Denitrifying bacteria quantitative feeding method and system for aquaculture

Abstract

The application discloses a nitrogen removal bacteria quantitative feeding method and system for aquaculture, and belongs to the technical field of aquaculture, and the technical scheme points of the application are as follows: the excretion amount of each type of fish in each fish pond and the bait residual amount in a fixed time are obtained; the ammonia nitrogen content in each fish pond in a preset time is predicted according to the excretion amount, the bait residual amount and a long-short time sequence model; the bacteria feeding pond, the bacteria feeding amount and the bacteria feeding type of the bacteria feeding pond in each fish pond are determined according to the ammonia nitrogen content; and the quantitative bacteria feeding of the bacteria feeding pond is completed according to the bacteria feeding amount and the bacteria feeding type. In the aquaculture process, ammonia nitrogen and nitrite can affect the respiration, immunity and growth of fish, and the nitrogen removal bacteria can reduce the concentration of these substances and maintain the balance of the water ecosystem, so that the ammonia nitrogen content in each fish pond is predicted through bait feeding amount, and the bacteria feeding amount and the bacteria feeding type are reasonably planned in combination with the fish type and the growth stage in each fish pond, so that a suitable environment is provided for fish breeding.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of aquaculture, more particularly to a method and system for quantitatively feeding denitrifying bacteria in aquaculture. BACKGROUND

[0002] In the field of aquaculture, the healthy reproduction of fish is a key link to ensure the benefits of aquaculture and the sustainable development of the industry. Water quality, as one of the core factors affecting fish reproduction, directly affects the growth and development of fish, the maturation of gonads, and the success rate of reproduction. Denitrifying bacteria, as an important microorganism for regulating water quality, can convert harmful nitrogen-containing substances such as ammonia nitrogen and nitrite in water, which are highly toxic to fish, into nitrogen gas through nitrification and denitrification, effectively reducing nitrogen pollution in water, and creating a stable and suitable breeding environment for fish. However, the current feeding method of denitrifying bacteria in aquaculture is mostly empirical and extensive, lacking precise quantitative control means, so there are deficiencies in the prior art. SUMMARY

[0003] In view of the deficiencies in the prior art, the present application provides a method and system for quantitatively feeding denitrifying bacteria in aquaculture, which can effectively reduce costs by analyzing and determining the amount of bacteria to be fed based on basic data such as feeding amount, combined with long and short time sequence models.

[0004] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0005] The present application provides a method for quantitatively feeding denitrifying bacteria in aquaculture, comprising:

[0006] Obtaining the excretion amount and the remaining amount of bait of each type of fish in each fish tank within a fixed time;

[0007] According to the excretion amount, the remaining amount of bait, and the long and short time sequence model, the ammonia nitrogen content in each fish tank within a predetermined time is predicted;

[0008] According to the ammonia nitrogen content, determining the bacteria-feeding tank, the amount of bacteria to be fed, and the type of bacteria to be fed in each fish tank;

[0009] According to the amount of bacteria to be fed and the type of bacteria to be fed, completing the quantitative feeding of the bacteria-feeding tank.

[0010] As a further improvement of the present application, the obtaining of the excretion amount and the remaining amount of bait of each type of fish in each fish tank within a fixed time comprises:

[0011] Obtaining the feeding amount, feeding rate, and absorption rate of each type of fish in each fish tank within the fixed time, and the cleaning frequency and water changing frequency of each fish tank;

[0012] According to the feeding amount, the feeding rate, the cleaning frequency and the water changing frequency, the bait residual amount is obtained;

[0013] According to the feeding amount, the absorption rate, the cleaning frequency and the water changing frequency, the excretion amount is obtained.

[0014] As a further improvement of the present application, according to the excretion amount, the bait residual amount and a long-short time sequence model, the ammonia nitrogen content in each fish tank within a preset time is predicted, comprising:

[0015] According to historical breeding data, the long-short time sequence model corresponding to each type of fish is trained;

[0016] The excretion amount and the bait residual amount of each type of fish in each fish tank within the fixed time are input into the long-short time sequence model corresponding to each type of fish, and the ammonia nitrogen content in each fish tank within the preset time is predicted.

[0017] As a further improvement of the present application, the long-short time sequence model corresponding to each type of fish is trained according to historical breeding data, comprising:

[0018] The historical breeding data is divided into a validation set and a test set;

[0019] According to the validation set, an initial long-short time sequence model is trained to obtain an optimized long-short time sequence model;

[0020] According to the test set, the optimized long-short time sequence model is evaluated to obtain the long-short time sequence model corresponding to each type of fish.

[0021] As a further improvement of the present application, according to the validation set, an initial long-short time sequence model is trained to obtain an optimized long-short time sequence model, comprising:

[0022] According to historical breeding data, a relationship curve between the excretion amount, the bait residual amount and the ammonia nitrogen content is obtained;

[0023] According to the relationship curve, a prediction training set is obtained;

[0024] According to the prediction training set, the initial long-short time sequence model is trained to obtain a training model;

[0025] According to the training model and the validation set, the optimized long-short time sequence model is obtained.

[0026] As a further improvement of the present application, according to the training model and the validation set, the optimized long-short time sequence model is obtained, comprising:

[0027] The parameters in the training model are divided into a plurality of parameter groups, and the plurality of parameter groups are sequentially arranged to obtain a first queuing sequence;

[0028] A first iteration operation is performed, which includes solidifying all parameter groups except a first parameter group, updating the parameters in the current training model according to the verification set, until the training model meets a first termination condition, updating the first queuing sequence and repeating the solidification and updating steps until the first queuing sequence is empty, and outputting the current training model as the optimized long-short time sequence model; the first parameter group is the parameter group located at the first position of the first queuing sequence.

