Growth rate prediction confidence driven regulation method and system for hippopotamus baboon breeding

Abstract

The present application relates to the technical field of intelligent cultivation of leeches, in particular to a growth rate prediction confidence driven regulation method and system for the cultivation of Limnatis filiformis. The method comprises collecting environmental monitoring data, cultivation operation data and growth sampling data, performing time alignment, abnormality elimination and effectiveness screening, identifying the growth stage of the cultivation object, and constructing a basic cultivation state representation. Further, the growth driving force index and the prediction confidence index are calculated, and the target regulation scheme is determined from the preset regulation candidate set according to the two indexes. The feeding, water changing, oxygen increasing, temperature control, pool division and manual review are regulated. The method can improve the accuracy and stability of the regulation of Limnatis filiformis cultivation.

Description

Technical Field

[0001] This invention relates to the field of intelligent leech farming technology, specifically to a growth rate prediction confidence-driven control method and system for Hirudo medicinalis farming. Background Technology

[0002] As an important species in medicinal leech farming, the growth of *Hirudo nipponia* is influenced by a combination of factors, including water temperature, dissolved oxygen levels, pH, stocking density, feeding rhythm, water change frequency, and stage-specific management practices. Particularly during the juvenile, young, and marketable stages, individuals exhibit significant differences in their sensitivity to environmental conditions, nutrient supply, and stocking density. Therefore, in actual farming, timely adjustments to feeding, water changes, oxygenation, temperature control, and pond separation based on different growth stages and site conditions directly impact the growth rate, survival rate, and farming stability of *Hirudo nipponia*.

[0003] Currently, the management of medicinal leeches, especially Hirudo nipponia, primarily relies on environmental monitoring, manual judgment, and experience-based adjustments. Specifically, on-site management typically involves monitoring environmental parameters such as water temperature, dissolved oxygen, and pH, combined with data on feeding, water changes, sampling and weighing results, feeding activity, and abnormal inspection records. Based on this information, farmers assess the current breeding status and decide whether to implement measures such as water changes, oxygenation, temperature control, residue removal, pond separation, or adjustments to feeding rhythm. While some existing technologies have incorporated sensor monitoring and automatic control devices, their primary implementation still largely involves comparing threshold values ​​for single environmental parameters and triggering corresponding actions. For example, activating oxygenation devices when dissolved oxygen falls below a preset threshold, or performing cooling or water changes when water temperature exceeds a preset threshold. While this approach can alleviate the problem of environmental parameters exceeding limits to some extent, it is essentially a passive adjustment mode based on fixed thresholds.

[0004] Furthermore, some existing technologies have begun to explore the use of multi-source aquaculture data for growth prediction or trend assessment. For example, they estimate growth trends over a specific period based on environmental parameters, feeding frequency, density levels, and sampling and weighing results. However, such approaches generally suffer from two limitations: first, they often only provide a single prediction result without further reflecting its reliability under current operating conditions; second, there is a lack of a stable and clear connection mechanism between the prediction result and actual control actions. Even when a growth trend assessment is formed, it usually still relies on human experience to determine whether to adopt it, to what extent to adopt it, and whether manual verification is necessary. Consequently, a true closed loop is not formed between prediction and control.

[0005] From a practical perspective, the existing technologies mentioned above have at least the following shortcomings. First, existing solutions typically do not fully consider the differences in optimal growth conditions between different growth stages, tending to treat juvenile, young, and marketable leech stages within a relatively uniform judgment framework, thus weakening the stage-specific growth response. In other words, while existing technologies can obtain multiple aquaculture parameters, their application often manifests as uniform threshold management or uniform empirical rule management, making it difficult to model and differentiate growth drivers in a targeted manner based on stage changes.

[0006] Secondly, existing technologies generally lack explicit evaluation mechanisms for data quality. Environmental monitoring data from aquaculture sites may suffer from sensor drift, short-term gaps, abnormal spikes, and time lags. Aquaculture operation data and growth sampling data may also be incomplete due to delays in manual entry, limited sampling frequency, or discontinuous inspections. In such cases, directly using raw data or data that has not been adequately screened for growth judgment and control decisions can easily lead to calculation results that deviate from the actual aquaculture conditions, resulting in misadjustments or overadjustments. While existing technologies may simply remove or supplement some abnormal data, they typically do not further address the crucial question of whether the current prediction results are reliable under the current data quality conditions.

[0007] Furthermore, even when existing technologies possess some predictive capability, they largely lack a technical approach that integrates both the prediction results and their reliability into control decisions. In actual aquaculture scenarios, growth predictions alone are insufficient to directly support high-intensity regulatory actions, as the same prediction may correspond to entirely different data bases and operational stability. For example, when the environment is relatively stable, data is complete, and recent prediction errors are small, the prediction results are suitable as a basis for proactive fine-tuning; however, when many key fields are missing, manual recording has a long lag, or recent prediction errors fluctuate significantly, even seemingly similar prediction results are not suitable for direct use in aggressive regulation. Existing technologies typically do not quantify this difference, resulting in whether regulation is possible, how much to regulate, and whether manual verification is necessary still largely relying on experience-based judgment.

[0008] Furthermore, existing technologies generally suffer from numerous single-variable adjustments and insufficient coordinated decision-making at the control output end. In aquaculture, feeding, water exchange, aeration, temperature control, and pond division are not independent processes; changes in one control variable often affect the reasonable range of values ​​for others. For example, increasing feeding frequency may require a corresponding increase in water exchange intensity and aeration margin; increasing stocking density constraints may necessitate simultaneously shortening feeding intervals or arranging pond division in advance. In existing technologies, many schemes still rely on single-parameter alarms triggering single-action execution, lacking a mechanism for constructing candidate control schemes around multiple control dimensions and conducting comprehensive evaluations. Therefore, it is difficult to achieve a optimal balance between growth promotion benefits and execution risks. Summary of the Invention

[0009] The purpose of this invention is to provide a growth rate prediction confidence-driven regulation method and system for Hirudo medicinalis farming, in order to solve the technical problems mentioned in the background art.

[0010] Based on the above ideas, the present invention provides the following technical solution:

[0011] Growth rate prediction confidence-driven regulation methods for Hirudo medicinalis farming include:

[0012] S1. Collect environmental monitoring data, aquaculture operation data and growth sampling data of each aquaculture unit. The environmental monitoring data shall include at least water temperature, dissolved oxygen and pH. The aquaculture operation data shall include at least feeding records, water change records and stocking density. The growth sampling data shall include at least sampling and weighing results, feeding activity records and abnormal inspection records.

[0013] S2. Perform time alignment, anomaly removal, and validity screening on the environmental monitoring data, aquaculture operation data, and growth sampling data, and identify the stage of the aquaculture object according to its juvenile leech stage, young leech stage, or commercial leech stage.

[0014] S3. Based on the stage identification results, construct the corresponding stage breeding state sequence to form the basic breeding state representation of each breeding unit under the current stage;

[0015] S4. Based on the target growth conditions at the current stage and combined with the basic aquaculture status characterization, calculate the growth driving force index characterizing the achievability of the current aquaculture unit's growth rate, and obtain the growth rate prediction result for the future target period accordingly.

[0016] S5. Based on the target growth conditions of the current stage, the basic aquaculture status characterization, and the data quality status, calculate the prediction confidence index corresponding to the growth rate prediction result.

[0017] S6. Based on the growth driving force index and the prediction confidence index, determine the target control scheme from the preset control candidate set, and control at least one of the following according to the target control scheme: feeding rhythm, water exchange intensity, oxygenation retention margin, temperature control target, pond separation timing, and manual verification triggering conditions.

[0018] Based on collected environmental monitoring data, aquaculture operation data, and growth sampling data, a basic aquaculture status characterization is first established. Then, growth rate prediction results and their corresponding prediction confidence indices are further output. This allows regulatory decisions to move beyond single threshold trigger levels and transform into trend-based regulation targeting future periods. It also enables earlier identification of potential growth deviation risks.

[0019] By incorporating prediction confidence into the control decision chain, the system can not only determine how future growth might change, but also whether the prediction is reliable enough to support actual control actions. This avoids the problems of misadjustment, overadjustment, and frequent intervention caused by directly translating low-reliability predictions into strong control actions, and is particularly suitable for the complex conditions in Hirudo nipponia farming where seasonal changes, feeding disturbances, pond switching, and data quality fluctuations coexist.

