A method for predicting aquaculture water quality based on time difference data fusion and COS-SSA-FBPNN
By using time difference data fusion and the COS-SSA-FBPNN method, the problems of insufficient accuracy and generalization ability of water quality prediction models in aquaculture were solved, enabling accurate prediction of water quality conditions and improving the sustainability and management level of aquaculture.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HUAIYIN INSTITUTE OF TECHNOLOGY
- Filing Date
- 2026-02-06
- Publication Date
- 2026-05-26
AI Technical Summary
Existing water quality prediction models lack accuracy and generalization ability in aquaculture, failing to meet the requirements of refined management, resulting in low efficiency and low accuracy in aquatic parameter prediction.
A method for predicting aquaculture water quality based on time difference data fusion and an improved COS-SSA-FBPNN was adopted. By introducing the Cotton Tulip Algorithm (COS-SSA) with density and adaptive cosine inertia weights, the characteristic coefficient F of the time difference sequence was optimized, and feedback processing was introduced into the feedback probability neural network to establish the COS-SSA-FBPNN model for prediction.
It has improved the accuracy and generalization of water quality forecasting, enabled precise judgment of water quality conditions, reduced fish mortality, and promoted the sustainability and refined management of aquaculture.
Smart Images
Figure CN122087531A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of aquaculture water quality prediction, specifically to an aquaculture water quality prediction method based on time difference data fusion and COS-SSA-FBPNN. Background Technology
[0002] Water quality parameters are crucial for monitoring fish health. Since fish rely on water for respiration, a significant decline in water quality can severely impact their survival. Therefore, monitoring water quality parameters plays a critical role in aquaculture systems. Dissolved oxygen (DO), pH, and temperature are important water quality indicators for sustainable fish farming, affecting aquaculture yields. Currently, unpredictable changes in water quality parameters pose a serious threat to aquaculture, leading to increased stress levels in fish and ultimately hindering yield increases and mortality rates. The lack of reliable technologies for monitoring aquaculture water quality has hampered aquaculture development. Recently, driven by sensor, controller, and wireless communication technologies, intelligent systems have emerged and are being applied in various fields. While current research primarily focuses on real-time monitoring of water quality indicators, this is insufficient for effective testing in aquaculture. Machine learning and predictive models offer efficient, economical, non-destructive, and objective detection and analysis techniques, significantly contributing to the automation and intelligentization of aquaculture. Machine learning algorithms and predictive models can predict water quality parameters, simulate dynamic water quality behavior, and detect outliers, providing ample time for preventative measures. Therefore, effectively combining machine learning, predictive models, and fish husbandry requirements can help achieve intelligent aquaculture, facilitate early detection of anomalies and reduce risks, and further enhance the sustainability of aquaculture. Currently, scholars have developed various models to predict aquatic environmental parameters, but most of these models are complex and suffer from two key drawbacks: poor accuracy and generalization ability. This leads to low efficiency and low precision in predicting aquatic parameters, thus failing to fully meet the requirements of refined management in aquaculture. Summary of the Invention
[0003] Purpose of the invention: To address the problems in the background technology, this invention discloses an aquaculture water quality prediction method based on time difference data fusion and COS-SSA-FBPNN, which more effectively captures dynamic information of water quality parameters, captures the complex correlations between water quality parameters, enhances prediction accuracy, and thus obtains higher prediction results for aquaculture water quality, showing good performance in predicting aquaculture water quality conditions.
[0004] Technical Solution: This invention discloses a method for predicting aquaculture water quality based on time difference data fusion and COS-SSA-FBPNN, comprising the following steps:
[0005] Step 1: Collect aquatic environmental parameter data, including temperature, pH value, and dissolved oxygen, perform normalization processing, and establish a dataset;
[0006] Step 2: Calculate the time difference sequence characteristic coefficient F of the environmental parameters, wherein the time difference sequence characteristic coefficient F is the weighted value of the time difference sequence data of temperature, pH value and dissolved oxygen parameters;
[0007] Step 3: Improve the SSA algorithm by introducing density-based leader position update and designing an adaptive adjustment weight based on the cosine function. Introduce a follower position update strategy based on adaptive cosine function inertial weight adjustment to obtain the COS-SSA algorithm based on density and adaptive cosine inertial weight.
