Fish-farming tail water nitrogen content dynamic prediction method based on learning algorithm

Abstract

The invention provides a dynamic prediction method for the nitrogen content of fish culture tail water based on a learning algorithm, and relates to the technical field of dynamic prediction, and the method comprises the steps: carrying out water quality simulation of different culture densities, and extracting the simulated culture density of the simulated water quality; collecting multivariate water quality parameters, and performing feature extraction to obtain a feature data set; constructing a nitrogen content prediction model based on the simulated breeding density and the feature data set; and collecting real-time water quality parameters, inputting the real-time water quality parameters into the nitrogen content prediction model, and outputting a nitrogen content prediction value. According to the method and the device, the technical problem that in the prior art, due to low prediction accuracy, the real-time change of the nitrogen content in the fish culture tail water cannot be accurately mastered, and reasonable application of the inorganic nitrogen fertilizer is influenced can be solved, the technical aims of improving the prediction accuracy and the real-time monitoring capability of the nitrogen content are achieved, and the application prospect is wide. The technical effects of precise fertilization and improvement of crop growth efficiency and resource utilization efficiency are achieved.

Description

Technical Field

[0001] The present application relates to the field of dynamic prediction technology, and in particular to a method for dynamically predicting the nitrogen content of tail water in fish farming based on a learning algorithm. Background Art

[0002] The nitrogen content in the tail water of factory-farmed fish is often high and cannot be discharged directly into the water or soil, otherwise it will damage the ecological environment. However, if nitrogen is removed through water treatment, the cost is high and the treatment efficiency is low. Therefore, vegetables can be used to absorb and transform nitrogen, reducing the nitrogen content in the tail water of fish farming. At the same time, the tail water of fish farming can provide nitrogen nutrients for crops, realizing the planting and breeding cycle.

[0003] At present, in the process of planting and breeding cycle, since the nitrogen in the tail water of fish farming cannot completely replace inorganic fertilizers, it is often necessary to mix it with inorganic nitrogen fertilizers to achieve the nitrogen concentration required by crops. However, the nitrogen content in the tail water of fish farming is affected by the breeding density, which leads to changes in the nitrogen content in the tail water of fish farming, which brings difficulties to the application of inorganic fertilizers. Therefore, real-time dynamic prediction of the nitrogen content in the tail water of fish farming becomes the key to accurate application. In the prior art, due to the low prediction accuracy, it is impossible to accurately grasp the real-time changes of the nitrogen content in the tail water of fish farming, thereby affecting the technical problem of the reasonable application of inorganic nitrogen fertilizers. Summary of the invention

[0004] The purpose of this application is to provide a method for dynamically predicting the nitrogen content in fish tail water based on a learning algorithm, so as to solve the technical problem in the prior art that the real-time changes of the nitrogen content in fish tail water cannot be accurately grasped due to low prediction accuracy, thereby affecting the rational application of inorganic nitrogen fertilizers.

[0005] In view of the above problems, the present application provides a method for dynamically predicting the nitrogen content in fish tail water based on a learning algorithm, including: performing water quality simulation at different breeding densities, and extracting simulated breeding density of simulated water quality; collecting multivariate water quality parameters, and performing feature extraction to obtain a feature data set; constructing a nitrogen content prediction model based on the simulated breeding density and the feature data set; collecting real-time water quality parameters and inputting them into the nitrogen content prediction model, and outputting a predicted nitrogen content value.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0007] By simulating water quality at different breeding densities, the simulated breeding density of the simulated water quality is extracted; multiple water quality parameters are collected, and feature extraction is performed to obtain a feature data set; a nitrogen content prediction model is constructed based on the simulated breeding density and the feature data set; real-time water quality parameters are collected and input into the nitrogen content prediction model, and a nitrogen content prediction value is output. In other words, by achieving the technical goal of improving the prediction accuracy of nitrogen content and the real-time monitoring capability, the technical effect of precise fertilization, improving crop growth efficiency and resource utilization efficiency is achieved.

