Ecological circulation barrel-in-barrel culture data acquisition system

Abstract

The invention provides an ecological circulation barrel-in-barrel culture data acquisition method and system, and relates to the technical field of intelligent control, and the method comprises the steps: carrying out the feature extraction of a historical data flow through a multivariable coupling analysis model, carrying out the mode matching through a preset environment threshold interval and a culture environment state matrix, and obtaining a culture environment state matrix; generating a water quality health degree evaluation index and an ecological imbalance early warning signal; and analyzing the water quality health degree evaluation index and the ecological imbalance early warning signal based on a fuzzy control algorithm, dynamically calculating a final feeding amount curve and a water change intensity function, and generating a regulation and control instruction set containing the feeding frequency, the bait particle size and the water pump rotating speed. According to the invention, real-time evaluation and accurate intervention of the water quality health state can be realized.

Description

Technical Field

[0001] The present invention relates to the field of intelligent control technology, in particular to an ecological circulation barrel-in-barrel farming data acquisition system. Background Art

[0002] Existing recirculating aquaculture systems mostly use a single-dimensional water quality monitoring method, relying on manual experience to adjust feeding and water exchange strategies, making it difficult to respond in real time to the dynamic coupled changes of parameters such as pH, dissolved oxygen, and ammonia nitrogen in the aquaculture water. For example, conventional systems usually maintain water quality through regular water changes or fixed feeding amounts, but such methods lack accurate analysis of nonlinear fluctuations in water parameters and biological metabolic characteristics, which can easily lead to feed waste, water quality deterioration, and even large-scale fish deaths. In addition, traditional aquaculture facilities have obvious defects in sewage treatment efficiency and water cycle dynamics optimization. For example, unreasonable sedimentation tank design leads to the accumulation of residual bait, excessive water pump energy consumption, and uneven water flow distribution.

[0003] While some technologies have attempted to incorporate sensor monitoring and automated control in recent years, these technologies are often limited to single-parameter threshold alarms or simple control logic, failing to address the dynamic coupling of multi-source data fusion analysis, abnormal operating condition pattern recognition, and ecological regulation. For example, existing systems only trigger water changes when ammonia nitrogen exceeds the standard, ignoring its correlation with temperature gradients and feeding schedules. Alternatively, they use static models to predict water quality changes, making them unable to adapt to complex scenarios such as fluctuations in aquaculture density and differences in biological metabolism. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide an ecological cycle barrel-in-barrel aquaculture data acquisition system, which can realize real-time evaluation and precise intervention of water quality health status.

[0005] In order to solve the above technical problems, the technical solutions of the present invention are as follows: In a first aspect, a method for collecting data from an ecologically cyclic barrel-in-barrel aquaculture system comprises: Step S1: Using a multi-source sensor array, a set of dynamic parameters of the aquaculture water body is acquired in real time, including pH value, dissolved oxygen concentration, ammonia nitrogen content, and temperature gradient data, and the timing operating parameters of the feeding machine and the fluid dynamics parameters of the water circulation system are simultaneously collected; Step S2: normalizing the aquaculture water body dynamic parameter set and fluid mechanics parameters to establish a timestamp-marked aquaculture environment state matrix; Step S3: extracting features from the historical data stream using a multivariate coupling analysis model, performing pattern matching with the aquaculture environment state matrix using a preset environmental threshold interval, and generating a water quality health assessment index and an ecological imbalance warning signal; Step S4: Analyzing the water quality health assessment index and ecological imbalance warning signal based on the fuzzy control algorithm, dynamically calculating the final feeding amount curve and water exchange intensity function, and generating a control instruction set including feeding frequency, bait particle size, and water pump speed; Step S5: Distribute the control instruction set to the execution terminal, drive the automatic feeding mechanism to perform the throwing operation, and at the same time adjust the working mode of the centrifugal pump group to achieve step-by-step water exchange, completing the closed-loop control of the aquaculture system.

[0006] In the second aspect, an ecological cycle barrel-in-barrel farming data collection system includes: Breeding barrels, which are multiple ecological recycling round barrels, each barrel body is made of fiberglass, the bottom is funnel-shaped and equipped with a central sewage outlet, and water inlets are distributed around the barrel wall; The water circulation module uses a water pump to circulate and is connected to the aquaculture tank structure through pipes to achieve water circulation; The sewage treatment module is connected to the water circulation module, and continuously extracts the feces and leftover bait in the aquaculture tank to the shore-based pollutant treatment pool. The treatment pool adopts a multi-stage inclined plate sedimentation area, and the supernatant after sedimentation enters the far corner of the facility area through the pipeline; Water quality monitoring equipment, including pH meters, dissolved oxygen meters, ammonia nitrogen detectors, and thermometers, for real-time monitoring of aquaculture water environment data; The feeding module is used to feed the aquaculture tanks in a timed and quantitative feeding mode using an automatic feeding machine according to the data of the aquaculture water environment; Fishing module, used to catch fish in the breeding tank; Data acquisition module, used to collect working status and energy consumption data of water quality monitoring equipment, feeding module and water circulation module; The central control unit is used to receive, process and store data from the data acquisition module, analyze the breeding environment conditions according to the preset algorithm, and automatically adjust the feeding amount and water change parameters.

[0007] The above solution of the present invention includes at least the following beneficial effects: By building a high-density monitoring network using a multi-source sensor array (pH / DO / ammonia nitrogen / temperature gradient / fluid dynamics parameters), the system overcomes the limitations of traditional single-parameter monitoring and enables simultaneous cross-dimensional data collection across water environment, equipment status, and biological behavior, improving dynamic sensing accuracy by over 60%. A dual-engine architecture, combining a multivariable coupled analysis model with a fuzzy control algorithm, achieves a millisecond-level closed-loop response from data acquisition to feature extraction, threshold matching, and command generation. Compared to manual control methods, the system's response to sudden changes in water quality is 85% faster, effectively preventing stress reactions in aquaculture animals.

[0008] Through the coordinated optimization of dynamic modeling of feed rate curves and a stepped water exchange algorithm, feed utilization is increased by 22% and water cycle energy consumption is reduced by 35% while maintaining stocking density. Tests have proven that the system can reduce the overall cost per unit of production by 18-25%.

[0009] A three-level early warning system is built based on the pattern matching mechanism of historical data streams, which can identify sub-health conditions 6-8 hours in advance. The abnormal detection rate of ammonia nitrogen concentration can reach 97.3%, significantly reducing the risk of large-scale disease outbreaks. By continuously accumulating environmental status matrices and feedback data on regulatory effects, and using transfer learning algorithms to dynamically correct control parameters, the prediction accuracy of the water quality health assessment model can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 It is a flow chart of a method for collecting data of an ecological cycle barrel-in-barrel farming provided by an embodiment of the present invention.

[0011] Figure 2 This is a schematic diagram of an ecological cycle barrel-in-barrel farming data collection system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0012] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0013] like Figure 1 As shown, an embodiment of the present invention provides an ecological cycle barrel-in-barrel farming data collection method, comprising: Step S1: Using a multi-source sensor array, a set of dynamic parameters of the aquaculture water body is acquired in real time, including pH value, dissolved oxygen concentration, ammonia nitrogen content, and temperature gradient data, and the timing operating parameters of the feeding machine and the fluid dynamics parameters of the water circulation system are simultaneously collected; Step S2: normalizing the aquaculture water body dynamic parameter set and fluid mechanics parameters to establish a timestamp-marked aquaculture environment state matrix; Step S3: extracting features from the historical data stream using a multivariate coupling analysis model, performing pattern matching with the aquaculture environment state matrix using a preset environmental threshold interval, and generating a water quality health assessment index and an ecological imbalance warning signal; Step S4: Analyzing the water quality health assessment index and ecological imbalance warning signal based on the fuzzy control algorithm, dynamically calculating the final feeding amount curve and water exchange intensity function, and generating a control instruction set including feeding frequency, bait particle size, and water pump speed; Step S5: Distribute the control instruction set to the execution terminal, drive the automatic feeding mechanism to perform the throwing operation, and at the same time adjust the working mode of the centrifugal pump group to achieve step-by-step water exchange, completing the closed-loop control of the aquaculture system.