[0008] Step 4: By introducing feedback processing into the probabilistic neural network and utilizing COS-SSA to smooth the parameters of the feedback probabilistic neural network. Optimization was performed to establish an aquaculture water quality prediction model based on COS-SSA-FBPNN;
[0009] Step 5: Optimize the three time weights of temperature, pH, and dissolved oxygen parameters in the time difference sequence characteristic coefficient F using COS-SSA. Then, use the optimized F and other aquaculture water quality parameters as input, and use the COS-SSA-FBPNN aquaculture water quality prediction model to make predictions and complete the judgment of water quality status.
[0010] Furthermore, in step 1, multiple sets of aquaculture water quality parameters, including temperature, pH value, and dissolved oxygen, are collected and normalized. According to the water quality, the original dataset is divided into five levels: poor, bad, average, good, and relatively good. The data of each level are then divided into training set, validation set, and test set in a ratio of 5:2:3.
[0011] Furthermore, the time difference sequence and characteristic coefficient F of the environmental parameters calculated in step 2 are specifically as follows:
[0012] ;
[0013] in, , , The time weights of the three sampling parameters, , , These are time-difference series data for temperature, pH value, and dissolved oxygen parameters, respectively.
[0014] Furthermore, step 3 introduces density-based leader position updates, as follows:
[0015] The leader's Dimensional position The update is performed according to the following formula:
[0016] ;
[0017] in, It is the first The target position value of dimension, the first The upper bound of the dimensional search range is , No. The lower bound of the dimensional search range is , , These are control parameters. Determines the direction and step size for leader position updates, taking a random number between [0, 1]. The convergence factor, used to balance the convergence speed of the algorithm during the iterative process, is obtained by the following formula:
[0018] ;
[0019] In the formula, Indicates the current iteration number. Indicates the maximum number of iterations. Leader density is defined as the ratio between the number of leaders and the number of followers in each iteration. The definition is as follows:
[0020] ;
[0021] in, It is the number of leaders in each iteration cycle. The number of followers in each iteration cycle.
[0022] Furthermore, the follower position update based on the adaptive cosine function inertia weight adjustment strategy is as follows:
[0023] The follower's position is updated based on the leader's position, using the following formula:
[0024]
[0025] in, Indicates the first A follower of the sea sheath Position in dimensions Indicates the first A follower of the sea sheath Position in dimensions This represents the current iteration number. For inertial weights, ;
[0026] The specific adaptive adjustment strategy involves introducing a cosine function and an iterative threshold. , This is used to ensure adaptive adjustment of the inertia weight. The definition is as follows:
[0027]
[0028] in, The minimum inertia weight up to the current iteration. The maximum inertia weight up to the current iteration. This represents the maximum number of iterations.
[0029] Furthermore, step 4 is specifically as follows:
[0030] Step 4.1: Introduce feedback processing, feed the output back to the mode layer, and solve for the output of the mode layer;
[0031] In the pattern layer, neurons are divided into different categories, and the first neuron is calculated using the following formula. The first mode Neuron-like output :
[0032] ;
[0033] in, It is the input data. It is output feedback. It is the kernel center. To output the classification pattern center, It is a smoothing parameter used to determine the kernel domain size;
[0034] Step 4.2: Solve the feedback probability function using the summation layer:
[0035] probability function Through the The summation is obtained as follows:
[0036] ;
[0037] in, It is the total number of patterns. Total number of categories;
[0038] Step 4.3: The output layer calculates the specific category, determines the maximum output of the summation layer, and uses this to determine the final type;
[0039] Step 4.4: Use COS-SSA to adjust the smoothing parameters Optimization is performed to obtain a more suitable kernel size.
[0040] Furthermore, in step 5, the time difference sequence characteristic coefficient F is optimized using COS-SSA, specifically by the three time weights in the time difference sequence characteristic coefficient F: , , Then, the COS-SSA-FBPNN aquaculture water quality prediction model is used to integrate the optimized F with other aquaculture water quality parameters, namely temperature, pH value, and dissolved oxygen, until the termination condition is met, thereby completing the judgment of water quality status.