[0020] This approach differentiates between the juvenile, young, and marketable leech stages, ensuring that the regulatory logic aligns with the physiological characteristics and management priorities of each growth stage. Since the sensitivity of Hirudo nipponia to temperature, dissolved oxygen, density, and feeding rhythm varies at different stages, this method avoids the distortion caused by simply applying uniform empirical rules to different stages, thereby improving the targetedness and effectiveness of the system's regulation.

[0021] Preferably, S4 specifically includes:

[0022] The target water temperature center, allowable deviation of water temperature, target pH center, allowable deviation of pH, minimum dissolved oxygen requirement, baseline growth rate, and coupling strength between density and feeding rhythm are determined for the juvenile leech stage, the young leech stage, and the commercial leech stage, respectively.

[0023] The deviation between the current water temperature and the target water temperature center of the corresponding stage is used as a temperature suitability characterization; the deviation between the current pH and the target pH center of the corresponding stage, as well as the relationship between the current dissolved oxygen and the minimum dissolved oxygen requirement of the corresponding stage, are used as a physicochemical environment suitability characterization; and the combination relationship between the current stocking density and the current feeding rhythm is used as a crowding feeding coupling characterization.

[0024] The growth driving force index is calculated based on the temperature suitability characterization, the physicochemical environment suitability characterization, and the crowding-feeding coupling characterization.

[0025] The originally discrete and scattered multi-source aquaculture factors such as temperature, pH, dissolved oxygen, density, and feeding rhythm are reorganized into a multi-dimensional characterization system with clear physiological significance according to the target working conditions of different growth stages.

[0026] Specifically, by separately determining the target water temperature center, allowable water temperature deviation, target pH center, allowable pH deviation, minimum dissolved oxygen requirement, baseline growth rate, and the coupling strength between density and feeding rhythm for each stage, the system can avoid mechanically processing raw monitoring values ​​during calculations. Instead, it first maps the raw data into suitability and coupling representations consistent with the stage requirements. In this way, environmental constraints and aquaculture pressures at different stages are explicitly introduced into the same technical framework, which is beneficial to improving the scenario adaptability of growth judgment.

[0027] Furthermore, by simultaneously considering temperature suitability, physicochemical environmental suitability, and the coupling relationship between crowding and feeding, the growth driving force indicators can better reflect the true growth status of Hirudo medicinalis under the combined influence of multiple factors. Especially when there are frequent changes in stocking density and feeding rhythm adjustments, this multidimensional characterization method can better avoid the one-sidedness caused by judging a single parameter.

[0028] Preferably, the growth driving force index is calculated in the following manner:

[0029] The baseline growth rate corresponding to the current growth stage is used as the base value;

[0030] The baseline value is attenuated and corrected based on the degree of deviation of the current water temperature from the target water temperature center of the current growth stage and the degree of deviation of the current pH from the target pH center of the current growth stage. The greater the deviation of the current water temperature or current pH from the target center value of the corresponding stage, the greater the attenuation of the baseline value.

[0031] Based on the relationship between the current dissolved oxygen level and the minimum dissolved oxygen requirement at the current growth stage, the oxygen supply response is corrected after attenuation correction. The closer the current dissolved oxygen level is to or higher than the oxygen requirement level at the current growth stage, the stronger the oxygen supply response.

[0032] Furthermore, crowding-feeding constraint correction is performed by combining the coupling relationship between the current stocking density and the current feeding interval. The higher the current stocking density and the longer the current feeding interval, the stronger the inhibitory effect on the result.

[0033] The final result is a growth driving force index used to characterize the achievability of the current growth rate of the breeding unit.

[0034] By employing a progressive approach that incorporates baseline growth capacity, environmental deviation attenuation, oxygen supply response correction, and crowding-feeding constraint correction, growth driver indicators are made more physiologically reasonable and physically interpretable.

[0035] Using the baseline growth rate corresponding to the current growth stage as the basic value reflects the objective fact that the growth capacity of Hirudo nipponia varies at different stages. This ensures that the driving force calculation does not start from a uniform initial value, but is based on the baseline growth characteristics of each stage. This helps improve the comparability and rationality of the judgment results between different growth stages.

[0036] By attenuating and correcting for deviations in water temperature and pH, the actual aquaculture pattern of inhibited growth as the environment deviates further from the suitable zone can be directly reflected in the indicator formation process, thus avoiding the distortion caused by simply relying on linear superposition of environmental factors. Furthermore, by correcting for dissolved oxygen response and incorporating oxygen supply capacity into the evaluation system, the direct impact of water physicochemical conditions on the growth activity and survival status of Hirudo nipponiae can be further reflected.

[0037] Introducing a coupling constraint between stocking density and feeding interval exerts a stronger inhibitory effect on high-density conditions with long feeding intervals, which helps to accurately reflect the objective law of declining growth potential in Hirudo medicinalis farming when density is too high and feed intake is insufficient. The resulting growth driving force index is not a simple empirical score, but a composite index that can comprehensively characterize the degree of environmental suitability, oxygen supply support, and farming stress level.

[0038] Preferably, S5 specifically includes:

[0039] Extract the model prediction residuals used to characterize prediction stability, the sensor drift used to characterize sensor reliability, the recording delay used to characterize the timeliness of manual recording, and the missing proportion used to characterize data integrity;

[0040] The model prediction residual, the sensor drift, the recording delay, and the missing proportion are normalized to form a data uncertainty load.

[0041] The prediction confidence index is calculated by combining the current temperature suitability and physicochemical environmental suitability. The greater the degree of environmental deviation and the higher the data uncertainty load, the lower the prediction confidence index.

[0042] By systematically incorporating data quality factors such as model prediction residuals, sensor drift, manual recording lag, and the proportion of missing data into aquaculture control decisions, the system can be equipped with the ability to identify bad data, late data, and low-reliability data.

[0043] In existing aquaculture management scenarios, sensors may drift over long periods of operation, manual recording may be delayed, and inspection and weighing data may be incomplete or discontinuous. If this data is used directly for growth judgment and control decisions without quality assessment, the system is highly likely to overreact based on distorted data. By constructing a data uncertainty load, these various data quality problems are uniformly expressed and quantified, thereby imposing prior constraints on the reliability of the data source at the algorithm level.

[0044] Furthermore, the reliability of predictions is not judged solely based on data quality, but rather by combining the data uncertainty load with the suitability of the current temperature and physicochemical environment. The benefit of this approach is that when the environment has significantly deviated from its suitable range and the data quality is poor, the system can more sensitively reduce the prediction confidence; conversely, when the environment is relatively stable and the data quality is good, a high level of reliability can be maintained. This resulting judgment mechanism better aligns with the comprehensive reliability requirements of actual control scenarios.

[0045] Preferably, the prediction confidence index is calculated in the following manner:

[0046] The basic environmental suitability value is determined based on the degree of deviation of the current water temperature from the target water temperature center of the current growth stage and the degree of deviation of the current pH from the target pH center of the current growth stage. The closer the current water temperature and current pH are to the target center value of the corresponding stage, the higher the basic environmental suitability value.

[0047] Based on the relationship between the current dissolved oxygen level and the minimum dissolved oxygen requirement at the current growth stage, the oxygen supply reliability of the aforementioned environmental suitability baseline value is corrected. The higher the current dissolved oxygen level can meet the oxygen requirement of the corresponding stage, the higher the oxygen supply reliability.

[0048] The model prediction residual, sensor drift, recording delay, and data missing ratio are then converted into data uncertainty load. The larger the model prediction residual, the more obvious the sensor drift, the longer the recording delay, and the higher the data missing ratio, the higher the data uncertainty load.

[0049] Using the baseline environmental suitability value and the oxygen supply reliability correction result, the data uncertainty load is subjected to inverse constraints to obtain the prediction confidence index, so that the more suitable the environmental state and the lower the data uncertainty, the higher the prediction confidence index.

[0050] By organically combining the suitability of environmental conditions with the quality of data, the prediction confidence index is no longer an abstract probability result, but a comprehensive technical quantity that can directly reflect whether the current prediction is trustworthy.

