Fish spawning behavior intelligent prediction method

Abstract

The invention discloses a fish spawning behavior intelligent prediction method. The intelligent prediction method comprises the following steps: collecting and sorting data information of a fish breeding period; constructing a zero-expansion two-part model according to the data information; according to the zero-expansion two-part model, determining a time period T when the egg laying occurrence proportion is highest and the peak occurrence probability is highest; constructing a fish spawning prediction chaotic system model based on attractor reconstruction; and according to the time period T, using the reconstructed fish spawning prediction chaotic system model to predict the spawning scale of the test data set V under the environmental condition. And statistical inference and state space reconstruction are carried out, so that the future spawning dynamic state of the fishes can be predicted.

Description

Technical Field

[0001] This invention relates to the field of aquatic ecology, and in particular to an intelligent prediction method for fish spawning behavior. Background Technology

[0002] Fish are indicator species in aquatic ecosystems, and their community structure and population size play a crucial role in maintaining the balance of these ecosystems. Natural fish reproduction is an organically organized life phenomenon closely related to environmental factors. In the natural environment, fish reproduction depends on the regulation of environmental stimuli and is jointly regulated by abiotic factors such as water flow, water quality, water temperature, and spawning ground substrate. A driving force-state-response interaction exists between hydrological processes, the river environment, and fish spawning behavior.

[0003] River aquatic ecosystems constitute a highly coupled and complex nonlinear system, with mutual feedback among various environmental factors. A significant characteristic of nonlinear complex systems is their non-additive and emergent nature; that is, the overall phenomenon is not a simple accumulation of internal attributes. Conversely, when the lower-level components of the system form a whole in a certain way, new properties emerge. Therefore, analyzing the influence of individual environmental factors to understand the phenomena of nonlinear systems easily yields "illusory" correlations. Neither the linear assumptions of correlation analysis nor the variable independence assumptions of regression analysis can truly describe the causal relationships between coupled variables.

[0004] Actual field observations cannot monitor all relevant variables. The monitoring data only represents the most common observable environmental factor sequences. Some important factors affecting fish spawning behavior may be missing. In this case, using mathematical equations with multiple parameters to fit the nonlinear relationship between key environmental factors and biological behavior, such as data mining and machine learning methods, shows great limitations. Summary of the Invention

[0005] In view of the above problems, the present invention is proposed to provide an intelligent prediction method for fish spawning behavior that overcomes or at least partially solves the above problems.

[0006] According to one aspect of the present invention, an intelligent prediction method for fish spawning behavior is provided, the intelligent prediction method comprising:

[0007] Collect data on fish breeding seasons, screen key environmental factors that affect fish spawning, and compile them into a dataset;

[0008] Construct a zero-inflation two-part model based on the dataset;

[0009] Based on the zero-inflation two-part model, the time period T with the highest spawning rate and the highest probability of peak occurrence is determined;

[0010] Construct a chaotic system model for predicting fish spawning based on attractor reconstruction;

[0011] Based on the time period T, a reconstructed chaotic system model for predicting fish spawning is used to predict the spawning scale under environmental conditions in the test dataset V.

[0012] Optionally, the collection of data on fish breeding season, screening of key environmental factors affecting fish spawning, and compilation into a dataset specifically includes:

[0013] Collect and organize data such as daily spawning volume, daily flow rate, and daily water temperature during the fish breeding season;

[0014] Based on the ecological habits that affect fish reproduction and the characteristics of environmental changes during the breeding season, environmental factors that affect fish spawning and reproduction were initially selected.

[0015] The environmental factors include the current daily average flow rate, the previous day's flow rate, the previous two days' flow rate, the 1-day flow rate variation, the 2-day flow rate variation, the current daily average water temperature, the 1-day water temperature variation, the 2-day water temperature variation, the previous day's flow rate, and the previous two days' flow rate.

[0016] Independence analysis was performed on the environmental factors to eliminate factors with collinearity.

[0017] Converting daily fish spawning data into a percentage of total annual spawning eliminates inter-year spawning variations and aligns with the characteristics of semi-continuous data.