[0029] As a further improvement of the application, the initial long-short time sequence model is trained according to the prediction training set to obtain a training model, including:

[0030] The prediction training set is divided into a plurality of training groups, and the plurality of training groups are sequentially arranged to obtain a second queuing sequence;

[0031] A second iteration operation is performed, which includes inputting a first training group to the current long-short time sequence model for forward propagation calculation, calculating a loss function according to the forward propagation calculation result and the historical breeding data, calculating the gradient of the parameters in the current long-short time sequence model according to the loss function and the back propagation algorithm, updating the current long-short time sequence model according to the gradient to obtain an updated long-short time sequence model, updating the second queuing sequence until a second termination condition is met, and outputting the current updated long-short time sequence model as the training model; the first training group is the training group located at the first position of the second queuing sequence; the loss function is a regularization loss function.

[0032] As a further improvement of the application, according to the ammonia nitrogen content, the fish tank for bacteria injection is determined, and the amount and type of bacteria injection in the fish tank for bacteria injection, including:

[0033] According to the ammonia nitrogen content and a preset threshold value, the fish tank for bacteria injection is determined;

[0034] According to the type and growth stage of the fish in the fish tank for bacteria injection, the type of bacteria injection is determined, and the type of bacteria injection includes nitrifying bacteria, denitrifying bacteria and anaerobic ammonia oxidation bacteria;

[0035] According to the denitrification capacity, action time and safety factor corresponding to the type of bacteria injection, the amount of bacteria injection is calculated.

[0036] As a further improvement of the application, according to the amount and type of bacteria injection, the quantitative bacteria injection into the fish tank for bacteria injection is completed, including:

[0037] According to the bacteria-adding amount, the bacteria-adding type, the fish type and the position of the bacteria-adding pool, a comprehensive score of the bacteria-adding pool is determined.

[0038] According to the comprehensive score, the quantitative bacteria-adding of the bacteria-adding pool is sequentially completed.

[0039] In another aspect, the present application provides a method for quantitatively adding denitrifying bacteria to aquaculture, comprising:

[0040] An acquisition module is configured to acquire the excretion amount and the bait remaining amount of each type of fish in each fish pool within a fixed time;

[0041] A calculation module is configured to predict the ammonia nitrogen content in each fish pool within a preset time according to the excretion amount, the bait remaining amount and a long-short time sequence model, and determine a bacteria-adding pool, the bacteria-adding amount and the bacteria-adding type in each fish pool according to the ammonia nitrogen content.

[0042] A control module is configured to complete the quantitative bacteria-adding of the bacteria-adding pool according to the bacteria-adding amount and the bacteria-adding type.

[0043] The present application predicts the ammonia nitrogen content in each fish pool by the bait-adding amount, and reasonably plans the bacteria-adding amount and the bacteria-adding type according to the fish type and the growth stage in each fish pool, so as to provide a suitable environment for fish breeding. Compared with the method of determining the ammonia nitrogen content by water quality detection, the method provided by the present application can effectively reduce the detection cost and improve the decision-making efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 The figure is a schematic diagram of the method steps of the present application;

[0045] Figure 2 The figure is a schematic diagram of the step of training the initial long-short time sequence model of the present application;

[0046] Figure 3 The figure is a schematic diagram of the implementation scenario of the present application Figure 1 ;

[0047] Figure 4 The figure is a schematic diagram of the implementation scenario of the present application Figure 2 ;

[0048] Figure 5 The figure is a schematic diagram of the implementation scenario of the present application Figure 3 . DETAILED DESCRIPTION

[0049] The technical solution of the present application will be described in detail below by means of the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present application and the specific features in the embodiments are detailed descriptions of the technical solution of the present application, rather than limitations of the technical solution of the present application.

[0050] The term "and / or" in the following description is only a description of the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can represent the three cases of A alone, A and B together, and B alone. In addition, the character " / " generally represents that the associated objects before and after are an "or" relationship.

[0051] As Figure 1 shown, the embodiment of the present application provides a method for quantitatively feeding denitrifying bacteria for aquaculture, comprising:

[0052] obtaining the excretion amount and the bait remaining amount of each type of fish in each fish tank within a fixed time;

[0053] According to the excretion amount, the bait remaining amount and the long-short time sequence model, the ammonia nitrogen content in each fish tank within a predetermined time is predicted;

[0054] According to the ammonia nitrogen content, the fish tank for feeding bacteria, the amount of bacteria and the type of bacteria in the fish tank for feeding bacteria are determined;

[0055] According to the amount of bacteria and the type of bacteria, the quantitative feeding of the fish tank for feeding bacteria is completed.

[0056] The fixed time and the predetermined time are determined according to actual needs, for example, the fixed time can be designed as 30 days, that is, the excretion amount and the bait remaining amount of each type of fish in each fish tank within 30 days are obtained, and the predetermined time is designed as 3 days or 5 days.

[0057] The embodiment can predict the ammonia nitrogen content in each fish tank in real time by feeding amount, and reasonably plan the amount of bacteria and the type of bacteria according to the type and growth stage of fish in each fish tank, which can provide a suitable environment for fish breeding. Compared with the way of determining the ammonia nitrogen content by water quality detection, the method provided by the present application can effectively reduce the detection cost and improve the decision efficiency.