[0051] Specifically, the baseline environmental suitability is first determined based on the deviation of water temperature and pH from the target center values ​​for each stage. Then, the oxygen supply reliability is corrected based on the current dissolved oxygen level's satisfaction of the stage's oxygen requirements. Thus, the system can determine from an environmental perspective whether the current operating conditions are within a suitable growth window. This process establishes a clear link between the prediction confidence level and the key environmental factors upon which the actual survival and growth of *Hirudo nipponia* depend, preventing the confidence level from becoming a purely mathematical quantity detached from the farming scenario.

[0052] Building upon this, the model prediction residuals, sensor drift, recording lag, and data missing ratio are collectively transformed into a data uncertainty load, which is then constrained by environmental suitability and oxygen supply reliability. The resulting prediction confidence index reflects information on both environmental stability and data reliability, providing more comprehensive support for subsequent regulatory decisions.

[0053] Preferably, S6 specifically includes:

[0054] A set of regulatory candidates is constructed according to the current growth stage. The set of regulatory candidates includes multiple candidate regulatory schemes. Each candidate regulatory scheme includes at least the target feeding interval, the target water exchange ratio, the target oxygenation retention margin, the target temperature control setpoint, and the target stocking density upper limit.

[0055] When the prediction confidence index reaches the high confidence interval, an active fine-tuning plan is allowed to be implemented to adjust the feeding rhythm, water exchange intensity, oxygenation retention margin and temperature control target in a coordinated manner.

[0056] When the prediction confidence index is in the middle confidence interval, only the limited adjustment scheme is allowed to be implemented, and the feeding rhythm and environmental control parameters are corrected in small steps.

[0057] When the prediction confidence index is in the low confidence interval, a conservative maintenance plan is implemented, and at least one of the following manual verification actions is triggered: manual sampling and weighing, close-range inspection, or pre-pooling verification.

[0058] By constructing a set of control candidates that includes target feeding intervals, target water exchange ratios, target oxygenation margins, target temperature setpoints, and target stocking density limits, the system's adjustment actions are no longer single-variable, single-step adjustments, but rather can form a complete set of control schemes around multiple key control parameters. This helps improve the synergy between control actions and avoids causing new imbalances by adjusting only one parameter.

[0059] The intensity of regulation is stratified into high-confidence, medium-confidence, and low-confidence intervals. Specifically, in the high-confidence state, proactive fine-tuning is permitted, which helps to fully utilize relatively reliable forecasts to optimize aquaculture conditions in a timely manner; in the medium-confidence state, only limited adjustments are implemented, which helps to maintain a certain degree of regulatory flexibility while reducing the risks associated with aggressive control; in the low-confidence state, a conservative maintenance plan is implemented and manual review is triggered, which helps to prevent the system from taking strong actions based on unreliable forecasts, thereby reducing misadjustments, overadjustments, and unnecessary interventions.

[0060] Preferably, the target control scheme is determined in the following manner:

[0061] For the current growth stage, candidate regulatory schemes are retrieved one by one from the corresponding regulatory candidate set;

[0062] For each candidate regulation scheme, the degree to which the candidate regulation scheme promotes the growth rate improvement in the future target period is evaluated in conjunction with the growth driving force index, and the feasibility and reliability of the candidate regulation scheme are evaluated in conjunction with the prediction confidence index.

[0063] When evaluating the degree of promotion, the matching degree between the target temperature control setpoint corresponding to the candidate control scheme and the target water temperature center at the current growth stage, the degree of improvement of the dissolved oxygen guarantee capacity by the target oxygenation retention margin corresponding to the candidate control scheme, and the degree of coupling constraint between the target stocking density upper limit and the target feeding interval corresponding to the candidate control scheme are also considered.

[0064] When assessing the feasibility and reliability of the proposed scheme, the deviation of the target water exchange ratio corresponding to the candidate control scheme from the conservative water exchange ratio at the current growth stage is also considered, as well as the deviation of the upper limit of the target stocking density corresponding to the candidate control scheme from the target stocking density at the current growth stage.

[0065] Considering the combined benefits of growth promotion and the execution risks associated with deviating from conservative operating conditions, the candidate control scheme with the best overall evaluation is determined as the target control scheme.

[0066] By comprehensively considering the benefits of growth promotion and the execution risks brought about by deviation from conservative operating conditions, the system is able to select the scheme with better overall performance from multiple candidate control schemes, rather than simply ranking them based on a single indicator.

[0067] On the one hand, when evaluating candidate solutions, the matching degree between the temperature control target and the stage target water temperature, the degree to which the oxygenation retention margin improves the dissolved oxygen guarantee capacity, and the degree of coupling constraint between the upper limit of the target stocking density and the target feeding interval are considered simultaneously. This can more realistically reflect the potential of candidate solutions to promote future growth rate improvement, so that the growth promotion benefits are no longer at the abstract level, but correspond one-to-one with the executable controllable quantities.

[0068] On the other hand, it requires that when assessing the feasibility and reliability, the degree to which the target water exchange ratio deviates from the conservative water exchange ratio, and the degree to which the upper limit of the target stocking density deviates from the phased target stocking density, should be considered simultaneously. The beneficial effect of doing so is that it can incorporate the practical problem that overly aggressive control measures, while potentially bringing higher short-term benefits, also come with higher implementation risks, into a unified evaluation framework.

[0069] A growth rate prediction confidence-driven control system for Hirudo nipponia farming, used to perform the method, comprising:

[0070] An environmental sensing unit is used to collect environmental monitoring data from each aquaculture unit. The environmental monitoring data includes at least water temperature, dissolved oxygen, and pH.

[0071] The breeding record unit is used to acquire breeding operation data and growth sampling data of each breeding unit. The breeding operation data includes at least feeding records, water change records and stocking density. The growth sampling data includes at least sampling and weighing results, feeding activity records and abnormal inspection records.

[0072] The execution control unit is used to perform at least one of the following control actions on each aquaculture unit: feeding regulation, water exchange regulation, oxygenation regulation, temperature control regulation, pond division regulation, and manual verification triggering;

[0073] The control unit is communicatively connected to the environmental sensing unit, the aquaculture recording unit, and the execution control unit, respectively. The control unit includes:

[0074] The data preprocessing module is used to perform time alignment, anomaly removal, and validity screening on the environmental monitoring data, aquaculture operation data, and growth sampling data;

[0075] The stage identification module is used to identify the stage of the leech according to whether the leech is in the juvenile leech stage, the young leech stage, or the commercial leech stage, and to construct the corresponding breeding status sequence for the stage.

[0076] The growth driving force calculation module is used to calculate the growth driving force index that represents the achievability of the current growth rate of the current breeding unit based on the target growth conditions and basic breeding status of the current stage, and to obtain the growth rate prediction results for the future target period.

[0077] The prediction confidence assessment module is used to calculate the prediction confidence index corresponding to the growth rate prediction result based on the target growth conditions, basic aquaculture status, and data quality status at the current stage.

[0078] The regulation scheme determination module is used to determine the target regulation scheme from a preset regulation candidate set based on the growth driving force index and the prediction confidence index.

[0079] The execution control module is used to control the execution control unit according to the target control scheme, and to regulate at least one of the following: feeding rhythm, water exchange intensity, oxygenation retention, temperature control target, pond separation timing, and manual verification trigger conditions.

[0080] By organically integrating environmental perception, aquaculture records, control decisions, and execution regulation into a complete closed-loop system combining hardware and software, the aforementioned methods have a clear implementation platform and engineering path.

[0081] First, the environmental sensing unit and the aquaculture recording unit together form the foundation for multi-source data input, enabling the simultaneous acquisition of automatic monitoring data and manual or semi-automatic recording data, thereby improving the system's comprehensive coverage of information at the aquaculture site. The execution and control unit directly addresses actual aquaculture actions such as feeding, water changing, oxygenation, temperature control, pond separation, and manual verification, ensuring that control results can be promptly fed back to the on-site production process.

[0082] Secondly, the data preprocessing module, stage identification module, growth driving force calculation module, prediction confidence assessment module, control scheme determination module, and execution control module within the control unit correspond one-to-one with the key steps in the method. This not only facilitates a clear division of system functions but also ensures that the entire process from data acquisition to control output has a clearly defined technical implementation entity, enhancing the completeness and feasibility of the solution.