[0018] The original observation sequence was z-score normalized.

[0019] For first- and second-order difference sequences, segmentation and standardization are performed based on the positive and negative values ​​of the variables;

[0020] The receiver operating curve method was used to identify a combination of key environmental factors affecting fish reproduction.

[0021] Data on daily egg production and key environmental factors were compiled to form dataset F.

[0022] Optionally, constructing a zero-inflation two-part model based on the data specifically includes:

[0023] The dataset is divided into a training dataset S and a test dataset V;

[0024] Based on whether the daily egg production is zero, the training dataset S is divided into zero-value data and non-zero-value data to construct a zero-inflation two-part model.

[0025] Optionally, determining the time period T with the highest spawning rate and the highest peak probability based on the zero-inflation two-part model specifically includes:

[0026] Using the zero-inflation two-part model, the probability of fish spawning response under environmental conditions of the test dataset V is estimated, and the time period T with the highest spawning rate and the highest peak probability is determined.

[0027] The posterior distribution of semi-continuous oviposition data in the training dataset S is described by a two-part model based on zero inflation: the first part describes the distribution of zero-value data, and the second part describes the distribution of non-zero-value oviposition data driven by environmental factors.

[0028] The zero-valued part uses a logistic regression model, and the non-zero-valued part uses a beta regression model. A generalized linear model is used to evaluate the probability p. i and mean μ i Perform regression analysis separately;

[0029] The parameters of the two parts of the model will be estimated separately using the Monte Carlo Markov chain algorithm;

[0030] Based on the environmental factors of the test dataset V, a zero-inflation beta regression model with converged parameters was used to obtain the response probability and peak proportion of fish spawning in the validation set, and the time period T with the highest spawning proportion and the highest peak probability was selected.

[0031] Optionally, the intelligent prediction method further includes: comparing the predicted egg-laying scale with the measured egg-laying data to evaluate the model performance.

[0032] Optionally, the construction of the fish spawning prediction chaotic system model based on attractor reconstruction specifically includes:

[0033] Based on the training dataset S, the embedding dimension E of the state space composed of environmental factors and spawning data is determined by applying the correlation dimension method.

[0034] From a set of n environmental factor variables, an E-dimensional combination of environmental factors is generated. Based on the multi-view embedding method, this combination is reconstructed in the training dataset S. Each attractor represents a neighboring trajectory in the state space of fish spawning driven by environmental factors, and the attractors are sorted according to the magnitude of the predicted Pearson correlation coefficient within the sample.

[0035] Optionally, the predicted spawning scale of the chaotic system model under the environmental conditions of the test dataset V specifically includes:

[0036] Select the top rankings The attractor is used for out-of-sample prediction of the test dataset V, employing the simplex projection method to predict... For the spawning behavior of a group of fish, the mean of all attractor predictions is taken as the final prediction result.

[0037] This invention provides an intelligent prediction method for fish spawning behavior. The intelligent prediction method includes: collecting and organizing data on fish breeding seasons; screening key environmental factors affecting fish spawning; constructing a zero-inflated two-part model based on the data; determining the time period T with the highest spawning rate and the highest peak probability based on the zero-inflated two-part model; constructing a fish spawning prediction chaotic system model based on attractor reconstruction; and applying the reconstructed fish spawning prediction chaotic system model to predict the spawning scale under environmental conditions of a test dataset V based on the time period T. By performing statistical inference and state space reconstruction, the future spawning dynamics of fish can be predicted.

[0038] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0039] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 A flowchart of an intelligent prediction method for fish spawning behavior provided by the present invention;

[0041] Figure 2 Here is a flowchart of an intelligent prediction method for fish spawning behavior provided in Embodiment 1 of the present invention;

[0042] Figure 3 The image shows the prediction results of an intelligent prediction method for fish spawning behavior provided in Embodiment 2 of the present invention. Detailed Implementation

[0043] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0044] The terms "comprising" and "having," and any variations thereof, in the specification, embodiments, claims, and drawings of this invention are intended to cover non-exclusive inclusion, such as including a series of steps or units.