[0008] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented according to the contents of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are specifically cited below. It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easy to understand through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the present application or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0010] Figure 1 A schematic diagram of a flow chart of a method for dynamically predicting nitrogen content in tail water of fish farming based on a learning algorithm for this application;

[0011] Figure 2 This is a flow chart of extracting features to obtain feature data sets in a method for dynamically predicting nitrogen content in tail water of fish farming based on a learning algorithm in this application. DETAILED DESCRIPTION

[0012] This application provides a dynamic prediction method for nitrogen content in fish tail water based on a learning algorithm, which solves the technical problem in the prior art that the real-time changes in nitrogen content in fish tail water cannot be accurately grasped due to low prediction accuracy, thereby affecting the reasonable application of inorganic nitrogen fertilizers. The technical goal of improving the prediction accuracy and real-time monitoring capability of nitrogen content is achieved, and the technical effect of precise fertilization, improving crop growth efficiency and resource utilization efficiency is achieved.

[0013] Below, the technical solutions in the present application will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments of the present application. It should be understood that the present application is not limited to the example embodiments described herein. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application. It should also be noted that, for the convenience of description, only the parts related to the present application are shown in the accompanying drawings, rather than all of them.

[0014] Please see attached Figure 1 The present application provides a method for dynamically predicting the nitrogen content in tail water of fish farming based on a learning algorithm, which specifically comprises the following steps:

[0015] Step 1: Conduct water quality simulation at different aquaculture densities and extract simulated aquaculture density with simulated water quality.

[0016] Specifically, by simulating environments with different aquaculture densities, the water quality changes in actual aquaculture environments can be reproduced, so as to more accurately analyze the nitrogen content at each density. Among them, multiple gradients of aquaculture density are set, for example, simulating the change from low density to high density to cover different aquaculture scenarios, so that the model can learn widely applicable features.

[0017] Next, under each density environment, by collecting water quality parameters such as dissolved oxygen, conductivity, redox potential, etc., the corresponding water quality characteristic values ​​are generated, thereby preliminarily determining the water quality characteristics under this environment. In order to further improve the accuracy of the simulation, the water quality characteristic values ​​are smoothed or averaged to reduce the impact of a single abnormal measurement.

[0018] Then, water quality data of known density are input into the corresponding relationship to generate a characteristic relationship that matches the actual aquaculture conditions, providing a basis for model prediction, and then accurately predicting the simulated aquaculture density of the water body based on any combination of water quality parameters.

[0019] Step 2: Collect multiple water quality parameters and perform feature extraction to obtain feature data sets.

[0020] Specifically, sensors collect data on a variety of water quality parameters, including conductivity, pH, temperature, dissolved oxygen, redox potential, etc., which provide the model with the basis for water quality characteristics under different environments. In order to ensure the representativeness and stability of the data, sensor data is usually collected in a time series, such as every 15 minutes or 30 minutes, continuously recording the water quality conditions at different time points.

[0021] Next, the collected multivariate water quality data is preprocessed to remove noise and abnormal data, thereby improving the quality of the model input data. Preprocessing methods include smoothing filtering and missing value filling, which can ensure the effectiveness of subsequent feature extraction. For example, if dissolved oxygen suddenly changes or is missing over a period of time, the gap can be filled by interpolation or mean filling to ensure data continuity.

[0022] After the data quality is processed, features that are helpful for nitrogen content prediction are extracted to form a feature data set, which provides the key input required for model training. Feature extraction can be based on the original water quality parameters, and some derived features can also be introduced, such as the average dissolved oxygen value or the maximum change rate of conductivity in the past 3 hours, which can capture the trend and periodicity of water quality changes.

[0023] As the characteristic data set is continuously enriched, including processed multivariate water quality parameters and various derived statistical characteristics, the characteristic data set formed has high representativeness and predictive power, and can provide reliable input for nitrogen content prediction. The final characteristic data set can help the model more accurately identify and predict the impact of water quality parameters on nitrogen content, thereby providing scientific basis and technical support for dynamic water quality prediction.

[0024] Step three: constructing a nitrogen content prediction model based on the simulated breeding density and the characteristic data set.

[0025] Specifically, the simulated aquaculture density and the generated characteristic data set are combined to ensure that the data used in the model can fully cover the water quality characteristics and nitrogen changes under actual aquaculture conditions. In order to conduct effective training based on the simulated data, the water quality characteristic data under different densities are combined to form a comprehensive training data set containing simulated aquaculture density. The training data set can better represent the nitrogen content in different density environments and provide a solid data foundation for model construction.