[0083] This system is not merely for environmental monitoring or simple automatic control; rather, it achieves a complete closed loop through the collaborative function of internal modules, encompassing phased growth assessment, predictive reliability evaluation, and tiered control execution. This system can more accurately respond to the multi-stage differences, multi-factor coupling, and data quality fluctuations in Hirudo medicinalis farming, offering benefits such as improved control accuracy, enhanced system robustness, and reduced reliance on human experience.

[0084] The technical solution of the present invention may include the following beneficial effects:

[0085] This invention addresses the long-standing reliance on manual experience, fixed threshold control, and periodic sampling in leech farming. It constructs a closed-loop technical solution that integrates environmental monitoring, farming records, growth assessment, reliability evaluation, and regulatory execution. By simultaneously collecting environmental monitoring data, farming operation data, and growth sampling data, and considering the differentiated needs of juvenile, young, and marketable leech stages, a more accurate baseline characterization of each farming unit's condition is established. This allows for the output of growth rate predictions for future target time periods. This transforms the traditional management approach, primarily based on post-hoc observation and manual correction, into a proactive regulatory approach focused on future growth trends. This improves the accuracy, timeliness, and continuity of growth assessment, thereby enhancing the level of refined leech farming management.

[0086] This invention not only predicts growth rate but also introduces a prediction confidence index corresponding to the prediction results, using it as a crucial basis for subsequent control intensity allocation and manual verification triggering. In this way, when facing complex conditions such as sensor drift, lag in manual recording, data loss, seasonal fluctuations, feeding disturbances, and pond switching, the system no longer directly implements strong actions based solely on a single predicted value. Instead, it can execute different levels of control schemes, such as proactive fine-tuning, limited adjustment, or conservative maintenance, depending on the reliability of the prediction results. This effectively reduces the probability of misadjustment, overadjustment, and frequent intervention, improves the stability and fault tolerance of the control process, and reduces reliance on long-term on-site judgment by experienced aquaculture personnel, demonstrating strong engineering practical value.

[0087] In this invention, stage identification, basic state characterization, growth driving force calculation, prediction confidence assessment, and target control scheme determination are not independent or simply set up in parallel. Instead, they establish a linked and mutually restrictive linkage mechanism based on the real breeding constraints of Hirudo medicinalis at different growth stages: front-end environmental conditions, density status, and feeding rhythm jointly affect the assessment of growth potential; data quality status further affects the reliability of prediction results; and the assessment of growth potential and the reliability of predictions jointly determine the selection range and execution intensity of back-end control schemes. Because the above links form a continuous coupling relationship in the same technical chain, the results obtained by this invention are not simply environmental monitoring results, simple prediction results, or simple control results, but rather a holistic optimized control result for Hirudo medicinalis farming scenarios. This can ensure the safety of control while taking into account the growth promotion effect, thereby achieving a comprehensive technical effect that better meets actual production needs. Attached Figure Description

[0088] Figure 1 This is the main flowchart of the confidence-driven regulation method for predicting growth rate in Hirudo medicinalis farming according to the present invention.

[0089] Figure 2 This is a system block diagram of the growth rate prediction confidence-driven control system for Hirudo nipponia farming according to the present invention.

[0090] Figure 3 This is a scatter plot showing the water temperature and observed specific growth rate of Hirudo nipponia at different growth stages in the example.

[0091] Figure 4 The graph shows the relationship between water temperature and growth driving force indices under different dissolved oxygen conditions at a fixed stage in the example.

[0092] Figure 5 The diagram shows the relationship between prediction confidence and uncertainty load in the example.

[0093] Figure 6 The image shows a multi-parameter radar chart of candidate control schemes in the embodiment.

[0094] Figure 7 This is a graph showing the comprehensive evaluation results of candidate control schemes in the examples. Detailed Implementation

[0095] Example 1

[0096] In this embodiment, taking the refined farming scenario of medicinal leeches as an example, the implementation of the "growth rate prediction confidence-driven control method and system for leech farming" is described. The farmed animals are distributed across multiple farming units, each with an independent number. The farming unit can be a farming pond, hatching container, or seedling container. For ease of description, it will be referred to as a farming unit below. The control objective is: during the continuous operation of each farming unit, based on environmental monitoring data, farming operation data, and growth sampling data, to identify its current growth stage, calculate the growth driving force index and prediction confidence index for the current stage, and further determine the target control scheme applicable to that farming unit, which will be applied to feeding, water exchange, oxygenation, temperature control, pond separation, and manual verification.

[0097] In this embodiment, environmental monitoring data includes at least water temperature, dissolved oxygen, and pH. Aquaculture operation data includes at least feeding records, water change records, and stocking density. Growth sampling data includes at least sampling and weighing results, feeding activity records, and abnormal inspection records. All of the above data are timestamped and indexed by the aquaculture unit number.

[0098] In one specific embodiment, water temperature is collected by a water temperature sensor, with a sampling period of 1 minute and a resolution of 0.1℃; dissolved oxygen is collected by a dissolved oxygen sensor, with a sampling period of 1 minute and a resolution of 0.01 mg / L; pH is collected by a pH sensor, with a sampling period of 5 minutes and a resolution of 0.01; feeding records, water change records, stocking density, sampling and weighing results, feeding activity records, and abnormal inspection records are recorded using event-triggered recording, recording the occurrence time and corresponding content. To balance online operating efficiency and the characteristics of field data fluctuations, in this embodiment, the control unit uses 30 minutes as a uniform operating time slice, and performs a complete calculation process on all aquaculture units every 30 minutes.

[0099] like Figure 1 A confidence-driven regulation method for predicting growth rate in Hirudo medicinalis farming includes:

[0100] S1. Collect environmental monitoring data, aquaculture operation data and growth sampling data of each aquaculture unit. The environmental monitoring data shall include at least water temperature, dissolved oxygen and pH. The aquaculture operation data shall include at least feeding records, water change records and stocking density. The growth sampling data shall include at least sampling and weighing results, feeding activity records and abnormal inspection records.

[0101] S2. Perform time alignment, anomaly removal, and validity screening on the environmental monitoring data, aquaculture operation data, and growth sampling data, and identify the stage of the aquaculture object according to its juvenile leech stage, young leech stage, or commercial leech stage.

[0102] S3. Based on the stage identification results, construct the corresponding stage breeding state sequence to form the basic breeding state representation of each breeding unit under the current stage;

[0103] The aquaculture state sequence refers to the time-series set of environmental monitoring data, aquaculture operation data, and growth sampling data arranged in a uniform time slice order at the current stage; the control unit can extract the basic aquaculture state representation corresponding to the current time slice from the aquaculture state sequence.

[0104] S4. Based on the target growth conditions at the current stage and combined with the basic aquaculture status characterization, calculate the growth driving force index characterizing the achievability of the current aquaculture unit's growth rate, and obtain the growth rate prediction result for the future target period accordingly.

[0105] S5. Based on the target growth conditions of the current stage, the basic aquaculture status characterization, and the data quality status, calculate the prediction confidence index corresponding to the growth rate prediction result.

[0106] S6. Based on the growth driving force index and the prediction confidence index, determine the target control scheme from the preset control candidate set, and control at least one of the following according to the target control scheme: feeding rhythm, water exchange intensity, oxygenation retention margin, temperature control target, pond separation timing, and manual verification triggering conditions.

[0107] In this embodiment, online operation is based on time slices. At each new time slice, the control unit first reads the continuous sampling data within that time slice from the environmental sensing side, and then reads the most recent valid event record up to the current time slice from the aquaculture record side. For continuously sampled variables, it is preferable to calculate the mean, maximum, minimum, and standard deviation within the current 30-minute time slice, and use the mean as the representative value for the current time slice. For event variables, if there are no new records in the current time slice, the most recent valid record is used, and the time lag between that record and the end of the current time slice is calculated.

[0108] In this embodiment, water temperature, dissolved oxygen, and pH are used as environmental monitoring data inputs; feeding records are converted into the current feeding interval, i.e., the time difference between the current time and the time when the most recent feeding was completed; water change records are converted into the most recent water change ratio and its occurrence time; stocking density is expressed as the last digit per unit area under the current statistical caliber; sampling and weighing results are converted into the average body mass of the most recent sample; feeding activity records are represented by graded numerical values; and abnormal inspection records are represented by abnormality category codes or binary abnormality identifiers. The purpose of this processing is to ensure that all input quantities have a unified time reference and clear numerical meaning before entering subsequent calculations.