[0045] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0046] like Figure 1 As shown, an intelligent prediction method for fish spawning behavior includes: collecting and organizing data on fish breeding seasons; screening key environmental factors affecting fish spawning; constructing a zero-inflation two-part model based on the data; determining the time period T with the highest spawning rate and the highest peak probability based on the zero-inflation two-part model; constructing a fish spawning prediction chaotic system model based on attractor reconstruction; and applying the reconstructed fish spawning prediction chaotic system model to predict the spawning scale under environmental conditions of the test dataset V based on the time period T.

[0047] Example 1

[0048] like Figure 2 As shown, an intelligent prediction method for fish spawning behavior coupled with a zero-inflation model and a chaotic system model is presented. The intelligent prediction method includes:

[0049] S1. Collect and organize data such as daily spawning volume, daily flow rate, and daily water temperature during the fish breeding season;

[0050] S2. Based on literature review and data mining, key environmental factors affecting fish spawning were screened.

[0051] S3. Organize daily egg production data and key environmental factors to form dataset F;

[0052] S4. Divide the dataset into a training dataset S and a test dataset V;

[0053] S5. Based on whether the daily egg production is zero, the training dataset S is divided into zero-value data and non-zero-value data, and a zero-inflation two-part model is constructed.

[0054] S6. Using the zero-inflation two-part model, estimate the probability of fish spawning response under environmental conditions of the test dataset V, and determine the time period T with the highest spawning rate and the highest peak probability.

[0055] S7. Construct a chaotic system model for predicting fish spawning based on attractor reconstruction;

[0056] S8. Based on the time period T determined in S6, use the fish spawning prediction chaotic system model reconstructed in S7 to predict the spawning scale under environmental conditions of the test dataset V.

[0057] S9. Compare the predicted egg-laying scale with the actual egg-laying data to evaluate the model performance.

[0058] Step S2, screening environmental factors affecting fish spawning, specifically includes:

[0059] Based on the ecological habits that affect fish reproduction and the characteristics of environmental changes during the breeding season, environmental factors that affect fish spawning and reproduction were initially selected.

[0060] Environmental factors include, but are not limited to, the current daily average flow rate, the previous day's flow rate, the previous two days' flow rate, the 1-day flow rate variation, the 2-day flow rate variation, the current daily average water temperature, the 1-day water temperature variation, the 2-day water temperature variation, the previous day's flow rate, and the previous two days' flow rate.

[0061] Independence analysis was performed on each environmental factor to eliminate factors with collinearity.

[0062] The daily spawning data of fish is converted into a proportion of the total spawning in a year, eliminating the differences in spawning scale between years and making it conform to the characteristics of semi-continuous data.

[0063] The original observation sequences, such as daily flow rate and daily water temperature, are standardized using z-score. First and second-order difference sequences are standardized in segments based on the positive and negative values ​​of the variables to preserve positive and negative information in the difference sequences.

[0064] The receiver operating characteristic (ROC) method was used to further identify key environmental factors affecting fish reproduction through multiple index combinations. The original observed variables, first-order difference variables, second-order difference variables, and 1-2 day lag values ​​of these variables were used to form index combinations. A joint discriminant model of multiple indicators was established using both logical sum and logical OR methods. The binary classification response index of the model under each index combination was calculated, and the index combination with the best comprehensive performance was selected as the key environmental factor for fish spawning.

[0065] S6 applies the zero-inflation two-part model to estimate the probability of fish spawning response under environmental conditions in the test dataset V, specifically including:

[0066] A two-part model with zero inflation is used to describe the posterior distribution of semi-continuous spawning data within the training dataset S: the first part describes the distribution of zero-valued data, and the second part describes the distribution of spawning data affected by a combination of multiple environmental factors. Y is used as the model's model. i Y represents the relative proportion of fish spawning. i The distribution is shown below:

[0067] Y i ~0,1-p i

[0068] Y i ~beta(μ i φ,(1-μ i )φ),pi

[0069] The zero-valued part uses a logistic regression model, and the non-zero-valued part uses a beta regression model. The density function of the beta distribution is expressed as:

[0070]

[0071] Using a generalized linear model to evaluate probability p i and mean μ i Perform regression analysis separately, i.e.