[0041] Beneficial effects:
[0042] 1. This invention proposes an improved Cosine-based Sea-Upper Algorithm (COS-SSA) based on density and adaptive cosine inertia weights. First, in the leader position update, this patent introduces a density-based leader position update. By defining the ratio between the number of leaders and followers in each iteration, it effectively controls the number of sea-ultra leading the global and local optima, thus achieving a better balance between the global and local optima represented by both, resulting in better optimization results. Furthermore, to avoid the problem that global optimization and convergence are often difficult to satisfy simultaneously, an adaptive adjustment weight based on a cosine function is designed in the leader-follower position update. This method completes the follower position update based on the adaptive cosine function inertia weight adjustment strategy, thereby ensuring that the method has strong optimization capabilities, which can improve both convergence speed and optimization accuracy, and avoid the algorithm getting trapped in local extrema.
[0043] 2. This invention proposes a prediction model based on time-difference data fusion and COS-SSA-FBPNN. First, feedback processing is introduced into the probabilistic neural network, feeding the output back to the pattern layer, thereby obtaining richer pattern forms and facilitating more accurate classification results. Furthermore, COS-SSA is used to smooth the parameters of the feedback probabilistic neural network. By performing optimization, a more suitable kernel size is obtained, thereby realizing a feedback probability neural network (COS-SSA-FBPNN) model based on COS-SSA. This model can automatically adjust the network structure and parameters, and can obtain a more flexible network structure, so that it gradually approaches the real data distribution and performs well in dealing with complex problems.
[0044] 3. This invention utilizes COS-SSA to optimize the characteristic coefficient F of the time difference sequence, thereby optimizing... The three time weights in the coefficients are then used to fuse the optimized F with other aquaculture water quality parameters (temperature, pH, dissolved oxygen) using COS-SSA-FBPNN, establishing an aquaculture water quality prediction model based on COS-SSA-FBPNN to determine the water quality status. Experimental results show that, for the validation set, the prediction accuracy of the proposed method for five different water quality levels (poor, bad, average, good, and relatively good) is 91.4%, 89.2%, 92.3%, 91%, and 89.2%, respectively. This demonstrates that the proposed method based on time-difference data fusion and COS-SSA-FBPNN performs well in predicting aquaculture water quality, reasonably predicting the water environment status, providing more accurate predictions for aquaculture, and laying the foundation for improving aquaculture farm management and early detection of water quality parameters, further promoting the green and sustainable development of aquaculture. Attached Figure Description
[0045] Figure 1 This is a flowchart illustrating the implementation of an aquaculture water quality prediction model based on time difference data fusion and COS-SSA-FBPNN.
[0046] Figure 2 It is an optimization process based on density and adaptive cosine inertia weighting of the Cosmos-SSA algorithm;
[0047] Figure 3 It is a feedback probability neural network architecture;
[0048] Figure 4 It is the COS-SSA-FBPNN neural network architecture. Detailed Implementation
[0049] To better explain and facilitate understanding of the present invention, the technical solution of the present invention is described in detail below. The following embodiments are illustrative of the present invention, but the present invention is not limited to the following embodiments.
[0050] This invention proposes a method for predicting aquaculture water quality based on time-difference data fusion and COS-SSA-FBPNN. First, an improved Cossula salina algorithm (COS-SSA) based on density and adaptive cosine inertia weights is proposed. Second, feedback processing is introduced into the mode layer of the probabilistic neural network, and COS-SSA is used to smooth the parameters of the feedback probabilistic neural network. Optimization is performed to implement a feedback probability neural network (COS-SSA-FBPNN) model based on COS-SSA. Finally, COS-SSA is used to optimize the characteristic coefficients of the time difference sequence. Then, COS-SSA-FBPNN is used to optimize the... By integrating predictions with other aquaculture water quality parameters (temperature, pH, dissolved oxygen), an aquaculture water quality prediction model based on COS-SSA-FBPNN is established to determine the water quality status.
[0051] The implementation block diagram of the proposed fusion method is as follows: Figure 1 As shown, the specific steps are as follows:
[0052] Step (1): Collect aquaculture water quality parameters, normalize them, and establish a dataset.