[0026] Next, in the process of building the model, a specific machine learning algorithm (such as BP neural network, support vector machine, etc.) is used to input the breeding density and feature data set into the model so that the model can learn the relationship between the features and the nitrogen content. The model will learn the influence of these features on the nitrogen content to form the ability to predict nitrogen concentration.

[0027] In the process of model training, by continuously iterating and optimizing model parameters, the prediction accuracy can be gradually improved, so that the model can more accurately reflect the impact of different water quality conditions on nitrogen content. The prediction performance of the model is tested by using a validation set or cross-validation method to ensure that it has good generalization ability on new data.

[0028] Step 4: Collect real-time water quality parameters and input them into the nitrogen content prediction model to output the nitrogen content prediction value.

[0029] Specifically, in order for the nitrogen content prediction model to provide real-time predictions, it is first necessary to continuously collect real-time data on water quality through sensors. Real-time data includes parameters such as dissolved oxygen, conductivity, temperature, pH value and redox potential, which are dynamically updated as aquaculture conditions and time change.

[0030] Next, the collected real-time water quality parameter data is input into the nitrogen content prediction model to ensure that the model predicts nitrogen content based on the latest environmental conditions. Since the model has learned the relationship between different water quality characteristics and nitrogen content during the construction process, after entering these parameters, the model can quickly calculate and generate a predicted nitrogen content value.

[0031] When the model outputs the predicted value of nitrogen content, it serves as real-time feedback to help aquaculture managers adjust water quality treatment measures in a timely manner. This prediction method can not only warn of potential water quality problems, but also effectively assist in the management of aquaculture density and the optimization of water quality, thereby avoiding the impact of excessive nitrogen content on the aquaculture environment.

[0032] The dynamic prediction method of nitrogen content in fish tail water based on a learning algorithm can achieve the technical goal of improving the prediction accuracy and real-time monitoring capability of nitrogen content, and achieve the technical effect of precise fertilization, improving crop growth efficiency and resource utilization efficiency.

[0033] Further, if Figure 2 As shown, this application also includes:

[0034] Extract original features of the multivariate water quality parameters, and generate derived features based on the original features, wherein the original features include conductivity, pH value, temperature, dissolved oxygen and redox potential; perform feature fusion on the original features and the derived features to obtain features to be predicted; perform feature data matching based on the features to be predicted to obtain the feature data set.

[0035] Specifically, when building a prediction model, it is first necessary to extract the original features of the collected multivariate water quality parameters, and use the conductivity, pH value, temperature, dissolved oxygen, redox potential, etc. directly measured by each sensor as the initial basic features. These parameters represent the direct characteristics of different water quality conditions. Conductivity can reflect the changes in ion concentration in the water body, while dissolved oxygen can reflect the level of oxygen content in the water body. In a high-density aquaculture environment, the conductivity may reach 900 microsilicates per centimeter and the pH value is about 7.5. These values ​​provide direct water quality information for the model.

[0036] In order to enable the model to capture the complex changing trends of water quality, derived features are further generated based on the original features. Derived features are obtained by performing different statistical operations or combinations on the original features, such as calculating the average value of dissolved oxygen or the rate of change of pH value in the previous 3 hours, which can reflect the dynamic trend of water quality. Derived features can help the model identify deeper laws, such as the relationship between the fluctuation frequency of dissolved oxygen and aquaculture density. When the dissolved oxygen in the previous hour is 5 mg / L and the previous 2 hours is 5.5 mg / L, its decline rate can be used as a derived feature for model reference.

[0037] After obtaining the original features and derived features, they are integrated together through feature fusion to generate the final features to be predicted. The feature fusion process combines the two types of features into a more comprehensive data structure, providing rich information for the model, so that the model can more accurately predict the nitrogen content. For example, combining the change rate of conductivity and dissolved oxygen can more clearly characterize the eutrophication level of the water body.

[0038] After feature fusion is completed, the model will perform feature data matching based on these features to be predicted, and compare them with historical data to build a feature data set. The feature data matching process ensures that the model can use similar data from the past to improve the accuracy of predictions. By matching similar feature data from the past, the model can obtain reliable training data.