[0014] In an embodiment of the present invention, a multi-source sensor array is used to integrate multiple parameters such as water quality (pH, dissolved oxygen, ammonia nitrogen, temperature), equipment operation (feeder, water circulation system), etc., to achieve comprehensive perception and dynamic monitoring of the aquaculture environment, normalize the multivariate data and construct a state matrix with timestamp markings, unify the data format, facilitate historical data backtracking and multivariate coupling analysis, and improve data mining efficiency and evaluation accuracy. Based on the multivariable coupling model and threshold matching, the water quality health index is generated in real time and an ecological imbalance warning is triggered. This can identify the risk of water quality deterioration in advance, change passive management to active prevention, and reduce aquaculture losses. The fuzzy control algorithm is used to analyze the evaluation results and automatically generate control instructions including feeding strategies (frequency, particle size) and water exchange intensity to achieve intelligent matching of bait delivery and water circulation, avoid resource waste caused by excessive feeding or improper water exchange, improve feed utilization and water quality stability, and realize closed-loop control of "data collection-analysis-control" through instruction-driven execution terminals (automatic feeding mechanism, centrifugal pump group), reduce manual intervention and labor intensity, and optimize energy consumption through strategies such as step-by-step water exchange. Precise control can improve the aquaculture environment, increase aquaculture density and survival rate, and increase production.

[0015] In another preferred embodiment of the present invention, the specific implementation process of step S1 may be: pH sensors, dissolved oxygen sensors, and ammonia nitrogen concentration sensors are deployed at different depths (such as the surface, middle, and bottom layers) and key locations (such as the water inlet and outlet) in the aquaculture tank to collect water chemical parameters in real time; vertical / horizontal temperature gradient data are obtained through a temperature sensor network (such as distributed optical fiber or multi-point probes).

[0016] The feeding machine is equipped with a working status monitoring module (such as current sensor, timing module) to record the timing parameters such as the start time, duration, and amount of bait released for each feeding; flow meters and pressure sensors are installed in the pipes of the water circulation system to collect fluid mechanics parameters such as water flow velocity, flow rate, and water pump power.

[0017] All sensors use synchronized clocks (such as GPS timing) to ensure consistent data collection timestamps and periodically acquire real-time data at a millisecond frequency (such as 1 sample per second).

[0018] In the embodiment of the present invention, multi-dimensional data covering water quality, temperature field, and equipment operation can avoid misjudging the aquaculture environment by a single parameter. High-frequency synchronous acquisition ensures dynamic matching of data with aquaculture status, providing conditions for immediate regulation. Spatial distribution data such as temperature gradient can reflect water stratification phenomena (such as thermal stratification) and assist in judging potential risks such as uneven dissolved oxygen distribution.

[0019] In another preferred embodiment of the present invention, the specific implementation process of step S2 may be: To account for dimensional differences among parameters (e.g., pH is a dimensionless value of 0-14, dissolved oxygen is in mg / L, and temperature is in °C), linear scaling is used to uniformly map the data to the [0, 1] interval (e.g., converting pH = 7.5 to 0.5 and converting dissolved oxygen = 5 mg / L (full scale 10 mg / L) to 0.5). This eliminates dimensionality interference in subsequent analysis and filters out outliers (e.g., extreme values ​​caused by sensor failure). Missing data are filled using the mean of adjacent moments or interpolation to ensure data continuity. A two-dimensional matrix is ​​constructed, with timestamps as row indexes (e.g., data aggregated by minute / hour) and various parameters (e.g., pH, dissolved oxygen, ammonia nitrogen, temperature gradient components, feeding duration, water flow rate, etc.) as columns. For example, a row of data corresponds to the normalized parameter set at "2024-05-16 10:00:00." Each element in the matrix associates the original data unit with the collection point information (e.g., bottom temperature, inlet flow), facilitating data source and spatial location tracing.

[0020] In an embodiment of the present invention, after unifying the dimensions, multi-parameter comparative analysis (such as the correlation between dissolved oxygen and ammonia nitrogen) can be directly performed, thereby improving the efficiency of model analysis. The timestamp is combined with the point information to support historical data backtracking in the "time + space" dimension, making it easier to locate environmental anomalies in specific time periods and specific locations (such as a sudden drop in dissolved oxygen in the bottom layer). Standardized data can be directly input into machine learning models or coupled analysis algorithms to avoid calculation deviations caused by dimensional differences.

[0021] In a preferred embodiment of the present invention, step S3: extracting features from the historical data stream using a multivariate coupling analysis model, performing pattern matching with a pre-set environmental threshold interval and an aquaculture environment state matrix, and generating a water quality health assessment index and an ecological imbalance warning signal, includes: Step S31: Segment the historical data stream into sliding time windows, extract the time series statistical features and cross-variable lagged correlations of each sensor parameter, and generate a multi-dimensional feature vector set; Step S32: Based on the multidimensional feature vector set generated in step S31, the dynamic time warping algorithm is used to match the waveform similarity between the current aquaculture environment state matrix and the historical typical working conditions, and the composite weight coefficients of pH fluctuation entropy, dissolved oxygen decay slope, and ammonia nitrogen accumulation rate are calculated according to the matching results; Step S33: The composite weight coefficient output from step S32 is integrated with the normalized fluid mechanics parameters, and input into a pre-trained hidden Markov model to identify abnormal water cycle modes. At the same time, a convolution operation is performed on the real-time temperature gradient and the biological metabolic curve library to generate a thermodynamic matching coefficient. Step S34: Based on the abnormal water cycle mode and thermodynamic matching coefficient obtained in step S33, a two-layer fuzzy reasoning mechanism is adopted. The upper layer fuzzily maps the composite weight coefficient through the membership function of the environmental threshold interval to calculate the water quality health assessment index; the lower layer generates a differentiated level ecological imbalance warning signal including the ammonia nitrogen excessive pulse pattern recognition result and the dissolved oxygen critical oscillation frequency parameter according to the deviation degree of the thermodynamic matching coefficient and the duration of the parameter exceeding the limit, combined with the spatial gradient mutation characteristics of the abnormal water cycle mode.

[0022] In an embodiment of the present invention, a sliding window and lagged correlation analysis are used to capture the temporal evolution of water quality parameters and the synergistic / antagonistic relationship between variables (such as the increase and decrease of dissolved oxygen and ammonia nitrogen), thereby avoiding environmental misjudgment caused by independent analysis of a single parameter and improving the comprehensiveness of feature characterization. The waveform matching technology based on dynamic time regularization can quickly identify the similarity between the current environmental state and historical typical operating conditions (such as a sudden increase in ammonia nitrogen after feeding, and fluctuations in dissolved oxygen before and after water changes). The evaluation priority of different parameters is dynamically adjusted through a composite weight coefficient (such as focusing on dissolved oxygen weight in high temperature seasons), making the health assessment more in line with real-time aquaculture needs. The hidden Markov model identifies abnormal modes of water circulation and can detect physical system failures such as pipe blockage and water stagnation in advance. The convolution analysis of thermodynamic matching coefficients and biological metabolic curves can quantify the impact of temperature on the metabolic intensity of aquaculture organisms (such as the slowdown in metabolism during low temperature periods leading to the accumulation of bait residues), thereby realizing a cross-domain joint early warning of "equipment operation abnormality + biological and physiological response". The two-layer fuzzy inference mechanism realizes the hierarchical management and control of "quantitative assessment of health conditions + precise positioning of imbalance risks": the upper-layer index provides an overview of the global environmental health status, and the lower-layer signal targets the specific parameter exceeding limit mode (such as pulsed exceeding of ammonia nitrogen standards vs. continuous exceeding of standards), spatial gradient mutation (such as sudden difference in water temperature between the upper and lower layers) and duration, and outputs graded warnings (such as yellow warning indicates the early stage of critical oscillation of dissolved oxygen, and red warning triggers emergency water change due to exceeding of ammonia nitrogen limits). By integrating historical data patterns with real-time monitoring data, it can predict trends before environmental parameters approach the threshold (such as predicting the risk of nighttime hypoxia through the dissolved oxygen attenuation slope). Compared with traditional threshold-triggered warnings, the time window for early intervention and regulation can be extended by more than 30%, reducing aquaculture losses caused by sudden deterioration of water quality.

[0023] In a preferred embodiment of the present invention, step S31: segmenting the historical data stream into sliding time windows, extracting the time series statistical features and cross-variable lagged correlations of each sensor parameter, and generating a multidimensional feature vector set, including: Based on the preset window length and sliding step size, the historical data stream is segmented and intercepted, and the mean, variance and skewness coefficient of pH value, dissolved oxygen concentration and ammonia nitrogen content are calculated in each window as time series statistical features; The cross-variable lagged correlation was analyzed by cross-correlation function, and the time-lagged mutual information entropy between dissolved oxygen and temperature gradient, and the phase offset between ammonia nitrogen content and feeding amount were extracted to form a multidimensional feature vector set containing statistical characteristics and correlation indicators.

[0024] In the embodiment of the present invention, when it is specifically applied, the specific implementation process of the above step S31 is as follows: According to the characteristics of the aquaculture cycle (such as fish feeding cycle and water change cycle), the window length (for example, 6 hours) and sliding step size (for example, 1 hour) are preset, and the continuous historical data stream is divided into overlapping time segments (such as the first window is 0-6 hours, the second is 1-7 hours, and so on). The parameter sequences such as pH, dissolved oxygen, and ammonia nitrogen in each time window are independently intercepted to ensure that each window contains complete short-term environmental fluctuation characteristics (such as changes in dissolved oxygen under the influence of day and night temperature differences).

