execute the intelligent pond ecological management method described in Example 1;

[0168] The processor 402 and the memory 401 are connected via a bus 403 , which includes but is not limited to an I2C (Inter-integrated Circuit) bus.

[0169] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. This application is intended to cover any modification, use or adaptation of the present invention, which follows the general principles of the present invention and includes common knowledge or customary techniques in the art that are not disclosed by the present invention. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present invention are indicated by the claims. It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. An intelligent pond ecological management method, characterized in that: include: Data collection: sensors placed inside and outside the pond are used to collect multi-dimensional data of the pond in real time. The multi-dimensional data includes pond environmental data and pond biological data, including but not limited to water quality parameters, meteorological conditions, and biological population density; Data processing: pre-process the multi-dimensional data of the pond and convert the multi-dimensional data into a unified three-dimensional data cube. , and extract the feature set F of the pond, the data preprocessing includes: processing outliers by combining the threshold method based on domain knowledge and the statistical 3σ criterion; identifying missing values ​​in the multi-dimensional data and filling them; performing time alignment by interpolation or downsampling; dedimensionalization; Abnormal diagnosis: Based on the feature set F, an intelligent diagnosis system is established. Based on the intelligent diagnosis system, real-time abnormal condition diagnosis of the pond ecosystem is performed, and abnormal diagnosis results and corresponding early warning information are generated and displayed; Ecological regulation: Based on the abnormal diagnosis results and early warning information, the ecological regulation strategy is generated in real time, and each actuator is controlled through a distributed control network to carry out actual regulation actions to complete the closed-loop ecological regulation; The establishment of the intelligent diagnosis system is specifically as follows: An ecological status assessment model is constructed based on a training data set, wherein the training data set includes historically collected multi-dimensional pond data, a feature set F corresponding to the multi-dimensional pond data, and a pond ecological status assessment result annotated by aquaculture experts, wherein the ecological status is divided into three categories: "healthy", "sub-healthy" and "abnormal"; The feature set is evaluated by the ecological status assessment model. Anomaly detection is performed to obtain the anomaly diagnosis result, which includes a health metric score of the pond ecosystem. The calculation formula of the health metric score is as follows: ,in, and are the probabilities of the pond being in "abnormal" and "sub-healthy" states, respectively, and the health metric score ranges from 0 to 100; The real-time generation of ecological regulation strategy includes: Construct a comprehensive regulatory decision space, abstract each possible regulatory measure for the pond ecosystem into a decision variable, and give it a corresponding numerical representation to form a high-dimensional decision vector D; Construct a dynamic model of the pond ecosystem, estimate and verify parameters based on the historical ecological data of the pond, and predict the ecological change trend of the pond in the short term; The environmental regulation problem of the pond ecosystem is transformed into a multi-objective optimization function maximum problem, and the following objectives are comprehensively considered: ecological benefits , calculated based on the health metric score and warning level, and used to evaluate the degree of improvement of the pond ecosystem by regulatory measures; economic benefits , used to calculate the costs and expected benefits of regulatory measures; operational complexity , which is used to measure the difficulty of implementing regulatory measures. The optimization problem of the pond ecological environment is formally expressed as: ;Satisfy the constraints: ,in, , , is the weight coefficient of each target; and Represent inequality constraints and equality constraints, respectively, to ensure that the control strategy meets various practical constraints; Based on the standard non-dominated sorting genetic algorithm, adaptive crossover and mutation operations are introduced to solve the multi-objective optimization function; Based on the multi-objective optimization function and the dynamic model, a set of Pareto solutions is output, each solution represents a possible regulation strategy, and a final ecological regulation strategy is selected from the set of Pareto solutions; The general form of the kinetic model is: ,in is the state vector of ecological factors, is the control measure vector, t is the time, is a nonlinear function that describes the dynamics of the system.

2. The intelligent pond ecological management method according to claim 1, characterized in that: The data preprocessing is to convert the multi-dimensional data into a unified three-dimensional data cube , specifically: A method combining a threshold method based on domain knowledge and a statistical 3σ criterion is used to identify and process outliers in the multi-dimensional data; Identify missing values ​​in the multi-dimensional data. For short-term random missing values, use linear interpolation to fill them. For long-term continuous missing values, use the mean of historical data for the same period to fill them. The historical data for the same period is preferably selected from the same time period within a week. The short time includes but is not limited to a time period of less than 15 minutes, and the long time includes but is not limited to a time period of more than 1 hour. Using the minimum time granularity as a benchmark, align the multi-dimensional data collected by all sensors to a unified time point through interpolation or downsampling; The Z-score standardization method is used to transform the data of different indicators in the multi-dimensional data into a standard normal distribution with a mean of 0 and a standard deviation of 1. The calculation formula is: ,in, is the original value, is the mean value of this indicator, is the standard deviation.

3. The intelligent pond ecological management method according to claim 2, characterized in that: The step of extracting the feature set F of the pond comprises: From the three-dimensional data cube Extract but not limited to time series features, spatial features and multi-dimensional features to form a feature set , where each feature All of them are three-dimensional data cubes An abstract representation in a particular form; The time series characteristics include: mean, variance, skewness, kurtosis, autocorrelation coefficient and partial autocorrelation coefficient; The spatial features include local features and spatial autocorrelation index, and the calculation formula is: , where N is the number of monitoring points, W is the sum of the spatial weight matrix, is the spatial weight, and are the observed values ​​at different locations, is the mean.