[0039] The original feature extraction, derived feature generation, feature fusion and data matching process together constitute the key steps of constructing the feature data set of the prediction model. Through this series of operations, the model can accurately predict the nitrogen content under different conditions based on comprehensive and rich feature data, providing a scientific basis for water quality management.

[0040] Furthermore, the present application also includes:

[0041] According to the original features, time features, sliding window features and interaction features are extracted to generate initial derived features; according to the initial derived features, correlation analysis is performed with the original features to obtain the derived features that meet the correlation threshold.

[0042] Specifically, in order to further enrich the input features of the model, time features, sliding window features and interaction features can be extracted based on the original features to generate initial derived features. Time features refer to some time-related information, such as hours, days of the week or seasons, which can help the model identify the periodic changes in nitrogen content. Sliding window features reflect the dynamic trend of features by calculating feature statistics over a period of time (such as the average or maximum value in the past 3 hours). Interaction features combine multiple original features, for example, multiplying conductivity and pH value to generate combined features that can capture the relationship between features.

[0043] Next, to ensure that the generated initial derived features contribute substantially to the prediction results, correlation analysis is performed between them and the original features to determine which features meet the correlation requirements. Correlation analysis can quantify the relationship between each derived feature and the target variable. Usually, a correlation threshold is set to screen out derived features that meet the threshold, thereby retaining the most useful features for the model to predict nitrogen content. For example, when the correlation of the sliding window feature exceeds 0.7, it indicates that it has a strong explanatory power for changes in nitrogen content and can be used as part of the final model input.

[0044] By selecting the most effective features from the abundant initial derived features, reliable data support is provided for the accurate prediction of nitrogen content.

[0045] Furthermore, the present application also includes:

[0046] The mutual information of the original feature is scored according to the derived feature to obtain a derived feature score; the derived features within a feature selection threshold are screened based on the derived feature score; and the original feature is regressed on the screened derived feature to obtain the feature to be predicted.

[0047] Specifically, the mutual information score of the original features based on the derived features can quantify the relationship between each derived feature and the original feature. The mutual information score is a statistic that reflects the degree of information sharing between two variables. For example, if the mutual information score between a derived feature and conductivity is zero, it means that the two features have no correlation, while a score of five indicates that there is a strong relationship between them.

[0048] Filtering derived features within the feature selection threshold based on derived feature scores means selecting features with mutual information scores above a certain threshold to improve the predictive performance of the model. The feature selection threshold can be set according to the specific data distribution. For example, in some cases, the threshold may be set to three, that is, only derived features with mutual information scores greater than three can be selected into the model, thereby removing redundant or irrelevant features.

[0049] By regressing the original features on the screened derived features, we can establish the relationship between the derived features and the target variables through regression analysis, thereby obtaining the features to be predicted, helping to identify important features, and improving the accuracy of the prediction model.

[0050] By quantifying the relationship between features through mutual information scoring, effective derived features are further screened out, and then the relationship with the original features is established through regression analysis to obtain the features to be predicted. This can more accurately predict nitrogen content and help aquaculture managers to effectively regulate water quality.

[0051] Furthermore, the present application also includes:

[0052] Collect fish tail water samples to collect sample nitrogen content, establish a density-content relationship between sample breeding density and sample nitrogen content; use fish feces water solution to simulate water quality at different breeding densities, and measure the simulated nitrogen content based on the simulated water quality; solve the simulated nitrogen content based on the density-content relationship to obtain the simulated breeding density.

[0053] Specifically, fish tail water samples are collected to collect the nitrogen content of the samples, and a density-content relationship between the sample farming density and the sample nitrogen content is established to obtain the true nitrogen content in the water samples under different farming densities. For example, when the farming density is 20 fish per cubic meter, the nitrogen content in the tail water sample can be measured to be three milligrams per liter. By collecting multiple sets of data, a relationship diagram between density and nitrogen content can be drawn to help subsequent analysis.

[0054] Using a fish manure water solution to simulate water quality at different stocking densities, and measuring the simulated nitrogen content based on the simulated water quality, means understanding the impact of different stocking densities on nitrogen content through a laboratory simulation environment. For example, if the stocking density is 50 fish per cubic meter, the nitrogen content in the simulated water sample is measured to be 8 mg per liter, providing data support for the model.