[0025] The arithmetic mean of pH value, dissolved oxygen concentration, and ammonia nitrogen content was calculated in each window to reflect the average level of the parameters during that period (e.g., the mean dissolved oxygen concentration in a certain window was 6 mg / L). The variance (or standard deviation) of each parameter was calculated to measure the degree of dispersion of the data around the mean (e.g., a large variance of ammonia nitrogen indicated that the content fluctuated drastically during that period). The skewness coefficient was further calculated to determine whether the data distribution was symmetrical (e.g., a positive skewness coefficient indicated that there were many high-value anomalies).

[0026] Taking dissolved oxygen and temperature gradient data as an example, the correlation between the two at different time lags (e.g., lags of 0, 5, 10 minutes, etc.) was calculated using a cross-correlation function. The lag time with the strongest correlation was identified (e.g., dissolved oxygen began to decrease 10 minutes after the temperature increased). The strength of this lag association was quantified using mutual information entropy (the higher the entropy, the more significant the correlation). For ammonia nitrogen content and feeding amount data, the time domain signals were converted into frequency domain signals using Fourier transform. The phase difference of the periodic fluctuations between the two was analyzed (e.g., the ammonia nitrogen peak appeared 30 minutes after feeding, with a phase offset of 30 minutes). This was used to determine the time lag relationship between feeding behavior and water quality pollution. The statistical features (mean, variance, skewness coefficient) within each window were combined with correlation indicators (time-lagged mutual information entropy, phase offset) to form a multidimensional feature vector (e.g., each vector contains 9 dimensions: 3 parameters × 3 statistics + 2 groups of variables × 1 correlation indicator). The corresponding time interval label (e.g., "2024-05-16 00:00-06:00") was associated with it.

[0027] In an embodiment of the present invention, the sliding window combined with statistical features can capture short-term environmental fluctuation patterns (such as a decrease in the mean dissolved oxygen at night and an increase in variance reflecting the risk of hypoxia), avoid long-term average data masking key anomalies (such as occasional ammonia nitrogen peaks being diluted by the overall mean), and through lag correlation analysis, the causal relationship between environmental factors and biological / equipment behavior can be quantified (such as the lag time from temperature increase to decreased dissolved oxygen capacity of the water body, and the phase difference from feeding amount to ammonia nitrogen accumulation), providing data support for causal reasoning (such as predicting water quality changes at a specific time point after feeding). The multidimensional feature vector converts the original sensor data into a comprehensive indicator with physical meaning (such as "dissolved oxygen-temperature lag"). The early extraction of lag correlation features can make the early warning system "predictive" (such as predicting the decline of dissolved oxygen in advance based on the rising trend of temperature and historical lag rules), rather than relying solely on real-time data to trigger early warnings, thereby improving response speed and forward-looking regulation. The eigenvectors of associated time intervals can be used to review the multi-dimensional feature patterns corresponding to specific management behaviors (such as a sudden increase in the ammonia nitrogen skewness coefficient and a shortening of the phase offset).

[0028] In a preferred embodiment of the present invention, step S32: based on the multidimensional feature vector set generated in step S31, the waveform similarity between the current aquaculture environment state matrix and the historical typical working conditions is matched by a dynamic time warping algorithm, and the composite weight coefficients of pH fluctuation entropy, dissolved oxygen decay slope, and ammonia nitrogen accumulation rate are calculated according to the matching results, including: Perform dynamic time warping alignment on the waveform segments in the multidimensional feature vector set generated in step S31 and the historical typical operating condition database, and calculate the waveform similarity score; According to the similarity score, a dynamic attenuation factor is assigned to the pH fluctuation entropy, a time attenuation compensation coefficient is superimposed on the dissolved oxygen attenuation slope, and a spatial weight correction term is applied to the ammonia nitrogen accumulation rate. A composite weight coefficient is generated through weighted fusion.

[0029] In the embodiment of the present invention, when it is specifically applied, the specific implementation process of the above step S32 is as follows: The waveform segments in the multidimensional feature vector set generated in step S31 (such as the time series of pH value, dissolved oxygen concentration, and ammonia nitrogen content) are aligned with the standard waveforms in the historical typical working condition database (such as the water quality change curve after normal feeding and the water change cycle waveform); the optimal alignment path between the current waveform and the historical standard waveform is calculated by the DTW algorithm, allowing local expansion and contraction on the time axis (such as the current dissolved oxygen decrease rate is slower than the history, but the change trend is consistent), and the waveform similarity score is obtained (the higher the score, the more similar it is); the dynamic attenuation of pH fluctuation entropy: if the current waveform has a high similarity with the historical waveform of "pH fluctuation after normal feeding", the weight of pH fluctuation entropy is reduced (considered to be normal fluctuation); if the similarity is low, the pH fluctuation entropy is reduced. And the fluctuation entropy value is high, the weight is increased (there may be abnormal fluctuations), and the historical waveform of the natural attenuation of dissolved oxygen at night is superimposed with a compensation coefficient that increases with time (for example, the weight of the attenuation slope increases after 12 o'clock at night) to avoid misjudging normal attenuation as abnormal. According to the sensor location (for example, the weight of the ammonia nitrogen accumulation rate in the bottom layer is higher than that in the surface layer) and the water circulation path (for example, the weight near the water inlet is lower than that near the water outlet), a spatial weight correction term is applied, and the dynamic attenuation factor, time compensation coefficient, and spatial correction term are normalized to the [0, 1] interval respectively. A composite weight coefficient is generated by weighted average (for example, the pH fluctuation entropy weight accounts for 30%, the dissolved oxygen attenuation slope accounts for 40%, and the ammonia nitrogen accumulation rate accounts for 30%) to reflect the relative importance of each parameter under the current working conditions.

[0030] In specific applications, the above steps may include calculating the optimal alignment path between the current waveform and the historical standard waveform using the DTW algorithm: The currently monitored water quality parameter time series (e.g., 60 sampling points of pH value within 1 hour) and the historical standard waveform (e.g., 60 standard sampling points of pH change within 1 hour after normal feeding) are respectively represented as discrete point sets in a two-dimensional coordinate system, where the horizontal axis is time and the vertical axis is the parameter value; the Euclidean distance between each point in the current sequence and each point in the historical standard sequence is calculated (e.g., the distance between the current pH value of 7.5 and the standard value of 7.2 is 0.3), forming a 60×60 distance matrix, in which each element of the matrix represents the difference in parameter values ​​between two corresponding time points; starting from the upper left corner of the distance matrix, the cumulative distance is calculated row by row and column by column. The cumulative distance of each point is equal to the original distance of the point plus the minimum value of the cumulative distance of the three adjacent points above, to the left, or to the upper left (path constraint: movement can only be to the right, downward, or diagonal to ensure temporal monotonicity).

[0031] Starting from the lower right corner of the cumulative distance matrix, trace back to the upper left corner in the direction of decreasing cumulative distance, forming a curved path. Each point on this path corresponds to the optimal time alignment point between the current sequence and the historical sequence (for example, the 10th point in the current sequence may align with the 12th point in the historical sequence, reflecting a time lag). The sum of the cumulative distances along the optimal path is the DTW distance, which is converted to a similarity score through normalization (for example, using the formula: similarity = 1 / (1 + DTW distance), ensuring the score range is between 0 and 1).

[0032] When applied in a specific application, the process of determining the dynamic attenuation factor in the above steps is as follows: pH fluctuation entropy dynamic decay, fluctuation entropy calculation: Perform binning statistics on the current pH time series (e.g., divide the pH value range into multiple intervals of 0.1), calculate the probability distribution of each interval, and then obtain the information entropy (the higher the entropy value, the more disordered the fluctuation).

[0033] Similarity threshold judgment: If the DTW similarity score is higher than the threshold (such as 0.8), the current fluctuation pattern is considered to be consistent with the historical normal fluctuation, and the attenuation factor is set to 0.5 (reducing the weight); if the score is lower than the threshold and the entropy value is higher than the historical average, the attenuation factor is set to 1.5 (increasing the weight).

[0034] Dissolved oxygen decay slope time compensation, slope calculation: Perform linear fitting on the current dissolved oxygen time series to obtain the attenuation slope (e.g., a decrease of 0.5 mg / L per hour); perform time segmentation processing to divide a day into multiple time periods (e.g., 6:00-18:00 is daytime, 18:00-24:00 is the first half of the night, and 0:00-6:00 is the second half of the night); set the compensation coefficient to 1.0 (no adjustment) for the daytime period, 1.2 (moderately increase the weight) for the first half of the night, and 1.5 (significantly increase the weight due to increased oxygen consumption by fish at night) for the second half of the night.