[0053] The min-max method is used to normalize the data. The purpose of normalization is to obtain a standardized dataset, ensuring that the data maintains a consistent range of values while preserving the relative differences between data values.
[0054] Assume the collected water environment data is The normalization formula is as follows:
[0055] (1)
[0056] in, , For the number of data points, Represents water quality parameters, including temperature pH value Dissolved oxygen ( ), The normalized detection values (including temperature) pH value Dissolved oxygen ), , This is a function to find the minimum and maximum values.
[0057] Step (2): Calculate the time difference sequence and characteristic coefficients of environmental parameters.
[0058] Step 2.1: Calculate the time difference series of environmental parameters
[0059] Let the time difference series data of temperature, pH value, and dissolved oxygen parameters be respectively , , The specific definition of the time difference sequence is as follows:
[0060]
[0061]
[0062] (2)
[0063] In the formula, , For being in time , Temperature parameters, , For being in time , pH value parameter, , For being in time , Dissolved oxygen parameters.
[0064] Step 2.2: Calculate the characteristic coefficients of the time difference series
[0065] To quantify the changes in water quality parameter characteristics over a specific time interval and record the influence of selected water quality parameters on the fusion process over time, this invention calculates characteristic coefficients based on the time difference sequence of water quality parameters. This effectively captures the dynamic data values associated with them, highlights differences between multiple features, and enhances the robustness of the measurements. The calculation process is as shown in equation (3):
[0066] (3)
[0067] in, , , The time weights of the three sampling parameters.
[0068] Step (3) involves introducing density-based leader position updates and designing adaptive weight adjustments based on a cosine function. We introduce a follower position update strategy based on adaptive cosine function inertia weight adjustment, resulting in an improved COS-SSA algorithm based on density and adaptive cosine inertia weight, as follows:
[0069] The Salamander Search (SSA) algorithm iteratively searches and evolves through the interaction of the salamander population. During optimization, individual salamanders continuously move, updating the group's position until it gathers at the location with the richest food resources, thus achieving the global optimum. The leader salamander changes its position only based on the location of food resources, while followers continuously move slowly. This process allows the group to continuously explore and develop optimal food resources. The interaction between the leader and followers serves as an adaptive mechanism in SSA, effectively balancing exploration and development, and obtaining an accurate estimate of the optimal solution during optimization. In each iteration, the group retains the best solution achieved so far, as well as the search trend and direction.
[0070] The SSA algorithm is simple to implement and has few control parameters. However, the main problem with traditional SSA algorithms is insufficient diversity, difficulty in balancing global and local optima, and frequent incompatibility between global optimization and convergence. Therefore, this invention proposes a density-based and adaptive cosine inertia weighted algorithm (COS-SSA). The main innovations are in two aspects: First, in leader position updating, this patent introduces density-based leader position updating. By defining the ratio between the number of leaders and followers in each iteration, it effectively controls the number of tunicates leading to both global and local optima, thus achieving a better balance between the two and resulting in better optimization. Second, to avoid the problem of frequent incompatibility between global optimization and convergence, in leader-follower position updating, a follower position updating strategy based on adaptive cosine function inertia weight adjustment is implemented, designing an adaptive adjustment weight based on the cosine function. This ensures that the method has strong optimization capabilities, which can improve both convergence speed and optimization accuracy, and avoid the algorithm getting trapped in local optima.
[0071] The specific steps are as follows:
[0072] Step 3.1: Initialize the salps population
[0073] Define the target environment 3D space, in which, Indicates the population size. The dimension representing the maximum space, the position of each salver is defined as... The target location of the food source is defined as The search scope for each dimension includes the upper bound. and the lower realm Finally, the initial population position is obtained according to formula (4):
[0074] (4)
[0075] Step 3.2: Complete density-based leader position update
[0076] In population optimization, the leader's position update needs to be related to the target position while also possessing a certain degree of randomness. This ensures the leader plays a leading role in the overall environmental search, thereby guiding followers towards the target. Therefore, the leader's first... Dimensional position The update is performed according to formula (5):
[0077] (5)
[0078] in, It is the first The target position value of dimension, the first The upper bound of the dimensional search range is , No. The lower bound of the dimensional search range is , , These are control parameters. Determines the direction and step size for leader position updates, taking a random number between [0, 1]. The convergence factor is used to balance the convergence speed of the algorithm during the iterative process, and is obtained from equation (6):
[0079] (6)
[0080] In equation (6), Indicates the current iteration number. Indicates the maximum number of iterations. Leader density is defined as the ratio between the number of leaders and the number of followers in each iteration. Specifically, The definition is as follows:
[0081] (7)
[0082] in, It is the number of leaders in each iteration cycle. The number of followers in each iteration cycle.