[0058] Further, the embodiment provides a step of obtaining the excretion amount and the bait remaining amount of each type of fish in each fish tank within a fixed time, comprising:

[0059] obtaining the bait feeding amount, the feeding rate, the absorption rate of each type of fish in each fish tank within a fixed time, and the cleaning frequency and the water changing frequency of each fish tank;

[0060] According to the bait feeding amount, the feeding rate, the cleaning frequency and the water changing frequency, the bait remaining amount is obtained;

[0061] According to the bait feeding amount, the absorption rate, the cleaning frequency and the water changing frequency, the excretion amount is obtained.

[0062] Specifically, assuming that the fixed time is 30 days, the bait remaining amount F ijx of the jth type of fish in the ith fish tank on the xth day is:

[0063] F ijx =(F ijx-1 +f ijx (1-r ij ))×a α ×b β

[0064] Wherein, F ijx-1 represents the i-th fish pool in the j-th fish species in the x-1 day of the remaining amount of feed, f ijx represents the i-th fish pool in the j-th fish species in the x day of the amount of feed, r ij represents the i-th fish pool in the j-th fish species of the feeding rate, alpha indicates the function related to the cleaning, if the x day to the i-th fish pool has been cleaned, alpha = 1, otherwise alpha = 0, a represents the proportion of the total un-eaten feed taken away by each cleaning, the total un-eaten feed is represented by (F ijx-1 +f ix (1-r j )), beta indicates the function related to the water change, if the x day to the i-th fish pool has been water changed, beta = 1, otherwise beta = 0, b represents the proportion of the total un-eaten feed taken away by each water change, since the water change is usually carried out after cleaning, therefore the total un-eaten feed at this time should be (F ijx-1 +f ix (1-r j ))×a α .

[0065] Wherein, the feeding rate refers to the ratio of fish feeding amount to fish body weight, according to the definition, the specific calculation formula of the feeding rate is:

[0066] r ij =X / J

[0067] Wherein X represents the i-th fish pool in the x day of the j-th fish species of the feeding amount, J represents the i-th fish pool in the j-th fish species of the weight, the fish weight can be obtained by weighing the fish. Specifically, the feeding amount is obtained according to the amount of feed f ijx and the remaining amount of feed F ijx-1 , the remaining amount of feed can be obtained according to the image, for example, the feed can be characterized by marking, such as using feed with specific color or shape, then taking the image of the fish pool after the fish finishes eating, and finally identifying the remaining amount of feed through image recognition.

[0068] The i-th fish pool in the x day of the j-th fish species of the excretion amount M ijx is:

[0069] M ijx =(M ijx-1 +(f ijx ×rij (1-η ij ))×c α ×d β

[0070] wherein, M ijx-1 represents the excretion amount of the jth type of fish in the ith fish tank on the x-1th day, η ij represents the absorption rate of the jth type of fish in the ith fish tank, c represents the proportion of excretion removed by each cleaning, and the total excretion is represented by (M ijx-1 +(f ijx ×r ijk )(1-η ij )), d represents the proportion of excretion removed by each water change, and the total excretion at this time is (M ijx-1 +(f ijx ×r ij )(1-η ij ))×c α .

[0071] The absorption rate reflects the proportion of the part that can be effectively absorbed and utilized by the fish from the ingested feed to the feeding amount. For example, if the absorption rate of a certain fish is 0.8, it means that it can absorb 80% of the ingredients in the ingested feed, and the remaining 20% is excreted in the form of excretion. The specific formula is:

[0072] η ij =I / X

[0073] wherein, I represents the total amount of the part absorbed by the fish from the feed, specifically, I can be obtained according to the ratio of fish weight gain and nutrient conversion coefficient, the fish weight gain can be obtained by weighing the fish, and the nutrient conversion coefficient represents the mass of fish weight gain per unit mass of feed, for example, if the nutrient conversion coefficient is 0.4, it means that the fish gains 0.4 grams of weight for every 1 gram of feed absorbed.

[0074] The nutrient conversion coefficient can be obtained according to the simulation experiment, specifically, according to the daily feeding amount and the protein content in the feed, the total amount of protein ingested by the fish during the experiment can be obtained, then according to the amount of feces collected during the experiment and the protein content in the feces, the total amount of protein excreted by the fish through feces during the experiment can be calculated, and the total amount of absorbed protein can be obtained according to the total amount of ingested protein and the total amount of excreted protein, then the weight gain of the fish body during the experiment is calculated, and the ratio of the weight gain of the fish body to the total amount of absorbed protein is the conversion coefficient of protein; wherein the protein content in the feces can be measured by using the Coomassie brilliant blue method and the Kjeldahl nitrogen determination method.

[0075] Therefore, for different fish species in different growth stages, simulation experiments can be carried out respectively to obtain the corresponding absorption rates of different fish species. Moreover, the present embodiment only gives one implementable manner, but the present embodiment is not limited thereto, and those skilled in the art can further improve the present embodiment to obtain more accurate absorption rates, for example, only one nutrient, protein, is considered in the present embodiment, but other nutrients such as fat and carbohydrate can be further considered in actual application.

[0076] It should be noted that the technical contribution of the present application does not lie in the accurate determination of the absorption rate and the feeding rate, and the absorption rate and the feeding rate of the present application can be within the allowable error range of the industry, and the determination of the above absorption rate and the feeding rate belongs to the commonly known technology in the art, and the present application does not limit and elaborate on this.