[0072]

[0073]

[0074] Where α and β represent the regression coefficients of the two parts, respectively. This represents the environmental factor vector. The parameters of the two parts of the model will be estimated separately using the Monte Carlo Markov Chain (MCMC) algorithm.

[0075] Based on the environmental factors of the test dataset V, a zero-inflation beta regression model with converged parameters was used to obtain the response probability and peak proportion of fish spawning in the validation set, and the time period T with the highest spawning proportion and the highest peak probability was selected.

[0076] S7 constructs a chaotic system model for predicting fish spawning based on attractor reconstruction, specifically including:

[0077] Based on the periods of highest spawning rate and highest peak probability predicted by the zero-inflation beta regression model, the spawning scale is further predicted using the attractor reconstruction method based on state space.

[0078] Based on the training dataset S, environmental factors and spawning data are combined to form a high-dimensional state space. The embedding dimension E of the state space is determined by applying the correlation dimension method, with spawning data and its own lag as variables.

[0079] Given a time series X = {x1, x2, ... x...} n}, extract E time-delay vectors of X with a delay time of τ, and construct the vector {x t ,x t-τ ,x t-2τ ,…,x t-(E-1)τ}, t=1,2,…,n;

[0080] Arbitrarily choose a set of E-dimensional state space points {X} t A point X in} i ={x i ,x i-τ ,…x i-(E-1)τ Using} as a reference point, calculate the distances from the remaining n-1 points to the reference point;

[0081] Defined by point X i For a volume element centered at a small scalar r with radius r, count the number of points falling outside the volume element to obtain the correlation function.

[0082]

[0083] In the formula, H represents the step function.

[0084]

[0085] The correlation function C(r) is a cumulative distribution function, which has the following properties:

[0086]

[0087] In the formula, D(E,r) is a constant related to E and r; d max Let C(r) be the maximum stretching distance of the attractor in E-dimensional space; for a certain appropriate range of values ​​of r, C(r) and the dimension E of the dynamical system satisfy a logarithmic linear relationship. If D(E,r) is the slope of the curve ln C(r) ~ ln r, then the correlation dimension is...

[0088] Given an initial embedding dimension E, take the segment with the best linearity from the double logarithmic curve ln C(r) ~ ln r and perform linear fitting. The slope of the fitted line is the correlation dimension.

[0089] Gradually increase the embedding dimension E and repeatedly calculate the correlation dimension D until the correlation dimension D no longer increases with the embedding dimension E, but fluctuates within a certain error range;

[0090] E-dimensional environmental factor combinations are generated from the variable set, the attractor manifold of multi-view embedding is reconstructed, and the simplex projection method is used to predict fish spawning behavior.

[0091] Simplex projection calculates the nearest neighbor of the target point in the reconstructed attractor and uses the average of the evolution trajectories of the nearest neighbors to predict the future state;

[0092] Given a univariate time series X = {x1, x2, ...}, use the three time delay vectors {x1, x2, ...} of X. t ,x t-τ ,x t-2τ ,…,x t-(E-1)τ}, forming an E-dimensional shadow manifold M X Where E < 2D + 1, D represents the spatial dimension of the system, and τ represents the time lag;

[0093] For the target point x(t) * ), calculate the shadow manifold M X All points xi With x(t) * The Euclidean distance d(x,y) is, i.e.