[0083] Step 3.3: Complete the follower position update based on adaptive cosine inertia weights.
[0084] The follower's position is updated based on the leader's position, using the following formula:
[0085] (8)
[0086] Among them, the The first follower Dimensional position is , For iteration time, It is the acceleration of an individual's motion. The definition is as follows:
[0087] (9)
[0088] in, This represents the initial velocity, the initial velocity of the follower at the start of each iteration. All are 0.
[0089] Due to follower position The update only matches the previous location of the sea squirt. Related, therefore, speed It is expressed as follows:
[0090] (10)
[0091] Therefore, the traditional formula for updating the position of tunicates is... Represented as:
[0092] (11)
[0093] In the formula, Indicates the first A follower of the sea sheath Position in dimensions Indicates the first A follower of the sea sheath Position in dimensions .
[0094] In the above algorithms, global optimization and convergence are often difficult to satisfy simultaneously. To ensure a balance between convergence and global optimization in the *Squirrel Algorithm*, this patent proposes a follower position update method based on adaptive cosine inertia weights, specifically considering the update characteristics of the follower. This involves proposing an adaptive inertia weight strategy and introducing the concept of inertia weights. Add it to the follower position update formula (11) to obtain formula (12):
[0095] (12)
[0096] in, Indicates the first A follower of the sea sheath Position in dimensions Indicates the first A follower of the sea sheath Position in dimensions This represents the current iteration number.
[0097] In formula (12), the design inertia weight is... Follow the number of iterations The value decreases adaptively with changes in the global optimum, thus ensuring that the global optimum better influences the updates of followers. The specific adaptive adjustment strategy is as follows: with the number of iterations... Changes in inertia weight Adaptively adjusted, during initial optimization, inertia weights When the inertia weight is relatively large, the algorithm can find the global optimum relatively quickly. However, as the number of iterations gradually increases, the inertia weight... It exhibits a non-linear descent state, with the optimization step size of each individual decreasing. Based on the above adaptive adjustment strategy, for... To determine the value of , this method introduces a cosine function and an iterative threshold. , This is used to ensure adaptive adjustment of the inertia weight. The definition is as follows:
[0098] (13)
[0099] in, The minimum inertia weight up to the current iteration. This represents the maximum inertia weight up to the current iteration.
[0100] Through the above process, the Cossula salina-SSA algorithm based on density and adaptive cosine inertia weights can be implemented. Experiments show that this method has strong optimization capabilities, can avoid getting trapped in local optima, and improves convergence speed and optimization accuracy. The specific execution process is as follows: Figure 2 As shown below:
[0101] Step 1: Set initial parameters, including the maximum number of iterations. Population size Dimensions Initialize the population To determine the initial position coordinates of individuals in the population.
[0102] Step 2: Calculate the individual fitness value of each tunicate based on the objective function.
[0103] Step 3: Select the initial location of the food source. Sort the fitness values of the salps and set the location coordinates of the salps with the best fitness as the target location, which is the location of the food source.
[0104] Step 4: Identify leaders and followers.
[0105] Step 5: Update the position of the leader of the tunicate according to formula (5).
[0106] Step 6: Update the position of the salver follower according to formula (12).
[0107] Step 7: Calculate the fitness value of each tunicate and compare it with the fitness value of the current target location to update the global optimal tunicate target location, which is to update the location of the food source.
[0108] Step 8: Determine if the iteration conditions are met. If yes, output the result; otherwise, go to Step 4 to continue iterative evolution, and finally output the current optimal fitness and target position coordinates.