[0077] Moreover, a, b, c and d need to be determined according to the actual breeding situation, for example, if the water flow is gentle and there is a filtering measure when changing water, the values of b and d should be within the range of 0.1-0.3, if the water flow is large and the water flow is relatively fast, the value of b should be within the range of 0.4-0.6, the value of d should be within the range of 0.3-0.5, if professional equipment such as siphon is used for cleaning, the value of a should be within the range of 0.3-0.6, the value of c should be within the range of 0.6-0.8, if only manual simple fishing cleaning is used, the value of a should be within the range of 0.1-0.3, and the value of c should be within the range of 0.3-0.5.

[0078] The present embodiment establishes a formula about the remaining amount of bait and the excretion amount based on the feeding rate of the fish species, the absorption rate and the water changing situation and the cleaning situation of the fish pond, and at the same time, the present embodiment considers that, in some large-scale aquaculture ponds, the water flow speed is relatively slow when changing water, it is difficult to flush and take away the bait or excretion that sinks in the low-lying part of the bottom of the pond or is sheltered, and part of the bait has certain viscosity or water absorption, which is easy to adhere to the bottom of the fish pond or the surface of other objects and is difficult to be flushed away by the water flow; and due to the limitation of cleaning tools and methods, the commonly used mesh tool cannot completely clean the small bait and excretion, the siphon device is difficult to clean the bait that is adsorbed on the bottom or wall of the pond, and the bait or excretion located in the suction port and the corner that is difficult to reach, therefore, by setting the indication function and the proportion of the bait and excretion taken away each time cleaning and water changing, the present embodiment can more accurately calculate the remaining amount of bait and the excretion amount.

[0079] Further, the present embodiment provides a step of predicting the ammonia nitrogen content in each fish pond within a preset time according to the excretion amount, the remaining amount of bait and the long-short time sequence model, comprising:

[0080] According to historical breeding data, the long-short time sequence model corresponding to each type of fish species is trained;

[0081] The excretion amount and the bait residual amount of each type of fish in each fish tank within a fixed time are input into the long-short time sequence model corresponding to each type of fish to obtain the ammonia nitrogen content in each fish tank within a preset time.

[0082] To improve the prediction accuracy, the historical aquaculture data used for training should be the aquaculture data within a long time, for example, the historical aquaculture data includes the bait residual amount, the excretion amount and the ammonia nitrogen content of each type of fish per day within a year.

[0083] The long-short time sequence model is used for prediction in this embodiment, which can more accurately capture the time sequence characteristics of the historical aquaculture data and make accurate prediction based on the time sequence characteristics.

[0084] Further, the embodiment provides a step of training the long-short time sequence model corresponding to each type of fish according to the historical aquaculture data, which includes:

[0085] The historical aquaculture data is divided into a validation set and a test set;

[0086] The initial long-short time sequence model is trained according to the validation set to obtain an optimized long-short time sequence model;

[0087] The optimized long-short time sequence model is evaluated according to the test set to obtain the long-short time sequence model corresponding to each type of fish.

[0088] To ensure that each set has sufficient data for the corresponding task, the embodiment sets the proportion of data in the validation set and the test set to the total data as 7:3, which can reserve sufficient data for the test set to optimize the performance of the long-short time sequence evaluation model without making the validation set too scarce, avoiding the case that the evaluation result is inaccurate or not representative due to too little data.

[0089] Further, before training, the embodiment needs to initially design the model to obtain an initial long-short time sequence model, which includes an input layer, an LSTM layer and at least one fully connected layer. The number of neurons in the input layer is equal to the number of input features. The features selected in this embodiment are the excretion amount and the bait residual amount. The number of neurons in the LSTM layer and the fully connected layer can be adjusted according to the actual situation, but the number of neurons in the last fully connected layer is determined according to the preset time. For example, when the preset time is 5 days, i.e., the ammonia nitrogen content in the next 5 days needs to be predicted, the number of neurons in the last fully connected layer is 5. The initial values of the weight parameters and the bias parameters in the model can be set by methods such as Xavier initialization, He initialization and random initialization.

[0090] Further, the embodiment provides a step of training the initial long-short time sequence model according to the validation set to obtain an optimized long-short time sequence model, which includes:

[0091] According to historical breeding data, a relationship curve of excretion amount, feed remaining amount and ammonia nitrogen content is obtained;

[0092] According to the relationship curve, a prediction training set is obtained;

[0093] According to the prediction training set, an initial long short time sequence model is trained to obtain a training model;

[0094] According to the training model and the verification set, an optimized long short time sequence model is obtained.

[0095] The ammonia nitrogen content is the dependent variable, and the excretion amount and the feed remaining amount are the independent variables. According to historical breeding data, the relationship curve of the excretion amount, the feed remaining amount and the ammonia nitrogen content can be obtained by linear fitting, polynomial fitting, etc. In this embodiment, this is not limited. Then, multiple sample points are selected in the relationship curve, the number of sample points is the same as the number of samples in the historical breeding data, and the interval between each sample point is the same. The feed remaining amount and the excretion amount corresponding to the multiple sample points are used as the prediction training set.

[0096] Further, the embodiment provides a step of training an initial long short time sequence model according to a prediction training set to obtain a training model, which includes:

[0097] The prediction training set is divided into multiple training groups, and the multiple training groups are arranged in sequence to obtain a second queue sequence;

[0098] A second iteration operation is performed, which includes inputting a first training group to a current long short time sequence model for forward propagation calculation, calculating a loss function according to the forward propagation calculation result and the historical breeding data, calculating the gradient of the parameters in the current long short time sequence model according to the loss function and a back propagation algorithm, updating the current long short time sequence model according to the gradient to obtain an updated long short time sequence model, updating the second queue sequence, until a second termination condition is met, and outputting the current updated long short time sequence model as the training model; the first training group is the training group located at the first position of the second queue sequence; and the loss function is a regularization loss function.