[0094] Assume x n(1) It is x(t) * The nearest neighbor of ) has a time index of t. n(1) ;

[0095] x n(i) The time index corresponding to the i-th nearest neighbor is t. n(i) ;

[0096] Each nearest neighbor is assigned a weight w based on the Euclidean distance from the target point to its nearest neighbor. i ,Right now

[0097]

[0098] M X middle The E+1 nearest neighbors form a simplex;

[0099] If we obtain the evolution values ​​of E+1 points forward in time, then x(t) * The future state after p time steps The prediction is:

[0100]

[0101] In-sample predictions are performed on the training set data based on different variable combinations. The correlation coefficients of the in-sample prediction results are calculated, and the variable combinations are ranked according to the magnitude of the correlation coefficients, with the top combinations ranked first. Combinations of these will be selected as valid attractors.

[0102] In S8, the chaotic system model for predicting fish spawning, reconstructed in S7, is used to predict the spawning scale under environmental conditions of the test dataset V. Specifically, this includes:

[0103] forward The combination of variables is used to perform out-of-sample prediction on the test set data using the simplex projection method to obtain the prediction results.

[0104] The average of the prediction results is used as the final prediction result.

[0105] Example 2

[0106] A smart prediction method for fish spawning behavior coupled with a zero-inflation model and a chaotic system model includes:

[0107] I. Construct a dataset and screen environmental factors that affect fish spawning.

[0108] The selection of exogenous environmental factors is determined based on the spawning habits of fish and their water flow requirements. Taking drifting fish as an example, spawning requires suitable water temperature and a rising water level. After the water temperature conditions are met during the breeding season, a suitable rising water level will induce drifting fish to begin spawning. Spawning behavior mostly occurs during the rising water period; fish begin spawning 0.5 to 2 days after the river water rises. The longer the rising water lasts, the larger the spawning scale. Among these factors, water level, flow rate, water temperature, water level variation, flow rate variation, and water temperature variation are the key environmental factors affecting spawning volume. Through independence analysis, water level variation and flow rate variation show a highly linear relationship, so water level variation is excluded. The absolute number of fish spawns is converted into a proportion of the annual breeding scale, ensuring that the non-zero portion is distributed within the range of 0 to 1.

[0109] II. Inferring Fish Spawning Response Probability Based on a Zero-Inflated Two-Part Model

[0110] The optimal combination of environmental factors derived from the ROC method is used as a covariate in the zero-inflation model. The proportion of fish spawning is logit-transformed to correct right skewness and to verify whether the non-zero portion of the data follows the aforementioned beta distribution. The MCMC algorithm is used to estimate the parameters of the zero-inflation model and the posterior distribution of the fish spawning data. Based on the changes in the combination of environmental factors in the test dataset V, the response probability of fish spawning in the validation set is predicted, and the time periods with the highest spawning proportion and the highest peak probability are selected for further refined predictions.

[0111] III. Predicting Fish Spawning Scale Based on Attractor Reconstruction using Chaotic System Model

[0112] Using key environmental factors and their time lags, the correlation dimension of fish spawning sequences calculated using the training dataset S converges to 5. An arrangement of the five environmental factors is then used to form... A combination of variables. Before obtaining the in-sample fitting effect using multi-view embedding within the training dataset S. The reconstructed environmental factors and spawning data form the state-space attractor. Based on the spawning time period predicted by the zero-inflation two-part model, the peak spawning value within the validation set is predicted by simplex projection.

[0113] like Figure 3 The figure shows the prediction results of an intelligent prediction method for fish spawning behavior that couples a zero-inflation model and a chaotic system model.

[0114] Beneficial effects: By using historical spawning data of drifting fish and based on water level, flow rate, and water temperature data from hydrological stations, attractors can be reconstructed in state space to predict future spawning dynamics of fish.

[0115] The above specific embodiments further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligently predicting fish spawning behavior, characterized in that, The intelligent prediction method includes: Collect data on fish breeding seasons, screen key environmental factors that affect fish spawning, and compile them into a dataset; Construct a zero-inflation two-part model based on the dataset; Based on the zero-inflation two-part model, the time period T with the highest spawning rate and the highest probability of peak occurrence is determined; Construct a chaotic system model for predicting fish spawning based on attractor reconstruction; Based on the time period T, a reconstructed chaotic system model for predicting fish spawning is used to predict the spawning scale under environmental conditions in the test dataset V.