[0055] Solving the simulated nitrogen content based on the density-content relationship to obtain the simulated stocking density shows that the specific stocking density can be inferred from the established density-content relationship. For example, if the simulated nitrogen content is known to be eight milligrams per liter, the corresponding stocking density can be calculated as fifty fish per cubic meter using the previously established relationship.

[0056] By collecting fish tail water samples to establish a density-content relationship, using fish feces water solution to simulate water quality, and using this relationship to reverse simulate breeding density, we can help accurately understand the changes in nitrogen under different breeding densities. This not only provides a scientific basis for water quality management, but also provides data support for the reasonable regulation of breeding density.

[0057] Furthermore, the present application also includes:

[0058] The simulated nitrogen content corresponding to the simulated breeding density is extracted according to the density-content relationship, and the stability of the nitrogen content is calculated; the preliminary sampling frequency is determined based on the stability of the nitrogen content; a self-regulating mechanism is introduced to monitor the rate of change of the water quality environment to adjust the preliminary sampling frequency and generate a sampling frequency; the multivariate water quality parameters are collected according to the sampling frequency to obtain the multivariate water quality parameters.

[0059] Specifically, the simulated nitrogen content corresponding to the simulated breeding density is extracted according to the density-content relationship, the stability of the nitrogen content is calculated, and the nitrogen content under a specific breeding density is found through the established relationship model, and its stability is further analyzed. For example, if the breeding density is thirty fish per cubic meter, the simulated nitrogen content is six milligrams per liter, and the stability can be obtained by calculating the change in nitrogen content at different time points. If the fluctuation range does not exceed one milligram per liter, it indicates that the nitrogen content is relatively stable.

[0060] Determining the initial sampling frequency based on the stability of nitrogen content means that the sampling frequency is determined by analyzing the stability of nitrogen content. If the stability is high, for example, the nitrogen content does not change by more than 0.5 mg / L within an hour, a longer sampling interval can be selected, such as sampling every three hours; on the contrary, if the stability is low, the sampling frequency needs to be increased, such as sampling once an hour, to capture the changes in nitrogen content in a timely manner.

[0061] The introduction of a self-regulating mechanism to monitor the rate of change of the water quality environment to adjust the preliminary sampling frequency and generate a sampling frequency means dynamically adjusting the sampling frequency according to real-time water quality data. For example, if the rate of change of water quality is monitored to be fast, such as the dissolved oxygen fluctuates by more than one milligram per liter in a short period of time, the sampling frequency may need to be increased, otherwise the frequency can be reduced, thereby ensuring that the sampling can effectively reflect the true situation of the water quality.

[0062] Collecting multiple water quality parameters according to the sampling frequency and obtaining multiple water quality parameters means that water quality data is systematically collected according to the adjusted sampling frequency. At this time, multiple water quality parameters such as conductivity, pH, temperature and dissolved oxygen can be measured regularly to obtain a comprehensive data set, such as recording conductivity and dissolved oxygen values ​​every hour to ensure the timeliness and accuracy of the data.

[0063] By extracting the nitrogen content corresponding to the simulated breeding density and calculating its stability, we can provide a basis for determining the initial sampling frequency. Then, we can dynamically adjust the sampling frequency through the self-regulating mechanism, and finally collect multiple water quality parameters according to this frequency, ensuring that water quality can be effectively monitored and managed under different breeding environments, and timely reflect water quality changes, thereby optimizing breeding conditions.

[0064] Furthermore, the present application also includes:

[0065] According to the simulated breeding density, a corresponding relationship is extracted in the density-content relationship to obtain the corresponding nitrogen content; a BP neural network architecture is constructed, and a training set and a verification set are generated with the corresponding nitrogen content as a label and the feature data set as a feature; the BP neural network architecture is trained based on the training set, and verified based on the verification set, and if the verification result of the BP neural network architecture is greater than a preset verification accuracy rate, the nitrogen content prediction model is obtained; if the verification result is less than / equal to the preset verification accuracy rate, the number of training times is increased until the training rounds are met, and the nitrogen content prediction model is used as the nitrogen content prediction model.