[0099] The second termination condition is that the second queue sequence is an empty sequence, and the step of updating the second queue sequence is to remove the current first training group to obtain an updated queue sequence. In the second iteration operation, in the first iteration, the current long short time sequence model is the initial long short time sequence model, and in each subsequent iteration, the current long short time sequence model is the updated long short time sequence model obtained through the last iteration.

[0100] Specifically, first, the first training set is input to the current long short time sequence model, the data in the first training set sequentially passes through the LSTM layer and the fully connected layer for forward propagation calculation, and the forward propagation calculation result is output through the fully connected layer, that is, the ammonia nitrogen content predicted according to the feed remaining amount and the excretion amount corresponding to a plurality of sample points, then the loss function is calculated through the predicted ammonia nitrogen content and the ammonia nitrogen content corresponding to a plurality of sample points in the relationship curve, and starting from the loss function, the gradient of the loss function to the parameters (weights and biases) of the last layer (fully connected layer) of the model is calculated according to the chain rule, and the gradient is back propagated to the LSTM layer to obtain the gradient of the loss function to all parameters (including the weights and biases of the LSTM layer and the fully connected layer) of the model, then the model parameters are updated combined with the gradient descent method, and an updated long short time sequence model is obtained.

[0101] The loss function can use a regularization loss function L, and the specific formula is as follows:

[0102]

[0103] n represents the number of predicted ammonia nitrogen contents, for example, when the ammonia nitrogen contents of the next 5 days need to be predicted, the number of predicted ammonia nitrogen contents is 5, that is, n = 5, y q represents the qth predicted ammonia nitrogen content, represents the qth predicted ammonia nitrogen content, q in the relationship curve, λ represents a regularization parameter, and ∑ p θ p is a regularization term, which represents the sum of the squares of all parameters of the model, and its function is to make the parameters of the model as small as possible, avoid the model being too complex, and thus reduce the risk of overfitting, θ p represents any one parameter in the current model, and p represents an index for traversing all parameters of the model.

[0104] In this embodiment, the regularization loss function is used for calculation, and by adding a regularization term to the basic loss function, the model complexity can be effectively constrained to prevent overfitting of the model and improve the generalization ability of the model.

[0105] Further, the embodiment provides a step of obtaining an optimized long short time sequence model according to a training model and a validation set, comprising:

[0106] The parameters in the training model are divided into a plurality of parameter groups, and the plurality of parameter groups are sequentially arranged to obtain a first queue sequence;

[0107] performing a first iteration operation, the first iteration operation comprising: solidifying all parameter groups except a first parameter group, updating parameters in the current training model according to the validation set until the training model meets a first termination condition, updating the first queuing sequence and repeating the above solidifying and updating steps until the first queuing sequence is an empty sequence, and outputting the current training model as an optimized long-short time sequence model; the first parameter group being a parameter group located at a first position of the first queuing sequence.

[0108] The specific steps of updating the parameters in the training model according to the validation set comprise: dividing the validation set into multiple validation groups, inputting each validation group into the current training model in turn for forward propagation and backward propagation calculation, and updating the parameters of the current training model based on the backward propagation result, outputting the updated training model, and the first termination condition being that all validation groups have completed the above input and output steps. For the first validation group performing the input step, the current training model is a training model obtained by training the initial long-short time sequence model according to the prediction training set. For subsequent validation groups, the current training model is an updated training model obtained by the adjacent previous validation group. In the first iteration operation, for each iteration operation, the adjacent next iteration operation is performed on the basis of the updated training model obtained by the last validation group in the adjacent previous iteration operation, that is, the current training model in the adjacent next iteration operation is the updated training model obtained by the last validation group in the adjacent previous iteration operation. The step of updating the first queuing sequence comprises: removing the current first parameter group to obtain an updated first queuing sequence.

[0109] Finally, the optimized long-short time sequence model is evaluated by the test set to obtain a long-short time sequence model corresponding to each fish species. Specifically, the test set is first input into the optimized long-short time sequence model to obtain a predicted ammonia nitrogen content, and an evaluation index is calculated based on the predicted ammonia nitrogen content. The evaluation index can be selected as a mean square error or a mean absolute error. If the root mean square error or the mean absolute error on the test set is small, it indicates that the model can better adapt to time series data and can be put into use, thereby obtaining a long-short time sequence model corresponding to each fish species. If the root mean square error or the mean absolute error is small, it indicates that the model fitting effect is poor, and the design of the model needs to be adjusted and retrained.

[0110] The plurality of smooth sample points are obtained through the relationship curve, and the model training based on the plurality of smooth points can make the long-short time sequence model converge faster and obtain the training model quickly. However, there is an error between the sample points obtained through the relationship curve fitting and the real data, which leads to low accuracy of the training model. The embodiment further adjusts the parameters in the training model by using the verification set and evaluates by using the test set, and then accurately obtains the long-short time sequence model corresponding to each fish species. Compared with directly training according to the historical breeding data, the method provided in the embodiment uses relatively smooth sample points, avoids the problem of low data quality and slow convergence speed caused by the existence of abnormal values in the historical breeding data, and avoids the problem of low model accuracy by adjusting the parameters and evaluating the model through the verification set and the test set.