2. The intelligent prediction method for fish spawning behavior according to claim 1, characterized in that, The collection of data on fish breeding seasons, screening of key environmental factors affecting fish spawning, and compilation of the dataset specifically include: Collect and organize data such as daily spawning volume, daily flow rate, and daily water temperature during the fish breeding season; Based on the ecological habits that affect fish reproduction and the characteristics of environmental changes during the breeding season, environmental factors that affect fish spawning and reproduction were initially selected. The environmental factors include the current daily average flow rate, the previous day's flow rate, the previous two days' flow rate, the 1-day flow rate variation, the 2-day flow rate variation, the current daily average water temperature, the 1-day water temperature variation, the 2-day water temperature variation, the previous day's flow rate, and the previous two days' flow rate. Independence analysis was performed on the environmental factors to eliminate factors with collinearity. Converting daily fish spawning data into a percentage of total annual spawning eliminates inter-year spawning variations and aligns with the characteristics of semi-continuous data. The original observation sequence was z-score normalized. For first- and second-order difference sequences, segmentation and standardization are performed based on the positive and negative values ​​of the variables; The receiver operating curve method was used to identify a combination of key environmental factors affecting fish reproduction. Data on daily egg production and key environmental factors were compiled to form dataset F.

3. The intelligent prediction method for fish spawning behavior according to claim 1, characterized in that, The construction of the zero-inflation two-part model based on the data specifically includes: The dataset is divided into a training dataset S and a test dataset V; Based on whether the daily egg production is zero, the training dataset S is divided into zero-value data and non-zero-value data to construct a zero-inflation two-part model.

4. The intelligent prediction method for fish spawning behavior according to claim 1, characterized in that, The specific time period T, which is determined based on the zero-inflation two-part model as having the highest spawning rate and the highest peak probability, includes: Using the zero-inflation two-part model, the probability of fish spawning response under environmental conditions of the test dataset V is estimated, and the time period T with the highest spawning rate and the highest peak probability is determined. The posterior distribution of semi-continuous oviposition data in the training dataset S is described by a two-part model based on zero inflation: the first part describes the distribution of zero-value data, and the second part describes the distribution of non-zero-value oviposition data driven by environmental factors. The zero-valued part uses a logistic regression model, and the non-zero-valued part uses a beta regression model. A generalized linear model is used to evaluate the probability p. i and mean μ i Perform regression analysis separately; The parameters of the two parts of the model will be estimated separately using the Monte Carlo Markov chain algorithm; Based on the environmental factors of the test dataset V, a zero-inflation beta regression model with converged parameters was used to obtain the response probability and peak proportion of fish spawning in the validation set, and the time period T with the highest spawning proportion and the highest peak probability was selected.

5. The intelligent prediction method for fish spawning behavior according to claim 1, characterized in that, The intelligent prediction method also includes: comparing the predicted egg-laying scale with the measured egg-laying data to evaluate the model performance.

6. The intelligent prediction method for fish spawning behavior according to claim 1, characterized in that, The construction of the fish spawning prediction chaotic system model based on attractor reconstruction specifically includes: Based on the training dataset S, the embedding dimension E of the state space composed of environmental factors and spawning data is determined by applying the correlation dimension method. From a set of n environmental factor variables, an E-dimensional combination of environmental factors is generated. Based on the multi-view embedding method, this combination is reconstructed in the training dataset S. Each attractor represents a neighboring trajectory in the state space of fish spawning driven by environmental factors, and the attractors are sorted according to the magnitude of the predicted Pearson correlation coefficient within the sample.

7. The intelligent prediction method for fish spawning behavior according to claim 1, characterized in that, The predicted spawning scale of the chaotic system model under the environmental conditions of the test dataset V specifically includes: Select the top rankings The attractor is used for out-of-sample prediction of the test dataset V, employing the simplex projection method to predict... For the spawning behavior of a group of fish, the mean of all attractor predictions is taken as the final prediction result.

Images