[0109] In this embodiment, the control unit performs time alignment, anomaly removal, and validity screening on the input data. Anomaly removal employs a two-level rule. The first category is physical boundary screening, for example: if the water temperature is less than 0℃ or greater than 45℃, dissolved oxygen is less than 0 mg / L or greater than 25 mg / L, or pH is less than 3 or greater than 11, the sampling point is directly determined to be invalid. The second category is sliding window anomaly identification, using the data from the most recent 24 hours as a window, and employing outlier detection based on the absolute deviation of the median to identify spike anomalies. Preferably, points deviating from the median by more than 3 times the absolute deviation of the median are marked as anomalies. For abnormal or missing data, if the length of consecutive anomalies or missing data does not exceed two consecutive sampling periods, linear interpolation is used for recovery; if the length of consecutive anomalies or missing data is greater than two consecutive sampling periods but not more than 12 consecutive sampling periods, the local mean of the same time period in the most recent 24 hours or the median of the same time period of the previous day is used for filling; if consecutive missing data exceeds 12 consecutive sampling periods, numerical filling is no longer performed, but the variable is recorded as a high-uncertainty input in the current time window and reflected in subsequent prediction confidence assessments.

[0110] In the stage identification stage, this embodiment preferably adopts a determination method that prioritizes the average body weight of the most recent sampling and supplements it with the number of rearing days. The stage boundary is not set directly based on experience, but is determined by historical sample statistics. Specifically, the average body weight, number of rearing days, and observed growth rate between adjacent samplings are first collected for at least one complete rearing cycle. Then, the samples are clustered or segmented according to the combination of body weight and number of rearing days, and the boundary where the distribution of observed growth rate changes significantly between two adjacent segments is taken as the stage threshold. The stage boundary obtained in this way is written into the parameter library for online stage identification. During online operation, if the current average body weight falls below the first stage threshold, it is identified as the juvenile leech stage; if it is between the first stage threshold and the second stage threshold, it is identified as the young leech stage; if it is above the second stage threshold, it is identified as the commercial leech stage. To avoid frequent switching near the stage boundary, a hysteresis judgment rule is preferably set: when the current body weight has just crossed the stage threshold but has not been maintained for two consecutive sampling cycles, the previous stage determination is maintained, and the prediction confidence level of the current time window is lowered.

[0111] After completing stage identification, the control unit constructs a basic aquaculture status representation for the current stage. This representation includes at least the current stage label, the average water temperature for the current time slice, the average dissolved oxygen for the current time slice, the average pH for the current time slice, the current stocking density, the current feeding interval, the proportion of the most recent water change, the average body weight of the most recent sample, the most recent feeding activity record, the most recent abnormal inspection status, the proportion of missing key data, and the time lag of the most recent key manual records. This basic status representation also serves the subsequent calculation of growth driving force indicators and prediction confidence indicators, thereby ensuring the consistency of the calculation chain on the input side.

[0112] In this embodiment, the online operation outputs three types of results: the first is the predicted growth rate for the future target period; the second is the corresponding prediction confidence index; and the third is the target control scheme. The control unit then maps the target control scheme into actual executable control actions or manual review prompts. In this way, the output results no longer remain at the pure calculation layer but directly enter the application end.

[0113] Specifically, S4 includes:

[0114] The target water temperature center, allowable deviation of water temperature, target pH center, allowable deviation of pH, minimum dissolved oxygen requirement, baseline growth rate, and coupling strength between density and feeding rhythm are determined for the juvenile leech stage, the young leech stage, and the commercial leech stage, respectively.

[0115] The deviation between the current water temperature and the target water temperature center of the corresponding stage is used as a temperature suitability characterization; the deviation between the current pH and the target pH center of the corresponding stage, as well as the relationship between the current dissolved oxygen and the minimum dissolved oxygen requirement of the corresponding stage, are used as a physicochemical environment suitability characterization; and the combination relationship between the current stocking density and the current feeding rhythm is used as a crowding feeding coupling characterization.

[0116] The growth driving force index is calculated based on the temperature suitability characterization, the physicochemical environment suitability characterization, and the crowding-feeding coupling characterization.

[0117] In this embodiment, the S4 refinement relies on a phased parameter library. This phased parameter library stores the following parameters for the juvenile leech stage, the young leech stage, and the commercial leech stage: the target water temperature center for the stage, the allowable deviation of the water temperature for the stage, the target pH center for the stage, the allowable deviation of the pH for the stage, the minimum dissolved oxygen requirement for the stage, the baseline growth rate for the stage, and the coupling strength between the stage density and the feeding rhythm.

[0118] These parameters are not arbitrarily set manually, but are obtained through calibration using historical samples. The historical sample set is constructed as follows: based on data from at least one complete breeding cycle, the basic state characteristics of each time slice are used as input samples, and the actual observed growth rate in the next 7 days after that time slice is used as the monitoring value. The observed growth rate is preferably calculated from the average body mass of two consecutive samples, for example, by determining the relative growth level of the corresponding time window based on the average body mass of the initial sample and the average body mass of the sample 7 days later. The reason for using the next 7 days as the target period is that this duration matches the sampling frequency at the breeding site, balancing the phased trend with data availability.

[0119] The target future period can be the next 3 days, the next 7 days, or the next 14 days, preferably the next 7 days.

[0120] In this embodiment, to avoid low-quality samples misleading the parameter database, a sample weight is assigned to each sample. The sample weight is determined by data integrity, recording timeliness, and sensor stability. Specifically, the key field completeness rate, key manual recording lag score, and sensor drift score within the corresponding time window of the sample are calculated first, and then the three are normalized to a unified scale. Preferably, the higher the completeness rate, the higher the weight; the shorter the lag, the higher the weight; and the smaller the drift, the higher the weight. In one specific implementation, the three normalized scores are multiplied to obtain the total weight; in another optional implementation, they can be weighted and summed according to a preset ratio before normalization. If a weighted summation method is used, the weight ratio of the three items can be determined by a validation set search, minimizing the prediction error obtained from subsequent parameter fitting. Regardless of the integration method used, the goal is to maximize the contribution of high-quality samples to the stage parameters.

[0121] Within each stage, samples are first sorted from highest to lowest based on observed growth rate, and the top 30% of samples are selected as the high-growth sample set. The target water temperature center for the stage is the weighted average of the water temperatures in this set; the allowable deviation of the water temperature for the stage is the weighted standard deviation of the water temperatures in this set, multiplied by a preset coverage coefficient, preferably 1.5; the target pH center for the stage and the allowable pH deviation for the stage are obtained in the same way; the minimum dissolved oxygen requirement for the stage is the weighted lower quartile of dissolved oxygen in the high-growth sample set; the baseline growth rate for the stage is the weighted average of the observed growth rates in the high-growth sample set. Once these parameters are determined, they are permanently written into a parameter library for direct use during online operation.

[0122] During online operation, the control unit reads the corresponding entries from the parameter library according to the current stage and forms three intermediate representations: temperature suitability representation, physicochemical environment suitability representation, and crowding-feeding coupling representation. Among them, the temperature suitability representation describes the degree of deviation of the current water temperature from the target water temperature center of the stage; the physicochemical environment suitability representation describes the degree of deviation of the current pH and the current dissolved oxygen saturation; and the crowding-feeding coupling representation describes the aquaculture pressure under the combined effect of stocking density and feeding interval.

[0123] like Figure 3 A scatter plot showing the relationship between water temperature and observed specific growth rate for Hirudo nipponia at different growth stages.

[0124] In this embodiment, historical samples are first divided into juvenile leech stage samples, young leech stage samples, and commercial leech stage samples according to the stage identification results. Then, the observed specific growth rate distribution of samples in each stage under different water temperature conditions is statistically analyzed. Figure 3 As can be seen, samples from different growth stages exhibit distinct clustering regions within the water temperature-observed growth rate plane. The juvenile leech stage samples generally show a higher observed growth rate, followed by the juvenile leech stage samples, while the commercial leech stage samples generally show a lower rate. Furthermore, high-growth samples at each stage are concentrated within a relatively limited water temperature range. Based on this scatter distribution, high-growth samples at each stage can be further screened, and the target water temperature center, allowable water temperature deviation, and baseline growth rate for each stage can be statistically determined.