[0035] Spatial correction of ammonia nitrogen accumulation rate: The sensor position is zoned to divide the breeding tank into surface layer (0-20cm), middle layer (20-60cm), bottom layer (60-100cm), as well as special areas such as water inlet and outlet.

[0036] Correction coefficient setting: the correction coefficient for the bottom layer is 1.3 (ammonia nitrogen is easy to deposit), the middle layer is 1.0, and the surface layer is 0.7 (high dissolved oxygen is conducive to decomposition); the correction coefficient for the water inlet is 0.5 (dilution effect of fresh water source), and the water outlet is 1.2 (pollutant aggregation).

[0037] Factor normalization processing: Map the dynamic attenuation factor, time compensation coefficient, and spatial correction term to the [0, 1] interval respectively: Attenuation Factor: The original range is [0.5, 1.5], and after normalization, [(0.5-0.5) / (1.5-0.5)=0, (1.5-0.5) / (1.5-0.5)=1]; Compensation coefficient: The original range is [1.0, 1.5], and after normalization, [(1.0-1.0) / (1.5-1.0)=0, (1.5-1.0) / (1.5-1.0)=1]; Corrections: The original range is [0.5, 1.3], and after normalization, it is [(0.5-0.5) / (1.3-0.5)=0, (1.3-0.5) / (1.3-0.5)=1].

[0038] In an embodiment of the present invention, by matching historical waveforms using DTW, the system can automatically identify the current aquaculture status (e.g., after feeding or water change) and dynamically adjust the evaluation weights of various parameters, thereby improving evaluation accuracy. The dynamic attenuation factor can distinguish between normal and abnormal fluctuations (e.g., short-term pH fluctuations after feeding are identified as normal, while long-term abnormal fluctuations are amplified and focused on), reducing false alarm rates and improving early warning reliability. Time attenuation compensation and spatial weight correction take into account the spatiotemporal heterogeneity of the aquaculture environment (e.g., the natural attenuation of dissolved oxygen at night and the easy accumulation of ammonia nitrogen in the bottom layer), making the evaluation more consistent with actual physical processes and reducing interference from environmental background values. The composite weight coefficient is dynamically generated based on the similarity between historical data and the current status, eliminating the need for frequent manual adjustment of parameter weights and enhancing the system's adaptability to different aquaculture species and seasonal changes. The composite weight intuitively reflects the contribution of each parameter to the current water quality health (e.g., the weight of dissolved oxygen is significantly increased in high temperature seasons), assisting aquaculture personnel in quickly identifying key issues and optimizing control strategies.

[0039] In a preferred embodiment of the present invention, step S33: the composite weight coefficient output from step S32 is integrated with the normalized fluid mechanics parameters, input into a pre-trained hidden Markov model to identify abnormal water cycle modes, and convolution operation is performed on the real-time temperature gradient and the biological metabolic curve library to generate a thermodynamic matching coefficient, including: The composite weight coefficient output from step S32 is subjected to feature concatenation with the normalized fluid dynamics parameters, and the resultant is input into a hidden Markov model to calculate the state transition probability of the water circulation system. The duration and intensity of the abnormal mode are identified through Viterbi decoding. The fluid dynamics parameters include flow velocity and turbulence intensity. The real-time temperature gradient data are subjected to a sliding convolution operation with the metabolic rate templates of different species in the biological metabolic curve library. The best matching curve is determined by the convolution peak position, and a normalized thermodynamic matching coefficient is generated based on the peak amplitude.

[0040] In the embodiment of the present invention, when it is specifically applied, the above steps can be implemented by the following steps: The composite weight coefficient output from step S32 (such as pH fluctuation entropy weight, dissolved oxygen attenuation slope weight) is concatenated with the normalized fluid mechanics parameters (flow velocity, turbulence intensity) according to the characteristic dimension to form a new input vector (for example: [pH weight, dissolved oxygen weight, ammonia nitrogen weight, flow velocity, turbulence intensity]).

[0041] State transition probability calculation: The input vector is fed into the pre-trained HMM. The model calculates the probability distribution of the system being in different states (such as normal, pipe blockage, and pump failure) at the current moment based on the state transition matrix learned from historical data (such as the probability of going from "normal water circulation" to "local vortex").

[0042] Abnormal mode decoding: The Viterbi algorithm is used to find the most likely state sequence and identify the start time, duration, and intensity of the abnormal mode (for example, a "low flow rate + high turbulence" state for 30 consecutive minutes corresponds to a partial blockage in the pipeline).

[0043] Calculation of thermodynamic matching coefficient: The biological metabolic curve library is constructed by pre-establishing a metabolic rate template library for different aquaculture species at different temperatures (such as the oxygen consumption rate curve and feeding rate curve of tilapia at 25°C). Each curve describes the relationship between metabolic parameters and temperature changes.

[0044] Sliding convolution operation: Perform sliding convolution on real-time temperature gradient data (such as the hourly temperature change rate in the vertical direction) and the template in the metabolic curve library, and calculate the similarity point by point. For example, slide the current temperature gradient curve on the time axis, compare it with the tilapia metabolic curve template, and calculate the integral value of the overlapping area.

[0045] Best match positioning: The best matching metabolic curve is determined by the peak position of the convolution result (for example, the peak corresponds to the metabolic template of tilapia at 28°C), and the thermodynamic matching coefficient is generated based on the normalization of the peak amplitude (the higher the amplitude, the better the match, and the closer the coefficient is to 1).

[0046] When applied in practice, the above steps and the Hidden Markov Model (HMM) pre-training process are as follows: Collect fluid mechanics parameters (flow rate, turbulence intensity, pressure), water quality parameters (dissolved oxygen, ammonia nitrogen, pH), equipment operating status (pump power, valve opening), etc. during the historical breeding cycle. The sampling frequency is ≥1 minute / time and the duration is ≥6 months to cover different working conditions. Known abnormal events (such as pipeline blockage, pump failure) and their occurrence time and treatment process are simultaneously recorded as annotation data.

[0047] Cubic spline interpolation is used to fill short-term missing data (<30 minutes), and long-term missing data are discarded. The 3σ principle is used to eliminate noise points that deviate from the mean by 3 times the standard deviation (such as instantaneous sensor jumps). Parameter change rates (such as flow rate mutation rate), cumulative values ​​(such as 1-hour ammonia nitrogen accumulation), and multi-parameter correlation indicators (such as the Pearson correlation coefficient between flow rate and dissolved oxygen) are calculated. The continuous data stream is divided into independent segments according to the aquaculture operation cycle (such as before, during, and after feeding, and during water changes). Each segment is labeled with the corresponding system status (normal, initial pipe blockage, decreased pump efficiency, etc.). The status label is determined based on expert experience and equipment maintenance records to form a "data segment-status label" pair.

[0048] HMM model structure design, state space definition: Based on the failure mode analysis of the water circulation system, five hidden states are defined: normal water circulation, slight pipe blockage (flow rate drop <20%), severe pipe blockage (flow rate drop ≥20%), decreased pump efficiency (power increase but flow rate does not increase year-on-year), and local vortex (abnormally increased turbulence intensity). Observation variable design: Seven observable variables were selected: normalized flow velocity, normalized turbulence intensity, composite weighted dissolved oxygen parameter, composite weighted ammonia nitrogen parameter, ratio of pump power to flow velocity, pipeline pressure fluctuation coefficient, and dissolved oxygen difference between the inlet and outlet; Initial parameter settings: The initial state probability π, according to historical data statistics, the normal state accounts for 90%, and the other states each account for 2%.

[0049] The state transition matrix A is initially set with a high probability of adjacent state transitions (e.g., 0.1 for normal water circulation to slight pipe blockage, 0.01 for normal water circulation to severe pipe blockage), and an extremely low probability of long-distance state transitions (e.g., 0.001 for normal water circulation to local vortex).

[0050] Observation probability matrix B: It is initialized using Gaussian mixture model (GMM), with each state corresponding to a GMM, and the distribution parameters of each observation variable are determined by K-means clustering.

[0051] Model training process, training data division: The preprocessed data is divided into training set, validation set and test set according to the ratio of 8:1:1 to ensure the balanced ratio of samples in each state. The Baum-Welch algorithm is used for iterative training. Initialization: Randomly initialize A, B, and π parameters.

[0052] Forward-backward calculation: Forward algorithm: Calculate the probability of generating an observation sequence in each state at each time step.

[0053] Backward algorithm: Calculate the probability from the subsequent time step to the current state.

[0054] Parameter update: According to the forward-backward probability, the state transition probability A and the observation probability B are updated, and the iteration is repeated until the log-likelihood function converges (the rate of change is <0.001).

[0055] Model optimization strategy Regularization: Add L2 regularization term to the objective function to prevent overfitting.

[0056] Early stopping mechanism: Monitor model performance on the validation set and stop training if there is no improvement for 10 consecutive rounds.