4. The intelligent pond ecological management method according to claim 1, characterized in that: The anomaly detection is accomplished through the three main components of the anomaly detector, state classifier and cause analyzer of the ecological status assessment model, specifically including: The anomaly detector adopts the isolation forest algorithm. For a data sample The anomaly score of is calculated as follows: ,in It is a sample The average path length, is given The average path length of samples, the closer the anomaly score is to 1, the data sample The greater the probability of being an outlier; The state classifier uses an ensemble learning method, combined with random forest and gradient boosting decision tree, to perform state classification prediction on the feature set F; The prediction results of random forest and gradient boosting decision tree for feature set F are fused by weighted average. The weights are dynamically adjusted according to their performance on the validation set. The final ecological status classification probability calculation formula of the pond ecosystem is as follows: ,in, and are the weights of random forest and gradient boosted decision tree, respectively. and are the probabilities of classifying sample x as y by random forest and gradient boosting decision tree respectively; The cause analyzer interprets the abnormal detection result of the ecological status assessment model based on the Shapley addition explanation value; For the features in the feature set F , its Shapley addition interpretation value is calculated as follows: ,in, Representation characteristics The Shapley addition interpretation value is, is the set of all features, Does not contain features A subset of is a feature set The anomaly detection results of the ecological status assessment model described below; by calculating the Shapley addition explanation value of each feature, determine which factors contribute most to the generation of abnormal status, and provide a basis for subsequent regulatory strategies.

5. An intelligent pond ecological management system, characterized in that: The system adopts a layered design, including: The data collection layer is used to collect multi-dimensional data of the pond in real time through sensors arranged inside and outside the pond. The multi-dimensional data includes pond environmental data and pond biological data, including but not limited to water quality parameters, meteorological conditions and biological population density; The data processing layer is used to pre-process the multi-dimensional data of the pond and convert the multi-dimensional data into a unified three-dimensional data cube. , and extract the feature set F of the pond, the data preprocessing includes: processing outliers by combining the threshold method based on domain knowledge and the statistical 3σ criterion; identifying missing values ​​in the multi-dimensional data and filling them; performing time alignment by interpolation or downsampling; dedimensionalization; The diagnosis and early warning layer is used to establish an intelligent diagnosis system based on the feature set F, and to perform real-time abnormal condition diagnosis on the pond ecosystem based on the intelligent diagnosis system, and to generate and display abnormal diagnosis results and corresponding early warning information; The ecological control layer combines the real-time diagnosis results and warning levels to generate the optimal control strategy. According to the generated control strategy, each executive agency is controlled to carry out actual control actions to complete the closed-loop ecological regulation. The establishment of the intelligent diagnosis system is specifically as follows: An ecological status assessment model is constructed based on a training data set, wherein the training data set includes historically collected multi-dimensional pond data, a feature set F corresponding to the multi-dimensional pond data, and a pond ecological status assessment result annotated by aquaculture experts, wherein the ecological status is divided into three categories: "healthy", "sub-healthy" and "abnormal"; The feature set is evaluated by the ecological status assessment model. Anomaly detection is performed to obtain the anomaly diagnosis result, which includes a health metric score of the pond ecosystem. The calculation formula of the health metric score is as follows: ,in, and are the probabilities of the pond being in "abnormal" and "sub-healthy" states, respectively, and the health metric score ranges from 0 to 100; Generate ecological control strategies in real time, including: Construct a comprehensive regulatory decision space, abstract each possible regulatory measure for the pond ecosystem into a decision variable, and give it a corresponding numerical representation to form a high-dimensional decision vector D; Construct a dynamic model of the pond ecosystem, estimate and verify parameters based on the historical ecological data of the pond, and predict the ecological change trend of the pond in the short term; The environmental regulation problem of the pond ecosystem is transformed into a multi-objective optimization function maximum problem, and the following objectives are comprehensively considered: ecological benefits , calculated based on the health metric score and warning level, and used to evaluate the degree of improvement of the pond ecosystem by regulatory measures; economic benefits , used to calculate the costs and expected benefits of regulatory measures; operational complexity , which is used to measure the difficulty of implementing regulatory measures. The optimization problem of the pond ecological environment is formally expressed as: ;Satisfy the constraints: ; ,in, , , is the weight coefficient of each target; and Represent inequality constraints and equality constraints, respectively, to ensure that the control strategy meets various practical constraints; Based on the standard non-dominated sorting genetic algorithm, adaptive crossover and mutation operations are introduced to solve the multi-objective optimization function; Based on the multi-objective optimization function and the dynamic model, a set of Pareto solutions is output, each solution represents a possible regulation strategy, and a final ecological regulation strategy is selected from the set of Pareto solutions; The general form of the kinetic model is: ,in is the state vector of ecological factors, is the control measure vector, t is the time, is a nonlinear function that describes the dynamics of the system.

6. An intelligent pond ecological management device, characterized in that: include: A memory for storing an intelligent pond ecological management program; A processor, wherein the processor is configured to execute the intelligent pond ecological management method according to any one of claims 1 to 4; The processor and the memory are connected via a bus, which includes but is not limited to an I2C bus.

Images