[0125] In one specific implementation, the samples whose observed specific growth rate is in the top 30% of the samples in each stage can be regarded as the high growth sample set. The weighted average of the water temperature corresponding to the set is taken as the target water temperature center of the stage. The weighted standard deviation of the water temperature corresponding to the set is multiplied by a preset coverage coefficient as the allowable deviation of the water temperature in the stage. The weighted average of the observed specific growth rate corresponding to the set is taken as the benchmark growth rate of the stage. Figure 3 The scatter distribution shown reflects the differences in the suitable temperature range of samples at different stages, and also provides a sample basis for determining the parameters of the stage target water temperature center, the stage water temperature allowable deviation, and the stage benchmark growth rate in the subsequent growth driving force index formula.

[0126] Specifically, the growth driving force index X is calculated according to the following formula:

[0127]

[0128] Where k represents the current growth stage; This represents the baseline growth rate at the k-th growth stage; T represents the current water temperature. ΔT represents the target water temperature center for the k-th growth stage; kThis indicates the permissible deviation of water temperature for the k-th growth stage; pH indicates the current acidity or alkalinity. ΔpH represents the target pH center for the k-th growth stage. k Indicates the allowable pH deviation for the k-th growth stage; O represents the current dissolved oxygen level; K O,k τ represents the oxygen response constant at the k-th growth stage; ρ represents the current stocking density; f Indicates the current feeding interval; η k This represents the density-feeding coupling coefficient at the k-th growth stage.

[0129] In this embodiment, the oxygen response constant K O,k and coupling coefficient η k It was obtained through weighted nonlinear least squares fitting. Specifically, the stage parameter library was determined. ΔT k , and ΔpH k Next, using the observed growth rate of historical samples as the target value and X calculated by the formula as the fitted value, a weighted squared error objective is constructed. The weights are the sample weights from the previous embodiment. During fitting, K is adjusted... O,k and η k Set reasonable boundaries, such as K O,k Between 0.5 mg / L and 10 mg / L, η k Between 0.0001 and 0.01m 2 Between / (tail·d); if the sample size is small in a certain stage, it is preferable to use the fitting results of the previous and next stages as initial values ​​for iteration. The preferred iteration termination condition is that the improvement in fitting error between two adjacent iterations is less than 10. -6 It may reach 200 iterations. After fitting, K at each stage will be... O,k and η k The data is then written into the parameter library.

[0130] In a set of online calculation examples, assuming the current breeding unit is identified as the juvenile leech stage, its parameter library entry is: stage baseline growth rate 0.024d. -1 The target water temperature center is 28.0℃, with an allowable temperature deviation of 2.0℃; the target pH center is 7.40, with an allowable pH deviation of 0.40; the oxygen response constant is 2.0 mg / L; and the coupling coefficient is 0.002m. 2 / (tails·d). Current online status: water temperature 28.1℃, pH 7.38, dissolved oxygen 8.0mg / L, stocking density 140 tails / m³. 2 Feeding interval is 1.5 days. Following the calculation order, first calculate the water temperature deviation term, i.e., ((28.1−28.0) / 2.0). 2The result is 0.0025; then the pH deviation term is calculated, i.e., ((7.38−7.40) / 0.40). 2 The result was 0.0025; adding the two together and taking the negative exponent, the environmental degradation term was approximately 0.995. The dissolved oxygen response term was then calculated as 8.0 / (2.0+8.0), resulting in 0.8. The crowding-feeding constraint term was then calculated as 1 / (1+0.002×140×1.5), resulting in approximately 0.704. Finally, multiplying 0.024, 0.995, 0.8, and 0.704 together yielded the growth driving force index X, approximately 0.01346d. -1 .

[0131] In this embodiment, the growth rate prediction result for the future target period directly uses X as the predicted value of the growth intensity per unit time in the next 7 days; if a more intuitive result is needed on site, it can be converted into the body mass growth trend in the next 7 days in the display layer, but this display method does not change the essence of obtaining the growth rate prediction result for the future target period based on the growth driving force index.

[0132] like Figure 4 The figure shows the relationship between water temperature and growth driving force indices under different dissolved oxygen conditions at a fixed stage.

[0133] In this embodiment, parameters such as the stage-based growth rate, target water temperature center, allowable water temperature deviation, target pH center, allowable pH deviation, and density-feeding rhythm coupling coefficient corresponding to the same growth stage are kept constant. The current pH, stocking density, and feeding interval are also fixed. Only the water temperature and dissolved oxygen values ​​are changed. The growth driving force index under different operating conditions is calculated according to the aforementioned growth driving force index formula. Figure 4 It is evident that, under fixed stage parameters and fixed aquaculture pressure, the growth driving force index exhibits a trend of first increasing and then decreasing with water temperature changes, reaching a relatively high value near the target water temperature center of that stage. Simultaneously, the overall height of the corresponding curves varies under different dissolved oxygen conditions, indicating that increasing dissolved oxygen can enhance the growth driving force index, but it does not alter its peak distribution pattern relative to the target water temperature center. This result demonstrates that the exponential decay term reflects the inhibitory effect of water temperature deviation from the target center on growth potential, while the dissolved oxygen response term reflects the promoting effect of improved oxygen supply on growth potential; both together determine the magnitude of the growth driving force index under the current operating conditions.

[0134] Specifically, S5 includes:

[0135] Extract the model prediction residuals used to characterize prediction stability, the sensor drift used to characterize sensor reliability, the recording delay used to characterize the timeliness of manual recording, and the missing proportion used to characterize data integrity;

[0136] The model prediction residual, the sensor drift, the recording delay, and the missing proportion are normalized to form a data uncertainty load.

[0137] The prediction confidence index is calculated by combining the current temperature suitability and physicochemical environmental suitability. The greater the degree of environmental deviation and the higher the data uncertainty load, the lower the prediction confidence index.

[0138] In this embodiment, the pre-input of the prediction confidence is not a single error term, but a data uncertainty load formed by the model prediction residual, sensor drift, recording delay, and data missing ratio.

[0139] The model prediction residuals are used to characterize the consistency between recent predicted outputs and actual posterior observations. Specifically, within each stage, several recently closed prediction windows are collected, preferably 10 to 30 closed windows. For each closed window, the difference between its historical predicted growth rate and the subsequent actual observed growth rate is taken as the prediction error. The weighted variance of these errors is then calculated to obtain the variance of the model prediction residuals for that stage. Closed windows closer to the current time are given higher weights to reflect the greater impact of recent operating conditions on the current confidence level.

[0140] Sensor drift is used to characterize the stability of the environmental sensing link. In this embodiment, short-term and long-term rolling statistics are established for water temperature, dissolved oxygen, and pH. If manual calibration records exist on-site, the drift score is preferably constructed using the deviation before and after calibration; if no manual calibration records exist, the drift score is preferably constructed using the offset between the median value of the past 24 hours and the median value of the past 7 days. The drift scores for water temperature, dissolved oxygen, and pH are normalized and combined, and their variance is calculated over the most recent several windows as the sensor drift variance.

[0141] The recording time lag is used to characterize the staleness of records related to human events. In this embodiment, feeding records, water change records, sampling and weighing results, feeding activity records, and abnormal inspection records are all timestamped. For the current time slice, the time lag of key human events from the current moment is calculated, and different importance is assigned based on their impact on the current decision. Preferably, the importance of sampling and weighing results and abnormal inspection records is higher than that of feeding activity records. The time lag, after being converted to a uniform scale, is then converted to a uniform scale, and the variance is calculated over the most recent windows to obtain the recorded time lag converted variance.

[0142] The data missing ratio indicates the degree of missing key fields within the current time window. In this embodiment, key fields include at least water temperature, dissolved oxygen, pH, stocking density, feeding interval, and the average body mass of the most recent sample. For each field, the missing ratio between the theoretically expected value and the actual effective value for the current time window is calculated, and then the total missing ratio between 0 and 1 is obtained. If any key field is completely missing in the current time window, it is preferable to directly raise the total missing ratio to no less than 0.2 to reflect that the window is significantly unstable.