[0111] Further, the embodiment provides a method for determining a bacteria-adding pool, a bacteria-adding amount and a bacteria-adding type of the bacteria-adding pool according to the ammonia nitrogen content, comprising the steps of:

[0112] determining the bacteria-adding pool according to the ammonia nitrogen content and a preset threshold value;

[0113] determining the bacteria-adding type according to the type and growth stage of the fish in the bacteria-adding pool, wherein the bacteria-adding type includes nitrifying bacteria, denitrifying bacteria and anaerobic ammonia oxidation bacteria;

[0114] calculating the bacteria-adding amount according to the denitrification capacity, action time and safety factor corresponding to the bacteria-adding type.

[0115] Specifically, according to the prediction result of the long-short time sequence model, the ammonia nitrogen content of each fish pool in a preset time can be obtained. The ammonia nitrogen content is compared with the preset threshold value. If the ammonia nitrogen content in the fish pool will exceed the preset threshold value in a shorter time, the fish pool is determined as a bacteria-adding pool that needs to add denitrifying bacteria. The value of the preset threshold value and the length of the shorter time are set according to the actual situation.

[0116] Then, the bacteria-adding type is determined according to the type and growth stage of the fish in the bacteria-adding pool, wherein the bacteria-adding type includes nitrifying bacteria, denitrifying bacteria and anaerobic ammonia oxidation bacteria, etc. For example, nitrifying bacteria are suitable for fish fry, juvenile fish and adult fish stages of Cyprinidae (carp, crucian carp, etc.), Perciformes (sea bass, grouper, etc.) and Salmonidae (salmon, etc.); denitrifying bacteria are suitable for juvenile fish, adult fish and breeding stages of ornamental fish (such as goldfish, tropical ornamental fish, etc.) and some high-grade edible fish (such as sparidae, etc.); anaerobic ammonia oxidation bacteria are suitable for some low-oxygen-tolerant fish, such as loach and swamp eel.

[0117] Finally, the bacteria-adding amount of each fish pool is calculated according to the denitrification capacity, action time and safety factor corresponding to the bacteria-adding type. For example, it is assumed that nitrifying bacteria need to be added to the i-th fish pool, and the volume of the pool water in the fish pool is V.i , the predicted ammonia nitrogen content is T i , the target ammonia nitrogen content is The ammonia nitrogen removal amount t i is:

[0118]

[0119] Wherein, V i is in liters, T i and are in milligrams / liter, and t i is in grams.

[0120] According to the ammonia nitrogen removal amount t i , the bacteria inoculation amount of the ith fish pond can be obtained as

[0121]

[0122] Wherein, u represents the denitrification capacity of nitrifying bacteria, in milligrams / liter·day, u d represents the action time of nitrifying bacteria, in days, and o represents the safety factor of nitrifying bacteria, which is used to reduce the risk of not meeting the denitrification standard or fluctuation of treatment effect due to various uncertain factors in the denitrification process. Preferably, the value range of the safety factor is 1.2-1.5.

[0123] Further, the embodiment of the present application provides a step of completing quantitative bacteria inoculation of the bacteria inoculation pond according to the bacteria inoculation amount and the bacteria inoculation type, comprising:

[0124] According to the bacteria inoculation amount, the bacteria inoculation type, the fish type and the position of the bacteria inoculation pond, the comprehensive score of the bacteria inoculation pond is determined;

[0125] According to the comprehensive score, the quantitative bacteria inoculation of the bacteria inoculation pond is completed in turn.

[0126] Specifically, the scoring standard can be determined according to the bacteria inoculation amount, the bacteria inoculation type, the fish type and the position of the bacteria inoculation pond. For example, for the fish type, 3 points are given to sensitive fish, and 1 point is given to general fish; for the bacteria inoculation type, 3 points are given to the bacteria inoculation type with strong specificity, and 1 point is given to ordinary bacteria inoculation; for the bacteria inoculation amount, 3 points are given to the bacteria inoculation amount with large demand, 2 points are given to the bacteria inoculation amount with moderate demand, and 1 point is given to the bacteria inoculation amount with small demand; for the position of the bacteria inoculation pond, 3 points are given to the position close to the bacteria inoculum tank, and 1 point is given to the general position. According to the above scoring standard, the comprehensive score of each bacteria inoculation pond can be determined, and the bacteria inoculation order of each bacteria inoculation pond can be obtained by arranging the comprehensive scores in descending order.

[0127] The embodiment calculates the bacteria amount of each bacteria-adding pool accurately by the type and growth stage of the fish seed, the corresponding denitrification ability, action time and safety factor of each bacteria type, establishes a scoring standard, determines the bacteria-adding sequence through the comprehensive score of each bacteria-adding pool, and efficiently completes the bacteria-adding work.

[0128] Further, the embodiment of the present application provides a denitrifying bacteria quantitative feeding system for aquaculture, comprising:

[0129] The acquisition module is configured to acquire the excretion amount and bait remaining amount of each type of fish seed in each fish pool within a fixed time.

[0130] The calculation module is configured to predict the ammonia nitrogen content in each fish pool within a preset time according to the excretion amount, bait remaining amount and long-short time sequence model, and determine the bacteria-adding pool, bacteria-adding amount and bacteria-adding type in each fish pool according to the ammonia nitrogen content.

[0131] The control module is configured to complete the quantitative bacteria-adding to the bacteria-adding pool according to the bacteria-adding amount and bacteria-adding type.