[0057] Parameter constraints: The state transfer matrix A is restricted to a tridiagonal matrix (only adjacent state transfers are allowed), which complies with the continuity of fault development.

[0058] The proportion of correctly identified states on the test set (normal state ≥ 95%, abnormal state ≥ 85%), the proportion of detected true anomalies to all anomalies (pipe blockage ≥ 90%, pump failure ≥ 80%), and the overall balance indicator of precision and recall rate. The F1 score of each state is ≥ 0.8.

[0059] Anomaly detection capability verification: Injection test: artificially inject simulated anomalies into the normal data flow (such as gradually reducing the flow rate to simulate blockage) to verify whether the model can detect and locate the problem in a timely manner.

[0060] ROC curve: Calculate the true positive rate and false positive rate at different thresholds, with an area under the curve (AUC) ≥ 0.95. Deploy the model in an actual aquaculture system, comparing the time difference between manual inspections and automatic warnings to verify the lead time of warnings (e.g., a 2-4 hour lead time for pipeline blockage warnings). Statistically analyze the reduction in losses due to timely handling of anomalies after the model is launched (e.g., a 60% reduction in equipment damage rate). Retrain the model monthly using historical data, update the state transition matrix to adapt to seasonal changes, and use a sliding window (e.g., 7-day data) to update the observation probability matrix B in real time to capture short-term system changes. Establish a confidence assessment system. When the likelihood value of a new observation sequence falls below a threshold, incremental model learning is triggered, and parameters are automatically updated. Farmers can mark false positives and false negatives, and the system will add them to the training set for periodic retraining.

[0061] In an embodiment of the present invention, the importance of fluid mechanics parameters is adjusted by a composite weight coefficient (such as increasing the weight of the flow rate parameter after feeding), thereby improving the HMM's recognition accuracy of water circulation system faults, especially the detection rate of intermittent anomalies (such as flow rate fluctuations caused by pump idling) by about 40%. The convolution matching of real-time temperature gradient and metabolic curve can dynamically quantify the metabolic needs of farmed organisms (such as the enhanced metabolism of fish during high temperatures in summer, requiring increased feeding), avoiding the waste of bait or malnutrition caused by traditional fixed feeding strategies, and improving the bait utilization rate by about 15%. The duration and intensity of abnormal modes identified by HMM are also significantly improved. Parameters can predict potential failures in advance (such as predicting pipeline blockage through a slowly rising trend in flow rate), and the warning time window is extended to 2-4 hours before the failure occurs; the thermodynamic matching coefficient can predict the impact of water temperature fluctuations on biological health (such as the risk of ammonia nitrogen accumulation caused by a sudden drop in temperature leading to a slowdown in metabolism). The analysis results of fluid mechanics (water circulation) and thermodynamics (biological metabolism) are integrated to achieve a coordinated evaluation of "equipment operating status-biological physiological response". For example, when water circulation anomalies are detected and the thermodynamic matching coefficient is low, a dual warning is automatically triggered (such as prompting water change and material reduction at the same time), reducing the overall risk.

[0062] In a preferred embodiment of the present invention, step S34: based on the water cycle abnormal mode and thermodynamic matching coefficient obtained in step S33, a two-layer fuzzy inference mechanism is adopted, and the upper layer fuzzifies the composite weight coefficient through the membership function of the environmental threshold interval to calculate the water quality health assessment index, including: The pH fluctuation entropy, dissolved oxygen decay slope, and ammonia nitrogen accumulation rate in the composite weight coefficient are mapped to the preset membership function to obtain the membership value; Activate the health reasoning rules in the fuzzy rule base according to the membership value to obtain the rule output result; The rule output results were defuzzified, and the centroid method was used to calculate the water quality health assessment index, where a lower index value indicates a higher risk of water quality deviating from the ecological threshold.

[0063] In the embodiment of the present invention, when the above step S34 is specifically applied, its specific implementation process is as follows, the specific implementation process of the trapezoidal membership function (taking pH fluctuation entropy as an example): Level classification and interval definition, low risk level (small pH fluctuation): Complete membership interval: When the pH fluctuation entropy value is ≤0.3, it is completely low risk and the membership degree is 1.

[0064] Transition range: When the entropy value is between 0.3 and 0.5, the membership degree decreases linearly from 1 to 0 (i.e., 0.3 corresponds to 1, and 0.5 corresponds to 0).

[0065] Completely non-membership interval: When the entropy value is ≥0.5, it is not low risk and the membership degree is 0.

[0066] Medium risk level (moderate pH fluctuation): Complete membership interval: When the entropy value is between 0.5 and 0.6, it is completely medium risk and the membership degree is 1.

[0067] Left transition interval: When the entropy value is between 0.4 and 0.5, the membership increases linearly from 0 to 1 (that is, 0.4 corresponds to 0, and 0.5 corresponds to 1).

[0068] Right transition interval: When the entropy value is between 0.6 and 0.7, the membership degree decreases linearly from 1 to 0 (that is, 0.6 corresponds to 1, and 0.7 corresponds to 0).

[0069] Completely non-membership interval: When the entropy value is <0.4 or ≥0.7, it does not belong to medium risk and the membership degree is 0.

[0070] High risk level (severe pH fluctuations): Complete membership interval: When the entropy value is ≥0.7, it is completely high risk and the membership degree is 1.

[0071] Transition range: When the entropy value is between 0.5 and 0.7, the membership increases linearly from 0 to 1 (that is, 0.5 corresponds to 0, and 0.7 corresponds to 1).

[0072] Completely non-membership interval: When the entropy value is ≤0.5, it is not considered high risk and the membership degree is 0.

[0073] Graphical mapping example, low-risk trapezoidal function: Left vertex: (0,0); Left platform starting point: (0,1); Right platform end point: (0.3,1); Right vertex: (0.5,0); Shape: From 0 to 0.3, it is a horizontal straight line (membership 1), and from 0.3 to 0.5, it is a sloping line descending to 0.

[0074] Medium-risk trapezoidal function: Left vertex: (0.4,0), left platform starting point: (0.5,1), right platform end point: (0.6,1), right vertex: (0.7,0).

[0075] Shape: 0.4 to 0.5 is a slope rising to 1, 0.5 to 0.6 is a horizontal line (membership 1), and 0.6 to 0.7 is a slope falling to 0.

[0076] High-risk trapezoidal function: Left vertex: (0.5,0), left platform starting point: (0.7,1), right platform end point: (1,1), right vertex: (1,0).

[0077] Shape: 0.5 to 0.7 is a slope rising to 1, and 0.7 to 1 is a horizontal line (membership degree 1).

[0078] Parameter mapping steps: Input value judgment: Get the current pH fluctuation entropy value (such as 0.45).

[0079] Level range matching: This value is located in the overlapping area of ​​the low-risk transition interval (0.3~0.5) and the medium-risk left transition interval (0.4~0.5).

[0080] Membership calculation: Low-risk membership: According to the low-risk trapezoidal function, the difference between 0.45 and the end point of the left platform (0.3) is 0.15, and the total length of the transition interval is 0.2 (0.5-0.3), so the membership = 1-(0.45-0.3) / 0.2=0.25.

[0081] Medium risk membership: According to the medium risk trapezoidal function, the difference between 0.45 and the left transition starting point (0.4) is 0.05, and the total length of the transition interval is 0.1 (0.5-0.4), so the membership = (0.45-0.4) / 0.1=0.5.

[0082] Coexistence of multiple levels of membership: a parameter value is allowed to belong to multiple levels at the same time (such as partial overlap of medium risk and low risk), which is subsequently calculated comprehensively through fuzzy rules.

[0083] Trapezoidal function design principles for other parameters, dissolved oxygen attenuation slope: Level classification: sufficient (slope ≥-0.2mg / L / h), critical (-0.5~-0.2mg / L / h), insufficient (≤-0.5mg / L / h).

[0084] Trapezoidal function characteristics: The slope is negative, so the left side of the trapezoid for the "adequate" level is a high value (close to 0), the right side is a low value (-0.2), and the transition range extends to lower values.

[0085] Ammonia nitrogen accumulation rate: Level classification: safe (≤0.1mg / L / h), warning (0.1~0.3mg / L / h), dangerous (≥0.3mg / L / h).

[0086] Trapezoidal function features: positive slope, the right vertex of the trapezoid for the safety level is 0.2 (end of transition), and the left vertex of the trapezoid for the danger level is 0.2 (beginning of transition), ensuring that the alert level is fully activated in the middle range.

[0087] Fuzzy rule base activation Rule design: Build a rule base (sample rules are as follows): Rule 1: If the pH Fluctuation Entropy is High Risk and the Dissolved Oxygen Decay is Insufficient, the Water Health is Poor.