[0143] In this embodiment, the reason for using the above four types of quantities as inputs to the prediction confidence score is that they respectively reflect whether the current prediction result is worthwhile for direct use in regulation from four aspects: prediction error, perceived stability, recording timeliness, and data integrity. This avoids giving a high confidence score simply because the environmental values ​​appear reasonable.

[0144] Specifically, the prediction confidence index is calculated as follows:

[0145] The prediction confidence index Y is calculated according to the following formula:

[0146]

[0147] in, This represents the variance of the predicted residuals at the k-th growth stage; Indicates the sensor drift variance; This indicates the variance calculated based on the time delay. denoted as the baseline variance; Mk represents the proportion of missing data in the k-th growth stage; μ represents the missing data penalty coefficient.

[0148] In this embodiment, the baseline variance The data was obtained through statistical analysis of stable operating condition samples. A stable operating condition sample refers to a window of historical data that simultaneously meets the following conditions: high completeness of key fields, minimal sensor drift, no major anomalies during abnormal inspections, and low recent prediction errors. The median of the sum of the three types of variances of these stable operating condition windows was calculated to obtain the baseline variance. The median, rather than the mean, was used to reduce the impact of extreme outliers on the baseline variance.

[0149] The missing value penalty coefficient μ is obtained through a validation set search. Specifically, a reliable prediction label is first defined: if the absolute deviation between the predicted growth rate and the actual observed growth rate of a historical window does not exceed a preset tolerance error, it is denoted as a reliable window; otherwise, it is denoted as an unreliable window. The tolerance error is preferably set to 10% to 25% of the baseline growth rate for that stage, more preferably 15%. Subsequently, multiple candidate values ​​for the missing value penalty coefficient are tried sequentially on the validation set, such as 0.2, 0.5, 1.0, 1.5, 2.0, 2.5, and 3.0. The prediction confidence index under each candidate value is calculated, and the consistency between the high-confidence samples and the reliable window label is compared. The candidate value with the highest consistency is selected as the final deployment parameter. This ensures that the missing value penalty coefficient is not determined empirically but is obtained through data validation.

[0150] In a set of numerical calculation examples, taking the juvenile leech stage as an example, let's assume the product of current environmental suitability and oxygen supply response is approximately 0.796. Let's assume the following statistical results from the most recently closed window: model prediction residual variance is 0.004, sensor drift variance is 0.003, and recording delay variance is 0.003; the baseline variance from stable condition samples is 0.08; the current window missing proportion is 0.02; and the missing penalty coefficient from validation set search is 0.5. The uncertainty suppression term is first obtained by adding the three types of variances to get 0.010, then dividing by 0.08 to get 0.125, then adding 0.5 and multiplying by 0.02 to get 0.01, and combining these gives 1.135, whose reciprocal is approximately 0.881. Multiplying 0.881 by the aforementioned 0.796 yields a prediction confidence index Y of approximately 0.701.

[0151] In this embodiment, to ensure that Y truly enters the control chain, it needs to be divided into confidence intervals. Preferably, a combination of high-confidence, medium-confidence, and low-confidence thresholds is determined through validation set search. For example, Y ≥ 0.65 is considered high confidence; 0.35 ≤ Y < 0.65 is considered medium confidence; and Y < 0.35Y is considered low confidence. The search objective for the thresholds is: the high-confidence window should correspond to low-error predictions as much as possible, and the low-confidence window should cover high-uncertainty conditions as much as possible. Once the thresholds are determined, they remain fixed within the same deployment version and are only updated overall after parameter recalibration.

[0152] like Figure 5 , Figure 5The graph illustrating the relationship between prediction confidence and uncertainty load reflects the trend of prediction confidence as the uncertainty load changes under different environmental suitability and oxygen supply response conditions. In this embodiment, several representative levels are taken for the environmental suitability and oxygen supply response items, with the uncertainty load gradually changing from low to high, and the prediction confidence index is calculated accordingly. When the environmental suitability and oxygen supply response items remain fixed, the prediction confidence index decreases as the data uncertainty load increases; when the environmental suitability and oxygen supply response items are high, the corresponding prediction confidence index is generally high; when the environmental suitability and oxygen supply response items are low, the corresponding prediction confidence index is generally low. Figure 5 The composition mechanism of the prediction confidence index is consistent, and it can be used to explain the inhibitory effect of uncertainty load on prediction confidence and the basic supporting role of environmental state on prediction confidence.

[0153] Specifically, S6 includes:

[0154] A set of regulatory candidates is constructed according to the current growth stage. The set of regulatory candidates includes multiple candidate regulatory schemes. Each candidate regulatory scheme includes at least the target feeding interval, the target water exchange ratio, the target oxygenation retention margin, the target temperature control setpoint, and the target stocking density upper limit.

[0155] When the prediction confidence index reaches the high confidence interval, an active fine-tuning plan is allowed to be implemented to adjust the feeding rhythm, water exchange intensity, oxygenation retention margin and temperature control target in a coordinated manner.

[0156] When the prediction confidence index is in the middle confidence interval, only the limited adjustment scheme is allowed to be implemented, and the feeding rhythm and environmental control parameters are corrected in small steps.

[0157] When the prediction confidence index is in the low confidence interval, a conservative maintenance plan is implemented, and at least one of the following manual verification actions is triggered: manual sampling and weighing, close-range inspection, or pre-pooling verification.

[0158] Specifically, target control scheme Determined according to the following formula:

[0159]

[0160] Among them, Z k Z represents the set of regulatory candidates corresponding to the k-th growth stage; j Let X represent the j-th candidate regulatory scheme; Y represent the growth driving force index; and T represent the prediction confidence index. j τ represents the target temperature setpoint for the j-th candidate control scheme; j ρ represents the target feeding interval for the j-th candidate control scheme; jδ represents the upper limit of the target stocking density for the j-th candidate control scheme; j ω represents the target oxygenation retention margin for the j-th candidate regulation scheme; j This represents the target water exchange ratio for the j-th candidate control scheme; This represents the conservative water exchange ratio for the k-th growth stage; This represents the target stocking density for the k-th growth stage.

[0161] In one specific implementation, the target temperature control setpoint can be discrete within ±0.5℃ and ±1.0℃, centered around the current value and the target water temperature for the period; the target feeding interval can be discrete within ±0.25d and ±0.5d, centered around the current feeding interval; the target water exchange ratio can be discrete within ±5% and ±10%, centered around the current value; the target oxygenation retention margin can be selected from discrete values ​​such as 0, 0.5, 1.0, and 1.5 mg / L; and the target stocking density upper limit can be discrete within ±5% and ±10%, centered around the current density. These combinations form a set of candidate schemes for the current time slice. To ensure real-time performance, the number of candidate schemes is preferably controlled between 5 and 15 groups.

[0162] In this embodiment, the prediction confidence level directly determines the executable control level. When the current window is in a high confidence interval, an active fine-tuning scheme is allowed, meaning that two or more key control variables can be adjusted simultaneously, such as adjusting the temperature setpoint, feeding interval, and oxygenation reserve simultaneously. When the current window is in a medium confidence interval, only a limited adjustment scheme is allowed, meaning that only one main control variable is allowed to change significantly within each control cycle, while other control variables are only corrected in small steps. When the current window is in a low confidence interval, only a conservative maintenance scheme is allowed, triggering at least one manual verification action among manual sampling and weighing, close-range inspection, or pre-pooling verification.

[0163] The so-called proactive fine-tuning, amplitude-limited adjustment, and conservative maintenance are all specific implementations of the hierarchical control logic. Their purpose is to avoid excessive intervention when prediction results are unreliable, and to prevent missed proactive optimization opportunities within high-confidence windows. Under boundary conditions, this embodiment also sets downgrade rules: if the current water temperature, dissolved oxygen, or pH exceeds the safe statistical boundary for that stage, or if the proportion of missing key fields exceeds a preset upper limit, then even if the Y value falls within the medium-high range, it will be preferentially handled according to a low-confidence or conservative maintenance strategy.