[0132] The acquisition module, the calculation module are located in the server, and the control module is located in the control cabinet. Figures 3-5 As shown in the figure, the embodiment does not limit the setting position of the bacteria tank.

[0133] The method and system for quantitative feeding of denitrifying bacteria for aquaculture provided by the embodiment of the present application can predict the ammonia nitrogen content in each fish pool in real time through the long-short time sequence model, reasonably plan the bacteria-adding amount and bacteria-adding type in combination with the type and growth stage of the fish seed in each fish pool, effectively reduce the detection cost and improve the decision-making efficiency compared with the way of determining the ammonia nitrogen content through water quality detection.

[0134] The present application is described with reference to flowcharts and / or block diagrams of the method, device (system) and computer program product according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be realized by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device realize the functions specified in the flowcharts and / or block diagrams. Figure 1 The device for realizing the functions specified in one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0135] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the flow Figure 1 The functions of a flow or multiple flows and / or a block or multiple blocks in conjunction with the disclosed methods can be implemented on practitioners' computers in computer software, firmware, hardware, or combinations of them. Figure 1

[0136] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flow Figure 1 The functions of a flow or multiple flows and / or a block or multiple blocks in conjunction with the disclosed methods can be implemented on practitioners' computers in computer software, firmware, hardware, or combinations of them. Figure 1

[0137] The above description is only preferred embodiments of the present application, and the protection scope of the present application is not limited to the above-mentioned embodiments. Any technical solutions falling within the concept of the present application shall be considered as falling within the protection scope of the present application. It should be noted that, for those skilled in the art, some improvements and refinements can be made without departing from the principles of the present application, and these improvements and refinements shall also be considered as falling within the protection scope of the present application.​​

Claims

1. A method for quantitative inoculation of denitrifying bacteria for aquaculture, characterized by, The method comprises the following steps: obtaining the excretion amount and the bait remaining amount of each fish species in each fish tank within a fixed time; predicting the ammonia nitrogen content in each fish tank within a preset time according to the excretion amount, the bait remaining amount and a long-short time sequence model, wherein the long-short time sequence model corresponding to each fish species is trained according to historical breeding data; determining a bacteria-adding tank, a bacteria-adding amount and a bacteria-adding species of the bacteria-adding tank in each fish tank according to the ammonia nitrogen content; completing quantitative bacteria adding to the bacteria-adding tank according to the bacteria-adding amount and the bacteria-adding species; wherein, the obtaining the excretion amount and the bait remaining amount of each fish species in each fish tank within a fixed time comprises: obtaining the bait feeding amount, the feeding rate, the absorption rate of each fish species in each fish tank within the fixed time, and the cleaning frequency and the water changing frequency of each fish tank; According to the feeding amount, the feeding rate, the cleaning times and the water changing times, the bait remaining amount is obtained as: F ijx = (F ijx-1 +f ijx (1-r ij ))×a α ×b β , wherein F ijx represents the bait remaining amount of the jth type of fish in the ith fish pond on the xth day, F ijx-1 represents the bait remaining amount of the jth type of fish in the ith fish pond on the (x-1)th day, f ijx represents the feeding amount of the jth type of fish in the ith fish pond on the xth day, r ij represents the feeding rate of the jth type of fish in the ith fish pond, α represents an indication function related to the cleaning, if the ith fish pond is cleaned on the xth day, then α = 1, otherwise α = 0, a represents the proportion of the uneaten bait taken away by each cleaning to the total uneaten bait, β represents an indication function related to the water changing, if the ith fish pond is changed on the xth day, then β = 1, otherwise β = 0, and b represents the proportion of the uneaten bait taken away by each water changing to the total uneaten bait. Based on the feeding amount, absorption rate, cleaning frequency, and water change frequency, the excretion volume is calculated to be: M ijx =(M ijx-1 +(f ijx ×r ij )(1-η ij ))×c α ×d β Among them, M ijx M represents the amount of excrement of fish species of type j in the i-th fishpond on day x. ijk-1 η represents the amount of excrement of the j-th type of fish in the i-th fishpond on day x-1. ij Let represent the absorption rate of the j-th type of fish in the i-th fishpond, c represent the proportion of excrement removed during each cleaning to the total excrement, and d represent the proportion of excrement removed during each water change to the total excrement.

2. The method for quantitative inoculation of denitrifying bacteria for aquaculture according to claim 1, characterized in that, the predicting the ammonia nitrogen content in each fish tank within a preset time according to the excretion amount, the bait remaining amount and a long-short time sequence model comprises: training the long-short time sequence model corresponding to each fish species according to historical breeding data; inputting the excretion amount and the bait remaining amount of each fish species in each fish tank within the fixed time into the long-short time sequence model corresponding to each fish species to predict the ammonia nitrogen content in each fish tank within the preset time.

3. The method for quantitative inoculation of denitrifying bacteria for aquaculture according to claim 2, characterized in that, The training the long-short time sequence model corresponding to each fish species according to historical breeding data comprises: dividing the historical breeding data into a verification set and a test set; training an initial long-short time sequence model according to the verification set to obtain an optimized long-short time sequence model; evaluating the optimized long-short time sequence model according to the test set to obtain the long-short time sequence model corresponding to each fish species.

4. The method for quantitative inoculation of denitrifying bacteria for aquaculture according to claim 3, characterized in that, The training an initial long-short time sequence model according to the verification set to obtain an optimized long-short time sequence model comprises: obtaining a relationship curve between the excretion amount, the bait remaining amount and the ammonia nitrogen content according to historical breeding data; obtaining a prediction training set according to the relationship curve; training the initial long-short time sequence model according to the prediction training set to obtain a training model; obtaining the optimized long-short time sequence model according to the training model and the verification set.