[0109] Step (4) establishes an aquaculture water quality prediction model based on COS-SSA-FBPNN, as detailed below:
[0110] First, a feedback probability neural network based on COS-SSA (COS-SSA-FBPNN) is established, and then the COS-SSA-FBPNN classifier is used for the prediction of aquaculture water quality.
[0111] Traditional deep learning methods often suffer from overfitting, meaning that when training data is insufficient or overtrained, even if the model performs well on the training set, it performs poorly on the test set, thus affecting the model's generalization ability. Therefore, this invention proposes a COS-SSA-FBPNN neural network. The innovation of this method is mainly reflected in the pattern layer, achieving two aspects of innovation and improvement: First, it introduces feedback processing, feeding the output back to the pattern layer to obtain richer pattern forms, facilitating more accurate classification results; second, it utilizes COS-SSA to smooth parameters. The method optimizes the network to obtain a more suitable kernel size. By automatically adjusting the network structure and parameters, it achieves a more flexible network structure, which gradually approximates the real data distribution. This makes the model perform well in handling complex problems. Therefore, compared with the existing technology, the model proposed in this invention has a significantly improved accuracy in predicting aquaculture water quality.
[0112] Feedback probability neural network architecture such as Figure 3 As shown, the COS-SSA-FBPNN neural network architecture is as follows: Figure 4 As shown, the specific steps are as follows:
[0113] Step 4.1: Introduce feedback processing to feed the output back to the mode layer and solve for the output of the mode layer.
[0114] The model consists of an input layer, a pattern layer, a summation layer, and an output layer. The input layer receives the input signal but does not perform calculations; these are handled by the summation layer and the pattern layer. The number of neurons in the output layer is the same as the target category. Specifically, in the pattern layer, neurons are divided into different categories, and the first neuron is calculated using equation (14). The first mode Neuron-like output .
[0115] (14)
[0116] in, It is the input data. It is output feedback. It is the kernel center. To output the classification pattern center, It is a smoothing parameter used to determine the kernel domain size.
[0117] Step 4.2: Solve the feedback probability function using the summation layer.
[0118] probability function Through the The summation is obtained as follows:
[0119] (15)
[0120] in, It is the total number of patterns. This represents the total number of categories.
[0121] Step 4.3: Calculate the specific category from the output layer.
[0122] Finally, the maximum output of the summation layer is determined to determine the final type, as shown in the following formula:
[0123] (16)
[0124] Step 4.4: Use COS-SSA to adjust the smoothing parameters Optimization was performed to obtain a more suitable kernel size, and a COS-SSA-FBPNN neural network was established.
[0125] The specific optimization steps are detailed in the COS-SSA optimization process described above (Step 1-Step 8). The main parameter settings during optimization are as follows: maximum number of iterations. Population size Dimensions Initialize the population to In this context, rows represent individual optimizations, and columns represent smoothing parameters. The initial values of individuals in the population are random numbers between 0 and 1. The input signal of the FBPNN neural network only takes the raw data of aquaculture water quality parameters (temperature). pH value Dissolved oxygen At this point, the time difference series data ( , , If not used as an input signal, the output consists of five water quality classification data, expressed as mean square error. As a fitness function, it is used to evaluate each smoothing parameter. The smaller the fitness value, the better the smoothing parameter performs. The better the predictive performance, The definition is as follows:
[0126] (17)
[0127] in, To output the number of items, This is the actual output value. This is the actual output value.
[0128] The selected water quality grade data were uniformly distributed, as detailed below:
[0129] Table 1. Distribution of water quality grade sample data
[0130] Water quality categories Sample size / group percentage Poor 1800 20% Difference 1800 20% generally 1800 20% good 1800 20% better 1800 20%
[0131] Step (5) optimizes the time difference sequence characteristic coefficient (F) using COS-SSA. Specifically, it optimizes the three time weights in the F coefficient. , , Then, COS-SSA-FBPNN is used to fuse the optimized F with other aquaculture water quality parameters (temperature, pH, dissolved oxygen) for prediction until the termination condition is met (greater than the maximum number of iterations or the prediction accuracy is reached), thereby completing the judgment of water quality status.