[0066] Specifically, the corresponding relationship extraction is performed based on the density-content relationship of the simulated breeding density to obtain the corresponding nitrogen content, which means that the nitrogen content under a specific breeding density can be found through the existing density-content model. For example, when the simulated breeding density is 40 fish per cubic meter, the model can tell that the corresponding nitrogen content is 5 mg / L, which provides important label data for subsequent model training.

[0067] Constructing a BP neural network architecture, using the corresponding nitrogen content as a label and the feature data set as a feature, generating a training set and a validation set, means that when designing a neural network, the nitrogen content is used as the prediction target, and the previously extracted multivariate water quality features are used as input features. The feature data set may contain data such as conductivity, pH value, and dissolved oxygen. The training set is used for model learning, while the validation set is used to test the generalization ability of the model.

[0068] The BP neural network architecture is trained based on the training set and verified based on the verification set. If the verification result of the BP neural network architecture is greater than the preset verification accuracy, a nitrogen content prediction model is obtained, indicating that after the training is completed, the performance of the model is tested using the verification set. If, after testing, the prediction accuracy of the model on the verification set reaches or exceeds a set threshold, such as a set threshold of 80%, then the model will be considered valid and can be used for practical applications.

[0069] If the verification result is less than or equal to the preset verification accuracy, the number of training rounds will be increased until the training rounds are met. As a nitrogen content prediction model, it means that when the performance of the model does not meet expectations, the number of training rounds needs to be increased to improve the learning ability of the model. For example, if the initial training is fifty times but the accuracy is less than 80%, the number of training rounds will be increased to one hundred times until the model reaches the required accuracy.

[0070] By extracting the corresponding nitrogen content from the density-content relationship, constructing a BP neural network architecture with the nitrogen content as a label, and evaluating the performance of the model through training and verification, an effective nitrogen content prediction model is finally formed, ensuring that the model can accurately predict the nitrogen content based on water quality characteristics, thereby providing a scientific basis for aquaculture management.

[0071] Furthermore, the present application also includes:

[0072] An online learning mechanism is introduced to adapt the nitrogen content prediction model according to the real-time water quality parameters and the predicted nitrogen content value; and the adaptive result is fed back to the nitrogen content prediction model.

[0073] Specifically, the introduction of an online learning mechanism to adapt the nitrogen content prediction model based on real-time water quality parameters and nitrogen content prediction values ​​means that the model can be dynamically adjusted based on the changing water quality data and real-time nitrogen content prediction results. By continuously monitoring water quality parameters and inputting real-time data into the model, it can learn and update its own prediction rules, thereby improving the accuracy of the prediction.

[0074] Feedback to the nitrogen content prediction model based on the adaptive results indicates that the online learning mechanism will feed back the adjusted model parameters and the new knowledge learned to the prediction model, which not only enhances the model's sensitivity to water quality changes, but also enables the model to better reflect the relationship between water quality and nitrogen content in subsequent predictions. For example, if the model learns from real-time data that nitrogen content tends to increase when conductivity is above a certain critical value, then in future predictions, the model will emphasize the impact of this feature more.

[0075] By introducing an online learning mechanism, the model can make adaptive adjustments based on real-time water quality parameters and predicted values, thereby continuously improving the accuracy of the prediction, and feeding the adaptive results back to the model to enhance its learning ability, ensuring that the model can still maintain good predictive performance when facing dynamically changing water quality conditions, providing real-time and accurate data support for aquaculture management.

[0076] In summary, the method for dynamically predicting nitrogen content in tail water of fish farming based on a learning algorithm provided in this application has the following technical effects:

[0077] By simulating water quality at different breeding densities, the simulated breeding density of the simulated water quality is extracted; multiple water quality parameters are collected, and feature extraction is performed to obtain a feature data set; a nitrogen content prediction model is constructed based on the simulated breeding density and the feature data set; real-time water quality parameters are collected and input into the nitrogen content prediction model, and a nitrogen content prediction value is output. In other words, by achieving the technical goal of improving the prediction accuracy of nitrogen content and the real-time monitoring capability, the technical effect of precise fertilization, improving crop growth efficiency and resource utilization efficiency is achieved.