5. The method for quantitative inoculation of denitrifying bacteria for aquaculture according to claim 4, characterized in that, The obtaining the optimized long-short time sequence model according to the training model and the verification set comprises: dividing the parameters in the training model into a plurality of parameter groups, and arranging the plurality of parameter groups in sequence to obtain a first queue sequence; performing a first iteration operation, wherein the first iteration operation comprises: solidifying all parameter groups except a first parameter group, updating the parameters in the current training model according to the verification set until the training model meets a first termination condition, updating the first queue sequence and repeating the solidifying and updating steps until the first queue sequence is empty, and outputting the current training model as the optimized long-short time sequence model; the first parameter group is the parameter group located at the first position of the first queue sequence.

6. The method for quantitative inoculation of denitrifying bacteria for aquaculture according to claim 4, characterized in that, The training the initial long-short time sequence model according to the prediction training set to obtain a training model comprises: dividing the prediction training set into a plurality of training groups, and arranging the plurality of training groups in sequence to obtain a second queue sequence; The second iteration operation includes inputting the first training group into the current long short time sequence model for forward propagation calculation, calculating a loss function according to a result of the forward propagation calculation and the historical breeding data, calculating a gradient of a parameter in the current long short time sequence model according to the loss function and a back propagation algorithm, updating the current long short time sequence model according to the gradient to obtain an updated long short time sequence model, updating the second queued sequence until a second termination condition is met, and outputting the current updated long short time sequence model as the training model; the first training group is a training group located at the first position of the second queued sequence; and the loss function is a regularization loss function.

7. The method according to claim 1, wherein the method is characterized by, According to the ammonia nitrogen content, a bacteria-adding pool and a bacteria-adding amount and a bacteria-adding type of the bacteria-adding pool are determined, including: According to the ammonia nitrogen content and a preset threshold value, the bacteria-adding pool is determined; According to the type and the growth stage of the fish in the bacteria-adding pool, the bacteria-adding type is determined, and the bacteria-adding type includes nitrifying bacteria, denitrifying bacteria and anaerobic ammonia oxidation bacteria; According to the denitrification capacity, the action time and the safety factor corresponding to the bacteria-adding type, the bacteria-adding amount is calculated.

8. The method for quantitative inoculation of denitrifying bacteria for aquaculture according to claim 1, characterized in that, According to the bacteria-adding amount and the bacteria-adding type, quantitative bacteria-adding to the bacteria-adding pool is completed, including: According to the bacteria-adding amount, the bacteria-adding type, the fish type and the position of the bacteria-adding pool, a comprehensive score of the bacteria-adding pool is determined; According to the comprehensive score, quantitative bacteria-adding to the bacteria-adding pool is completed in sequence.

9. A system for quantitative delivery of denitrifying bacteria for aquaculture, characterized by, Including: An acquisition module is configured to acquire excretion amounts of each type of fish and feed remaining amounts in each fish pool within a fixed time; A calculation module is configured to predict ammonia nitrogen contents in each fish pool within a preset time according to the excretion amounts, the feed remaining amounts and long short time sequence models, and determine a bacteria-adding pool and a bacteria-adding amount and a bacteria-adding type of the bacteria-adding pool according to the ammonia nitrogen contents, wherein a long short time sequence model corresponding to each type of fish is trained according to historical breeding data; A control module is configured to complete quantitative bacteria-adding to the bacteria-adding pool according to the bacteria-adding amount and the bacteria-adding type; The acquisition of the excretion amounts of each type of fish and the feed remaining amounts in each fish pool within the fixed time includes: The acquisition of the excretion amounts of each type of fish and the feed remaining amounts in each fish pool within the fixed time includes: Acquisition of a feeding amount, a feeding rate, an absorption rate of each type of fish in each fish pool and a cleaning frequency and a water changing frequency of each fish pool; According to the feeding amount, the feeding rate, the cleaning times and the water changing times, the bait residual amount is obtained as F ijx =(F ijx-1 +f ijx (1-r ij ))×a α ×b β , wherein F ijx represents the bait residual amount of the jth type of fish in the ith fish pond on the xth day, F ijx-1 represents the bait residual amount of the jth type of fish in the ith fish pond on the (x-1)th day, f ijx represents the feeding amount of the jth type of fish in the ith fish pond on the xth day, r ij represents the feeding rate of the jth type of fish in the ith fish pond, α represents an indication function related to cleaning, if the ith fish pond is cleaned on the xth day, then α=1, otherwise α=0, a represents the proportion of the total uneaten bait taken away by each cleaning, β represents an indication function related to water changing, if the ith fish pond is water changed on the xth day, then β=1, otherwise β=0, and b represents the proportion of the total uneaten bait taken away by each water changing. According to the feeding amount, the absorption rate, the cleaning frequency and the water changing frequency, the excretion amount is obtained as: M ijx = (M ijx-1 + (f ijx × r ij ) × (1-η ij )) × c α × d β , wherein M ijx represents the excretion amount of the jth type of fish in the ith fish pond on the xth day, M ijx-1 represents the excretion amount of the jth type of fish in the ith fish pond on the (x-1)th day, η ij represents the absorption rate of the jth type of fish in the ith fish pond, c represents the proportion of excretion removed by each cleaning to the total excretion amount, and d represents the proportion of excretion removed by each water changing to the total excretion amount.

Images