[0132] The specific optimization steps are detailed in the COS-SSA optimization process described above (Step 1-Step 8). The main parameter settings during optimization are as follows: maximum number of iterations. Population size Dimensions Initialize the population In this context, rows represent individual optimizations, and columns represent... The three time weights in the coefficient ( , , The initial values of individuals in the population are random numbers between 0 and 1. During this optimization period, the input signal of the neural network takes all data, including the raw data of aquaculture water quality parameters (temperature). pH value Dissolved oxygen ), and its corresponding time difference series data ( , , The output still consists of five water quality classifications, and the smoothing parameter is also adjusted. The fitness function is defined using the optimal value obtained after the previous optimization stage. The specific definition is the same as before. During the optimization period, the COS-SSA algorithm is based on... As a result, the population location is continuously updated and optimized. The three time weights in the coefficient ( , , The parameter combination, through multiple iterations, gradually approaches the optimal time weight. , , This improves the model's predictive performance.
[0133] An experiment was conducted to predict aquaculture water quality based on time-difference data fusion and COS-SSA-FBPNN, and the proposed method was compared with conventional methods. Details are as follows:
[0134] A dataset was obtained from an experimental fishpond in Huai'an, and the performance of the proposed model was evaluated using this data. An experimental validation of aquaculture water quality prediction based on time-difference data fusion and COS-SSA-FBPNN was completed. To evaluate the model's predictive performance, the original dataset was divided into five levels according to water quality: poor, bad, average, good, and relatively good. The data from each level were then divided into training, validation, and test sets in a 5:2:3 ratio. Furthermore, to examine the effectiveness and comparability of the proposed model, water quality prediction models were established using RBF networks, Support Vector Machines (SVM), Backpropagation (BP) networks, and the proposed time-difference data fusion and COS-SSA-FBPNN fusion method. These models predicted the aquaculture water quality environment, and the prediction effects of each model were compared. The experimental results are shown in Table 2.
[0135] Table 2 Comparison of prediction performance of different algorithms
[0136] Table 1 shows that for the validation set, the prediction accuracy of the method proposed in this patent for the five different water quality levels (poor, bad, average, good, and relatively good) is 91.4%, 89.2%, 92.3%, 91%, and 89.2%, respectively. For the test set, the prediction accuracy for the five different water quality levels is 86.5%, 81.3%, 85.1%, 86.1%, and 84.2%, respectively. The table also shows that this method exhibits better prediction performance compared to other conventional models. Therefore, the time-difference data fusion and COS-SSA-FBPNN method proposed in this patent demonstrates excellent performance in predicting aquaculture water quality. It can reasonably predict the state of the aquatic environment, providing more accurate predictions for aquaculture, thus laying the foundation for improving aquaculture farm management processes and early detection of water quality parameters, and further promoting the green and sustainable development of aquaculture.
[0137] The above embodiments are only for illustrating the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. They should not be construed as limiting the scope of protection of the present invention. All equivalent transformations or modifications made in accordance with the spirit and essence of the present invention should be covered within the scope of protection of the present invention.
Claims
1. A method for predicting aquaculture water quality based on time-difference data fusion and COS-SSA-FBPNN, characterized in that, Includes the following steps: Step 1: Collect aquatic environmental parameter data, including temperature, pH value, and dissolved oxygen, perform normalization processing, and establish a dataset; Step 2: Calculate the time difference sequence characteristic coefficient F of the environmental parameters, wherein the time difference sequence characteristic coefficient F is the weighted value of the time difference sequence data of temperature, pH value and dissolved oxygen parameters; Step 3: Improve the SSA algorithm by introducing density-based leader position update and designing an adaptive adjustment weight based on the cosine function. Introduce a follower position update strategy based on adaptive cosine function inertial weight adjustment to obtain the COS-SSA algorithm based on density and adaptive cosine inertial weight. Step 4: By introducing feedback processing into the probabilistic neural network and utilizing COS-SSA to smooth the parameters of the feedback probabilistic neural network. Optimization was performed to establish an aquaculture water quality prediction model based on COS-SSA-FBPNN; Step 5: Optimize the three time weights of temperature, pH, and dissolved oxygen parameters in the time difference sequence characteristic coefficient F using COS-SSA. Then, use the optimized F and other aquaculture water quality parameters as input, and use the COS-SSA-FBPNN aquaculture water quality prediction model to make predictions and complete the judgment of water quality status.