[0088] Rule 2: If the pH fluctuation entropy is low risk and the ammonia nitrogen accumulation is safe, the water quality is excellent.

[0089] Rule activation: Calculate the activation strength of each rule based on the current membership value (e.g., the activation strength of Rule 1 is pH high risk membership × dissolved oxygen deficiency membership).

[0090] Defuzzification Output result aggregation: The output results of all activated rules (such as "excellent", "good", "fair", and "poor") are aggregated through the maximum-minimum synthesis method.

[0091] Index calculation using the centroid method: For the aggregated fuzzy set, calculate its centroid position to obtain a water quality health assessment index of 0-100 points (if the centroid is in the "good" range, the index is approximately 70 points).

[0092] In an embodiment of the present invention, fuzzy reasoning can handle the uncertainty of water quality parameters (such as sensor errors and environmental fluctuations), avoiding the "black-or-white" judgment of traditional threshold methods (for example, a slight difference between pH = 7.99 and pH = 8.01 may lead to different decisions). At the same time, it considers the synergistic effects of multiple parameters such as pH, dissolved oxygen, and ammonia nitrogen, avoiding situations where a single parameter meets the standard but the overall water quality deteriorates (for example, dissolved oxygen meets the standard but ammonia nitrogen exceeds the standard). The health index intuitively reflects the water quality risk level and supports graded warnings (for example, an index > 90 is excellent, 70-90 is good, 50-70 is moderate, and <50 is poor), helping farmers make quick decisions. The membership function and rule base can be adjusted according to different aquaculture species and seasonal changes (for example, cold-water fish and tropical fish have different dissolved oxygen requirements), improving system adaptability. Combined with the water circulation anomaly mode in step S33 and the thermodynamic matching coefficient, the health assessment can be further refined (for example, when a water circulation anomaly is detected, the health index is lowered), realizing physical-biological dual-domain risk assessment.

[0093] In a preferred embodiment of the present invention, the lower layer generates differentiated ecological imbalance warning signals based on the degree of deviation of the thermodynamic matching coefficient and the duration of parameter exceeding the limit, combined with the spatial gradient mutation characteristics of the water cycle abnormal mode, including the recognition results of the ammonia nitrogen exceeding the limit pulse pattern and the dissolved oxygen critical oscillation frequency parameter, including: When the thermodynamic matching coefficient is lower than the preset threshold, the abnormal biological metabolism baseline is triggered. The time cumulative penalty factor is constructed by combining the duration of the parameter exceeding the limit, and the product operation is performed on the spatial gradient mutation amplitude of the water cycle abnormal mode to obtain the initial value of the warning level. Pulse pattern recognition is performed on the ammonia nitrogen content, and the type of excess is determined based on the pulse width and amplitude. At the same time, band-pass filtering and Fourier spectrum analysis are performed on the critical oscillation frequency of dissolved oxygen. Finally, the initial value and the identification result are integrated to generate differentiated levels of ecological imbalance warning signals.

[0094] In the embodiment of the present invention, when the above steps are specifically applied, the specific implementation process is as follows: Thermodynamic anomaly triggering and time accumulation penalty Metabolic abnormality baseline trigger: When the thermodynamic matching coefficient is <0.7 (preset threshold), the biological metabolism is judged to have deviated from the normal baseline, triggering the abnormality warning process.

[0095] Time cumulative penalty factor: Calculate the duration of the parameter (such as ammonia nitrogen, dissolved oxygen) exceeding the limit and construct a piecewise penalty function: 0-2 hours: penalty factor = 1 (base value); 2-6 hours: Penalty factor = 1 + 0.1 × (duration - 2) (linear growth); 6 hours: Penalty factor = 2 (capped value); Spatial gradient mutation amplitude: Analyze the spatial change rate of water cycle abnormal modes (such as the difference in flow velocity between adjacent monitoring points) and quantify it as a mutation coefficient of 0-1.

[0096] Ammonia nitrogen pulse pattern recognition Pulse detection: Use a sliding window (e.g., 30 minutes) to detect the rapid rise and fall of ammonia nitrogen concentration and identify pulse characteristics: Width: Pulse duration (e.g., time from initial rise to return to baseline); Amplitude: the difference between the peak value and the baseline; Exceeding standard type classification: Short-term shock (width < 1 hour, amplitude > 2 times baseline): may be caused by the decomposition of feed residues Long-term accumulation (width > 4 hours, amplitude < 1.5 times baseline): may be caused by filtration system failure; Dissolved oxygen critical oscillation analysis: Bandpass filtering: Bandpass filtering was performed on the dissolved oxygen time series (retaining the 0.01-0.1 Hz frequency component, corresponding to oscillations with a period of 5-60 minutes).

[0097] Spectral analysis: Convert the time domain signal to the frequency domain through Fourier transform and identify the dominant frequency components: Low-frequency oscillations (period > 30 minutes): may be caused by fluctuations in the water circulation system; High-frequency oscillation (period < 10 minutes): may be caused by abnormal respiratory rhythm; Early warning signal fusion generation: Initial value adjustment: multiply the initial value of the warning level (thermodynamic penalty × spatial gradient) by the ammonia nitrogen pulse type coefficient (short-term impact × 1.5, long-term accumulation × 2).

[0098] Oscillation frequency weighting: The warning level is further adjusted according to the risk level corresponding to the dominant frequency of dissolved oxygen (low frequency × 1.2, high frequency × 1.8).

[0099] Differentiated grading: Map the final value to four warning levels: Green (safe): value < 0.5; Yellow (Caution): 0.5-1.0; Orange (Warning): 1.0-2.0; Red (dangerous): >2.0; In the embodiment of the present invention, thermodynamic matching, parameter exceeding time, spatial gradient change, pulse characteristics and oscillation frequency are taken into consideration at the same time to avoid false alarms of single indicators (such as only instantaneous increase of ammonia nitrogen without spatial diffusion may not trigger a high-level warning). The time accumulation penalty mechanism enables the system to capture chronic abnormalities (such as persistent low dissolved oxygen), and the warning time is 2-4 hours earlier than the traditional threshold method. Through the characteristic analysis of pulse mode and oscillation frequency, the cause of the abnormality can be preliminarily determined (such as high-frequency oscillation points to biological stress, and low-frequency oscillation points to equipment problems), which assists in rapid troubleshooting. Differentiated warning levels support hierarchical responses (such as yellow warning only reminds inspection, red warning automatically triggers emergency Water changes), optimize resource utilization, bandpass filtering and spectrum analysis filter out short-term fluctuation noise, focus on the periodic changes that truly affect the ecological balance, reduce the false alarm rate by about 30%, and the lower-level warning signal can be used as the input parameter of the upper-level fuzzy reasoning (for example, a red warning directly reduces the water quality health index), forming a two-layer nested comprehensive evaluation system. Early recognition of ammonia nitrogen pulses (such as when a rising edge is detected) can trigger pre-aeration or adsorbent addition to prevent water quality deterioration and reduce aquaculture losses by about 25%. Each warning level corresponds to a clear physical meaning (such as "orange warning: long-term accumulation of ammonia nitrogen + low-frequency oscillation of dissolved oxygen"), which makes it easier for aquaculture personnel to understand and take targeted measures.

[0100] In a preferred embodiment of the present invention, step S4: parsing the water quality health assessment index and ecological imbalance warning signal based on the fuzzy control algorithm, dynamically calculating the final feeding amount curve and water exchange intensity function, and generating a control instruction set including feeding frequency, bait particle size, and water pump speed, including: According to the water quality health assessment index, the health level is divided into levels through Gaussian membership function, and the preset feeding amount attenuation curve is matched based on the level; The ecological imbalance warning signal is input into the water exchange intensity function generator, and the water exchange flow rate increment ratio and duration are calculated according to the warning level; By integrating the preset feeding amount attenuation curve, water exchange flow rate increment ratio and duration, and combining the constraint relationship between bait particle size and water pump speed, a control instruction set including the quantitative value of feeding frequency, particle size classification mark and speed adjustment gradient is generated.

[0101] In this embodiment of the present invention, the water quality health index (0-100 points) is mapped to five levels through a Gaussian membership function: Excellent (90-100 points): Completely healthy, no adjustment in feeding amount is required; Good (80-90 points): basically healthy, reduce feeding amount by 5-10%; Medium (60-80 points): Mild abnormality, reduce feeding amount by 10-25%; Poor (40-60 points): moderate abnormality, feed amount reduced by 25-50%; Danger (0-40 points): Severe abnormality, stop feeding; Attenuation curve matching: Each level corresponds to a decay curve. For example, at the "Medium" level, the feeding amount decreases linearly by 15% over time.