[0078] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

[0079] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the present application and its equivalent technology, the present application is also intended to include these modifications and variations.

Claims

1. A method for dynamically predicting nitrogen content in fish tail water based on a learning algorithm, characterized in that: include: Conduct water quality simulations at different stocking densities and extract simulated stocking densities with simulated water quality; Collect multiple water quality parameters and perform feature extraction to obtain feature data sets; Constructing a nitrogen content prediction model based on the simulated breeding density and the characteristic data set; Real-time water quality parameters are collected and input into the nitrogen content prediction model, and the nitrogen content prediction value is output.

2. A method for dynamically predicting nitrogen content in fish tail water based on a learning algorithm as claimed in claim 1, characterized in that: The step of extracting features to obtain a feature data set includes: Extracting original features of the multivariate water quality parameters and generating derived features based on the original features, wherein the original features include electrical conductivity, pH value, temperature, dissolved oxygen and redox potential; Performing feature fusion on the original features and the derived features to obtain features to be predicted; Feature data matching is performed according to the feature to be predicted to obtain the feature data set.

3. A method for dynamically predicting nitrogen content in fish tail water based on a learning algorithm as claimed in claim 2, characterized in that: The generating of the derived features based on the original features comprises: Extracting time features, sliding window features and interaction features according to the original features to generate initial derived features; A correlation analysis is performed on the initial derived features and the original features to obtain the derived features that meet a correlation threshold.

4. A method for dynamically predicting nitrogen content in fish tail water based on a learning algorithm as claimed in claim 2, characterized in that: The step of fusing the original features and the derived features to obtain the features to be predicted includes: Performing a mutual information score on the original feature according to the derived feature to obtain a derived feature score; Filtering the derived features within a feature selection threshold based on the derived feature scores; The original features are regressed on the filtered derived features to obtain the features to be predicted.

5. A method for dynamically predicting nitrogen content in fish tail water based on a learning algorithm as claimed in claim 1, characterized in that: The water quality simulation of different aquaculture densities and extraction of simulated aquaculture density of simulated water quality include: Collecting tail water samples of fish farming to collect nitrogen content of the samples, and establishing a density-content relationship between the farming density of the samples and the nitrogen content of the samples; Fish feces aqueous solution was used to simulate water quality at different stocking densities, and the simulated nitrogen content was measured based on the simulated water quality; The simulated nitrogen content is solved according to the density-content relationship to obtain the simulated breeding density.

6. A method for dynamically predicting nitrogen content in fish tail water based on a learning algorithm as claimed in claim 5, characterized in that: The collecting of multiple water quality parameters includes: Extracting the simulated nitrogen content corresponding to the simulated breeding density according to the density-content relationship, and calculating the nitrogen content stability; determining a preliminary sampling frequency based on the stability of the nitrogen content; Introducing a self-regulating mechanism to monitor the rate of change of the water quality environment and adjust the preliminary sampling frequency to generate a sampling frequency; The multivariate water quality parameters are collected according to the sampling frequency to obtain the multivariate water quality parameters.

7. A method for dynamically predicting nitrogen content in fish tail water based on a learning algorithm as claimed in claim 5, characterized in that: The step of constructing a nitrogen content prediction model based on the simulated breeding density and the characteristic data set includes: Extracting the corresponding relationship between the density and content according to the simulated breeding density to obtain the corresponding nitrogen content; Constructing a BP neural network architecture, using the corresponding nitrogen content as a label and the characteristic data set as a feature, to generate a training set and a validation set; The BP neural network framework is trained based on the training set, and verified based on the verification set, and if the verification result of the BP neural network framework is greater than a preset verification accuracy rate, the nitrogen content prediction model is obtained; If the verification result is less than / equal to the preset verification accuracy, the number of training times is increased until the training rounds are met, and the nitrogen content prediction model is used as the model.

8. A method for dynamically predicting nitrogen content in fish tail water based on a learning algorithm as claimed in claim 1, characterized in that: The method also includes: Introducing an online learning mechanism to adapt the nitrogen content prediction model according to the real-time water quality parameters and the nitrogen content prediction value; The adaptive result is fed back to the nitrogen content prediction model.

Images