2. The aquaculture water quality prediction method based on time difference data fusion and COS-SSA-FBPNN as described in claim 1, characterized in that, In step 1, multiple sets of aquaculture water quality parameters, including temperature, pH value, and dissolved oxygen, are collected and normalized. According to the water quality, the original dataset is divided into five levels: poor, bad, average, good, and relatively good. The data of each level are then divided into training set, validation set, and test set in a ratio of 5:2:
3.
3. The aquaculture water quality prediction method based on time difference data fusion and COS-SSA-FBPNN as described in claim 1, characterized in that, The time difference sequence and characteristic coefficient F of the environmental parameters calculated in step 2 are as follows: ； in, , , The time weights of the three sampling parameters, , , These are time-difference series data for temperature, pH value, and dissolved oxygen parameters, respectively.
4. The aquaculture water quality prediction method based on time difference data fusion and COS-SSA-FBPNN as described in claim 1, characterized in that, Step 3 introduces density-based leader position updates, as follows: The leader's Dimensional position The update is performed according to the following formula: ； in, It is the first The target position value of dimension, the first The upper bound of the dimensional search range is , No. The lower bound of the dimensional search range is , , These are control parameters. Determines the direction and step size for leader position updates, taking a random number between [0, 1]. The convergence factor, used to balance the convergence speed of the algorithm during the iterative process, is obtained by the following formula: ； In the formula, Indicates the current iteration number. Indicates the maximum number of iterations. Leader density is defined as the ratio between the number of leaders and the number of followers in each iteration. The definition is as follows: ； in, It is the number of leaders in each iteration cycle. The number of followers in each iteration cycle.
5. The aquaculture water quality prediction method based on time difference data fusion and COS-SSA-FBPNN according to claim 1, characterized in that, The follower position update based on the adaptive cosine function inertia weight adjustment strategy is as follows: The follower's position is updated based on the leader's position, using the following formula: in, Indicates the first A follower of the sea sheath Position in dimensions Indicates the first A follower of the sea sheath Position in dimensions This represents the current iteration number. For inertial weights, ; The specific adaptive adjustment strategy involves introducing a cosine function and an iterative threshold. , This is used to ensure adaptive adjustment of the inertia weight. The definition is as follows: in, The minimum inertia weight up to the current iteration. The maximum inertia weight up to the current iteration. This represents the maximum number of iterations.
6. The aquaculture water quality prediction method based on time difference data fusion and COS-SSA-FBPNN as described in claim 1, characterized in that, Step 4 is as follows: Step 4.1: Introduce feedback processing, feed the output back to the mode layer, and solve for the output of the mode layer; In the pattern layer, neurons are divided into different categories, and the first neuron is calculated using the following formula. The first mode Neuron-like output : ； in, It is the input data. It is output feedback. It is the kernel center. To output the classification pattern center, It is a smoothing parameter used to determine the kernel domain size; Step 4.2: Solve the feedback probability function using the summation layer: probability function Through the The summation is obtained as follows: ； in, It is the total number of patterns. Total number of categories; Step 4.3: The output layer calculates the specific category, determines the maximum output of the summation layer, and uses this to determine the final type; Step 4.4: Use COS-SSA to adjust the smoothing parameters Optimization is performed to obtain a more suitable kernel size.
7. The aquaculture water quality prediction method based on time difference data fusion and COS-SSA-FBPNN according to claim 3, characterized in that, In step 5, the COS-SSA optimization of the time difference sequence characteristic coefficient F specifically involves the three time weights in the time difference sequence characteristic coefficient F: , , Then, the COS-SSA-FBPNN aquaculture water quality prediction model is used to integrate the optimized F with other aquaculture water quality parameters, namely temperature, pH value, and dissolved oxygen, until the termination condition is met, thereby completing the judgment of water quality status.