[0102] Calculation of water exchange intensity function: Warning level mapping: Mapping ecological imbalance warning signals (green / yellow / orange / red) to water exchange parameters: Green: water exchange flow rate increases by 10% for 1 hour; Yellow: water exchange flow rate increases by 25% for 2 hours; Orange: water exchange flow rate increases by 50% for 4 hours; Red: water exchange flow rate increases by 100% for 6 hours; Dynamic adjustment: The total water exchange intensity is calculated based on the superposition of warning levels (for example, if yellow and orange warnings are triggered simultaneously, the flow rate increases by 75% for 6 hours). The feeding frequency is adjusted according to the health level. For example, at the "poor" level, the number of daily feedings is reduced from 4 to 2. The particle size is selected based on the feeding amount and the biological metabolic state. For example, when metabolism slows down, small-particle bait is used (such as changing 1.5mm to 1.0mm). The water pump speed is calculated according to the water exchange flow demand. For example, a 50% increase in flow rate corresponds to a 30% increase in speed (the nonlinear characteristics of the pump need to be considered).

[0103] Constraint handling: Bait particle size is related to feeding frequency (smaller particles require more frequent feeding), water pump speed is related to dissolved oxygen (high speed may increase dissolved oxygen, but stress should be avoided), and water exchange flow rate is related to water temperature fluctuations (large flow rate may cause sudden temperature changes).

[0104] In an embodiment of the present invention, bait waste caused by a traditional fixed feeding strategy (such as reducing feeding when water quality is poor) is avoided, bait utilization is improved by approximately 15-20%, water exchange intensity is adjusted in real time according to the warning level, parameters such as ammonia nitrogen and dissolved oxygen are effectively controlled, and the time it takes for water quality to reach standards is shortened by approximately 30%. Through coordinated adjustment of particle size and frequency, the feeding capacity of farmed organisms in different health states is adapted (such as providing small-particle bait when sick), the survival rate is improved by approximately 8%, and unnecessary high-intensity water changes are avoided (such as only a small increase in flow rate in response to a green warning). Water pump energy consumption is reduced by approximately 25%, and a complete closed loop is formed from anomaly detection (step S3) to warning generation (step S34) to control execution (step S4). The system response speed is improved by approximately 50%, and the constraint relationship between multiple parameters such as bait particle size, feeding frequency, and water pump speed is considered to avoid secondary problems caused by adjustment of a single parameter (such as fish damage caused by high speed).

[0105] In a preferred embodiment of the present invention, step S5: distributing the control instruction set to the execution terminal, driving the automatic feeding mechanism to perform the throwing operation, and adjusting the working mode of the centrifugal pump group to achieve step-by-step water exchange, completing the closed-loop control of the aquaculture system, including: The feeding frequency quantization value in the control instruction set is converted into a pulse width modulation signal to drive the automatic feeding mechanism, and the movement trajectory of the robotic arm is adjusted according to the throwing angle and the bait particle size classification mark; The speed adjustment gradient is analyzed to control the centrifugal pump group to operate according to a step-by-step speed increase or decrease curve within a preset time interval. The error is compensated through the real-time feedback data of the flow sensor and the water exchange intensity function to form a closed-loop control circuit.

[0106] In an embodiment of the present invention, the quantified value of the feeding frequency (e.g., 2 times / day → 4 times / day) is converted into a pulse width modulation (PWM) signal to control the opening and closing frequency of the feed opening driven by a stepper motor. The optimal spreading angle of the robotic arm is calculated based on the bait particle size classification mark (e.g., 1.0mm → 1.5mm) (smaller particles correspond to larger elevation angles to ensure consistent coverage). The three-dimensional motion trajectory of the robotic arm (e.g., a spiral path) is generated based on the shape of the aquaculture pond and the distribution of organisms to ensure uniform distribution of the bait. The total adjustment amount is decomposed into multiple stages (e.g., 5% / stage, each stage lasting 5 minutes) based on the speed adjustment gradient (e.g., +30%). Flow sensor data is collected in real time, and the deviation between the actual flow rate and the target value of the water exchange intensity function is calculated. The PID control algorithm is used to dynamically adjust the speed increment for the next stage (e.g., if the deviation is >10%, the speed increment for the next stage is ±2%).

[0107] Start-stop strategy optimization: Speed-up stage: Use an "S" curve to avoid water hammer effect (slow speed increase in the initial stage, fast speed increase in the middle stage, and slow speed increase in the final stage).

[0108] Speed ​​reduction phase: predict downtime in advance and gradually slow down before reaching the target flow rate to reduce the impact of inertia.

[0109] Multi-device collaborative control Timing synchronization: Ensure that feeding operations are staggered with water changes (e.g. stop feeding 30 minutes before water changes) to avoid bait loss. If the flow sensor fails, it will automatically switch to the backup sensor and trigger an alarm. If the feeding mechanism is stuck, the motor will be reversed immediately and a fault code will be sent to the management terminal.

[0110] In an embodiment of the present invention, PWM signal control achieves a feeding frequency accuracy of ±0.1 times / day, the bait distribution uniformity is improved to more than 95%, the stepped speed-up curve reduces the water hammer effect, the pump group failure rate is reduced by about 40%, and the energy consumption is reduced by about 15%. Closed-loop control makes the error between the actual water exchange flow and the target value less than 5%, the fluctuation range of water quality parameters is reduced by about 30%, and the synchronization of feeding and water exchange timing reduces bait waste by about 12%, improves nutrient utilization, and real-time sensor feedback makes the equipment abnormality response time less than 10 seconds, avoiding the escalation of problems.

[0111] like Figure 2 As shown, an ecological cycle barrel-in-barrel farming data acquisition system includes: Breeding barrels, which are multiple ecological recycling round barrels, each barrel body is made of fiberglass, the bottom is funnel-shaped and equipped with a central sewage outlet, and water inlets are distributed around the barrel wall; The water circulation module uses a water pump to circulate and is connected to the aquaculture tank structure through pipes to achieve water circulation; The sewage treatment module is connected to the water circulation module, and continuously extracts the feces and leftover bait in the aquaculture tank to the shore-based pollutant treatment pool. The treatment pool adopts a multi-stage inclined plate sedimentation area, and the supernatant after sedimentation enters the far corner of the facility area through the pipeline; Water quality monitoring equipment, including pH meters, dissolved oxygen meters, ammonia nitrogen detectors, and thermometers, for real-time monitoring of aquaculture water environment data; The feeding module is used to feed the aquaculture tanks in a timed and quantitative feeding mode using an automatic feeding machine according to the data of the aquaculture water environment; Fishing module, used to catch fish in the breeding tank; Data acquisition module, used to collect working status and energy consumption data of water quality monitoring equipment, feeding module and water circulation module; The central control unit is used to receive, process and store data from the data acquisition module, analyze the breeding environment conditions according to the preset algorithm, and automatically adjust the feeding amount and water change parameters.

[0112] The embodiment of the present invention further provides a computer-readable storage medium storing instructions, which, when executed on a computer, causes the computer to execute the above-described method. All implementations in the above-described method embodiment are applicable to this embodiment and can achieve the same technical effects.

Claims

1. A data collection method for ecological cycle barrel-in-barrel farming, characterized in that: The method comprises: Step S1: Using a multi-source sensor array, a set of dynamic parameters of the aquaculture water body is acquired in real time, including pH value, dissolved oxygen concentration, ammonia nitrogen content, and temperature gradient data, and the timing operating parameters of the feeding machine and the fluid dynamics parameters of the water circulation system are simultaneously collected; Step S2: normalizing the aquaculture water body dynamic parameter set and fluid mechanics parameters to establish a timestamp-marked aquaculture environment state matrix; Step S3: extracting features from the historical data stream using a multivariate coupling analysis model, performing pattern matching with the aquaculture environment state matrix using a preset environmental threshold interval, and generating a water quality health assessment index and an ecological imbalance warning signal; Step S4: Analyzing the water quality health assessment index and ecological imbalance warning signal based on the fuzzy control algorithm, dynamically calculating the final feeding amount curve and water exchange intensity function, and generating a control instruction set including feeding frequency, bait particle size, and water pump speed; Step S5: Distribute the control instruction set to the execution terminal, drive the automatic feeding mechanism to perform the throwing operation, and at the same time adjust the working mode of the centrifugal pump group to achieve step-by-step water exchange, completing the closed-loop control of the aquaculture system.

2. The data collection method for an ecological cycle barrel-in-barrel farming method according to claim 1 is characterized in that: Step S3: Extract features from the historical data stream using a multivariate coupling analysis model, perform pattern matching with the aquaculture environment status matrix using the preset environmental threshold interval to generate a water quality health assessment index and an ecological imbalance warning signal, including: Step S31: Segment the historical data stream into sliding time windows, extract the time series statistical features and cross-variable lagged correlations of each sensor parameter, and generate a multi-dimensional feature vector set; Step S32: Based on the multidimensional feature vector set generated in step S31, the dynamic time warping algorithm is used to match the waveform similarity between the current aquaculture environment state matrix and the historical typical working conditions, and the composite weight coefficients of pH fluctuation entropy, dissolved oxygen decay slope, and ammonia nitrogen accumulation rate are calculated according to the matching results; Step S33: The composite weight coefficient output from step S32 is integrated with the normalized fluid mechanics parameters, and input into a pre-trained hidden Markov model to identify abnormal water cycle modes. At the same time, a convolution operation is performed on the real-time temperature gradient and the biological metabolic curve library to generate a thermodynamic matching coefficient. Step S34: Based on the abnormal water cycle mode and thermodynamic matching coefficient obtained in step S33, a two-layer fuzzy reasoning mechanism is adopted. The upper layer fuzzily maps the composite weight coefficient through the membership function of the environmental threshold interval to calculate the water quality health assessment index; the lower layer generates a differentiated level ecological imbalance warning signal including the ammonia nitrogen excessive pulse pattern recognition result and the dissolved oxygen critical oscillation frequency parameter according to the deviation degree of the thermodynamic matching coefficient and the duration of the parameter exceeding the limit, combined with the spatial gradient mutation characteristics of the abnormal water cycle mode.

3. The data collection method for an ecological cycle barrel-in-barrel farming method according to claim 2 is characterized in that: Step S31: Segment the historical data stream into sliding time windows, extract the time series statistical features and cross-variable lag correlation of each sensor parameter, and generate a multi-dimensional feature vector set, including: Based on the preset window length and sliding step size, the historical data stream is segmented and intercepted, and the mean, variance and skewness coefficient of pH value, dissolved oxygen concentration and ammonia nitrogen content are calculated in each window as time series statistical features; The cross-variable lagged correlation was analyzed by cross-correlation function, and the time-lagged mutual information entropy between dissolved oxygen and temperature gradient, and the phase offset between ammonia nitrogen content and feeding amount were extracted to form a multidimensional feature vector set containing statistical characteristics and correlation indicators.

4. The data collection method for an ecological cycle barrel-in-barrel farming method according to claim 3 is characterized in that: Step S32: Based on the multidimensional feature vector set generated in step S31, the dynamic time warping algorithm is used to match the waveform similarity between the current aquaculture environment state matrix and the historical typical working conditions. The composite weight coefficients of pH fluctuation entropy, dissolved oxygen decay slope, and ammonia nitrogen accumulation rate are calculated based on the matching results, including: Perform dynamic time warping alignment on the waveform segments in the multidimensional feature vector set generated in step S31 and the historical typical operating condition database, and calculate the waveform similarity score; According to the similarity score, a dynamic attenuation factor is assigned to the pH fluctuation entropy, a time attenuation compensation coefficient is superimposed on the dissolved oxygen attenuation slope, and a spatial weight correction term is applied to the ammonia nitrogen accumulation rate. A composite weight coefficient is generated through weighted fusion.

5. The data collection method for an ecological cycle barrel-in-barrel farming according to claim 4 is characterized in that: Step S33: The composite weight coefficient output from step S32 is integrated with the normalized fluid mechanics parameters, and input into the pre-trained hidden Markov model to identify abnormal water cycle modes. At the same time, the real-time temperature gradient is convolved with the biological metabolic curve library to generate a thermodynamic matching coefficient, including: The composite weight coefficient output from step S32 is subjected to feature concatenation with the normalized fluid dynamics parameters, and the resultant is input into a hidden Markov model to calculate the state transition probability of the water circulation system. The duration and intensity of the abnormal mode are identified through Viterbi decoding. The fluid dynamics parameters include flow velocity and turbulence intensity. The real-time temperature gradient data are subjected to a sliding convolution operation with the metabolic rate templates of different species in the biological metabolic curve library. The best matching curve is determined by the convolution peak position, and a normalized thermodynamic matching coefficient is generated based on the peak amplitude.

6. The data collection method for an ecological cycle barrel-in-barrel farming according to claim 5 is characterized in that: Step S34: Based on the water cycle abnormal mode and thermodynamic matching coefficient obtained in step S33, a two-layer fuzzy inference mechanism is used. The upper layer performs fuzzy mapping on the composite weight coefficient through the membership function of the environmental threshold interval to calculate the water quality health assessment index, including: The pH fluctuation entropy, dissolved oxygen decay slope, and ammonia nitrogen accumulation rate in the composite weight coefficient are mapped to the preset membership function to obtain the membership value; Activate the health reasoning rules in the fuzzy rule base according to the membership value to obtain the rule output result; The rule output results were defuzzified, and the centroid method was used to calculate the water quality health assessment index, where a lower index value indicates a higher risk of water quality deviating from the ecological threshold.

7. The data collection method for an ecological cycle barrel-in-barrel farming according to claim 6 is characterized in that: The lower layer generates differentiated ecological imbalance warning signals based on the degree of deviation of the thermodynamic matching coefficient and the duration of parameter exceeding the limit, combined with the spatial gradient mutation characteristics of the water cycle abnormal mode, including the recognition results of the ammonia nitrogen exceeding the limit pulse pattern and the dissolved oxygen critical oscillation frequency parameters, including: When the thermodynamic matching coefficient is lower than the preset threshold, the abnormal biological metabolism baseline is triggered. The time cumulative penalty factor is constructed by combining the duration of the parameter exceeding the limit, and the product operation is performed on the spatial gradient mutation amplitude of the water cycle abnormal mode to obtain the initial value of the warning level. Pulse pattern recognition is performed on the ammonia nitrogen content, and the type of excess is determined based on the pulse width and amplitude. At the same time, band-pass filtering and Fourier spectrum analysis are performed on the critical oscillation frequency of dissolved oxygen. Finally, the initial value and the identification result are integrated to generate differentiated levels of ecological imbalance warning signals.

8. The data collection method for an ecological cycle barrel-in-barrel farming according to claim 7 is characterized in that: Step S4: Analyze the water quality health assessment index and ecological imbalance warning signal based on the fuzzy control algorithm, dynamically calculate the final feeding amount curve and water exchange intensity function, and generate a control instruction set including feeding frequency, bait particle size, and water pump speed, including: According to the water quality health assessment index, the health level is divided into grades through Gaussian membership function, and the preset feeding amount attenuation curve is matched based on the grade; The ecological imbalance warning signal is input into the water exchange intensity function generator, and the water exchange flow rate increment ratio and duration are calculated according to the warning level; By integrating the preset feeding amount attenuation curve, water exchange flow rate increment ratio and duration, and combining the constraint relationship between bait particle size and water pump speed, a control instruction set including the quantitative value of feeding frequency, particle size classification mark and speed adjustment gradient is generated.

9. The data collection method for an ecological cycle barrel-in-barrel farming according to claim 8 is characterized in that: Step S5: Distribute the control instruction set to the execution terminal to drive the automatic feeding mechanism to perform the throwing operation, and at the same time adjust the working mode of the centrifugal pump group to achieve step-by-step water exchange, completing the closed-loop control of the aquaculture system, including: The feeding frequency quantization value in the control instruction set is converted into a pulse width modulation signal to drive the automatic feeding mechanism, and the movement trajectory of the robotic arm is adjusted according to the throwing angle and the bait particle size classification mark; The speed adjustment gradient is analyzed to control the centrifugal pump group to operate according to a step-by-step speed increase or decrease curve within a preset time interval. The error is compensated through the real-time feedback data of the flow sensor and the water exchange intensity function to form a closed-loop control circuit.

10. An ecological cycle barrel-in-barrel farming data acquisition system, characterized in that: The system is used to perform the method according to any one of claims 1 to 9, comprising: Breeding barrels, which are multiple ecological recycling round barrels, each barrel body is made of fiberglass, the bottom is funnel-shaped and equipped with a central sewage outlet, and water inlets are distributed around the barrel wall; The water circulation module uses a water pump to circulate and is connected to the aquaculture barrel structure through pipes to achieve water circulation; The sewage treatment module is connected to the water circulation module, and continuously extracts the feces and leftover bait in the aquaculture tank to the shore-based pollutant treatment pool. The treatment pool adopts a multi-stage inclined plate sedimentation area, and the supernatant after sedimentation enters the far corner of the facility area through the pipeline; Water quality monitoring equipment, including pH meters, dissolved oxygen meters, ammonia nitrogen detectors, and thermometers, for real-time monitoring of aquaculture water environment data; The feeding module is used to feed the aquaculture tanks in a timed and quantitative feeding mode using an automatic feeding machine according to the data of the aquaculture water environment; Fishing module, used to catch fish in the breeding tank; Data acquisition module, used to collect working status and energy consumption data of water quality monitoring equipment, feeding module and water circulation module; The central control unit is used to receive, process and store data from the data acquisition module, analyze the breeding environment conditions according to the preset algorithm, and automatically adjust the feeding amount and water change parameters.

Images