[0164] In this embodiment, to facilitate the explanation of the differences in the composition of candidate control schemes and their comprehensive evaluation process, multiple candidate control schemes are first generated based on the growth stage and current aquaculture status of the current time window, according to the target temperature control setpoint, target feeding interval, target water exchange ratio, target oxygenation reserve, and target stocking density limit. From this set of candidate control schemes, three representative schemes are selected as example schemes for demonstration and comparison, denoted as Candidate Scheme A, Candidate Scheme B, and Candidate Scheme C, respectively. Candidate Scheme A represents a balanced scheme that considers both growth promotion and execution risk under the current conditions; Candidate Scheme B represents a more aggressive scheme with relatively large adjustments in temperature control, oxygenation, water exchange, or density constraints; and Candidate Scheme C represents a relatively conservative and robust scheme in terms of water exchange ratio, oxygenation reserve, or density constraints. It should be noted that Candidate Schemes A, B, and C all originate from the same set of candidate control schemes generated under the same current time window and are only used to illustrate the control characteristics and comprehensive evaluation results corresponding to different parameter combinations, and do not constitute a limitation on the number of candidate control schemes.

[0165] In this embodiment, to facilitate the explanation of the differences in the composition of candidate control schemes and their comprehensive evaluation process, multiple candidate control schemes are first generated based on the growth stage and current aquaculture status of the current time window, focusing on the target temperature control setpoint, target feeding interval, target water exchange ratio, target oxygenation retention margin, and target stocking density upper limit. From this set of candidate control schemes, three representative groups are selected as example schemes for demonstration and comparison, denoted as Candidate Scheme A, Candidate Scheme B, and Candidate Scheme C, respectively.

[0166] like Figure 6 This reflects the differences in parameter distribution across five control dimensions—target temperature setpoint, target feeding interval, target water exchange ratio, target oxygenation retention margin, and target stocking density upper limit—among candidate control schemes A, B, and C. In this embodiment, the control unit does not directly output a single control value. Instead, it first constructs multiple candidate control schemes based on the current operating conditions and the current stage target, and then comprehensively evaluates each candidate scheme. The results show that different candidate control schemes have different value combinations across multiple control dimensions, thus forming a set of schemes with differentiated control intensity and risk levels, providing an input basis for determining subsequent target control schemes.

[0167] like Figure 7This reflects the comparison results of the growth promotion items, risk penalty items, and comprehensive evaluation values ​​corresponding to candidate schemes A, B, and C. In this embodiment, the control unit evaluates multiple candidate control schemes generated in the current stage. Different candidate control schemes differ in terms of growth promotion benefits and execution risks. Among them, candidate scheme B has a larger risk penalty item, resulting in its lowest comprehensive evaluation value; although candidate scheme C has a smaller risk penalty item, its comprehensive evaluation value is still lower than that of candidate scheme A; candidate scheme A has the highest comprehensive evaluation value and is therefore selected as the target control scheme for the current time window. This result shows that this embodiment does not determine the control scheme based solely on a single control quantity or a single benefit indicator, but rather completes the selection of the target control scheme by comprehensively weighing the growth promotion capabilities and execution risks of the candidate schemes.

[0168] like Figure 2 A confidence-driven control system for predicting growth rate in Hirudo medicinalis farming.

[0169] In this embodiment, the growth rate prediction confidence-driven control system for Hirudo nipponia farming adopts an architecture consisting of a front-end acquisition layer, an edge control layer, an execution control layer, and a manual verification and interaction layer. The front-end acquisition layer corresponds to the environmental perception unit and the farming record unit. The edge control layer corresponds to the control unit, which preferably has a time-series data cache and a stage parameter library to store environmental monitoring data, farming operation data, and growth sampling data within the most recent preset time window, as well as the target parameters, response constants, coupling coefficients, and confidence interval thresholds for each stage. The execution control layer corresponds to the execution control unit, and the manual verification and interaction layer undertakes manual verification tasks such as manual sampling and weighing, close-range inspection, and pre-pool verification. Through this layered architecture, data acquisition, status identification, index calculation, scheme determination, and execution output in the method can form a closed loop.

[0170] In one specific embodiment, the environmental sensing unit is installed at each aquaculture unit to continuously collect environmental monitoring data. The environmental sensing unit includes at least a water temperature sensor, a dissolved oxygen sensor, and a pH sensor. Preferably, the environmental sensing unit is connected to the edge control node via a field acquisition link. Optionally, the field acquisition link uses an RS485 bus and transmits data using the Modbus protocol, enabling environmental monitoring data from each aquaculture unit to be uploaded to the control unit according to a unified address and time reference. The advantages of this method are stable transmission link, strong anti-interference capability, and suitability for continuous data collection from multiple points at the aquaculture site.

[0171] In one specific embodiment, the aquaculture recording unit is used to acquire aquaculture operation data and growth sampling data of each aquaculture unit. The aquaculture recording unit can be composed of a field input terminal, a mobile terminal, or an industrial control touch terminal, and is used to input feeding records, water change records, stocking density, sampling and weighing results, feeding activity records, and abnormal inspection records. To facilitate unified processing with environmental monitoring data, the aquaculture recording unit preferably adds an aquaculture unit number, record type, and timestamp to each record. The aquaculture recording unit and the control unit can optionally use MQTT message transmission for reporting, i.e., the aquaculture recording unit publishes event-type records to the corresponding topic, and the control unit subscribes to the relevant topic and receives the record content; alternatively, a direct connection via a local interface can be used to write to the database or cache of the edge control node. All of the above implementation methods are feasible architectures of this embodiment.

[0172] In this embodiment, the control unit is preferably an edge computing control node, deployed near the aquaculture site, used to complete data preprocessing, stage identification, growth driving force index calculation, prediction confidence index calculation, control scheme determination, and execution control locally. The edge computing control node can be composed of an industrial computer, an embedded industrial control host, or an edge server, on which data access services, time-series caching services, parameter library services, and corresponding functional modules are deployed. The purpose of adopting an edge control architecture is to enable environmental monitoring data and manually recorded data to be processed locally with low latency, and to complete calculation and output within the same control cycle, without relying on real-time results returned from a remote central location.

[0173] Furthermore, the calling relationships between the various modules within the control unit are as follows. The data preprocessing module first receives environmental monitoring data from the environmental sensing unit and aquaculture operation data and growth sampling data from the aquaculture record unit, and performs time alignment, anomaly removal, and validity screening on the data. The stage identification module receives valid data output by the data preprocessing module and state quantities such as the average body weight of the most recent sampling and the number of aquaculture days, identifies whether the aquaculture object belongs to the juvenile leech stage, the young leech stage, or the commercial leech stage, and constructs the aquaculture state sequence for the corresponding stage. The growth driving force calculation module calculates the growth driving force index of the current aquaculture unit based on the stage identification results and the basic aquaculture state characterization, calls the stage parameter library and corresponding formulas, and obtains the growth rate prediction result for the future target period. The prediction confidence assessment module calculates the prediction confidence index corresponding to the growth rate prediction result based on the basic aquaculture state characterization, the recent closed prediction window statistics, and the data quality status, calls the corresponding formulas. The control scheme determination module determines the target control scheme from the preset control candidate set based on the growth driving force index and the prediction confidence index. The execution control module generates control instructions based on the target control scheme and sends the control instructions to the execution control unit. The preset control candidate set is a set of candidate control schemes generated according to preset candidate construction rules, preset control boundaries, and preset step sizes, based on the current stage objectives and current operating conditions.

[0174] In this embodiment, the control unit preferably includes a local time-series data cache and a stage parameter library. The time-series data cache stores environmental monitoring data, aquaculture operation data, and growth sampling data within the most recent preset time window, allowing the data preprocessing module, prediction confidence assessment module, and control scheme determination module to access the data by time slice. The stage parameter library stores the target water temperature center, allowable water temperature deviation, target pH center, allowable pH deviation, minimum dissolved oxygen requirement, stage baseline growth rate, oxygen response constant, density-feeding rhythm coupling coefficient, baseline variance, missing value penalty coefficient, and confidence interval threshold for the juvenile leech stage, young leech stage, and marketable leech stage. During online operation, each module reads the corresponding stage parameters from the parameter library, avoiding the mixing of parameters between different stages.

[0175] In one specific embodiment, the execution control unit includes at least a feeding execution device, a water exchange execution device, an oxygenation execution device, and a temperature control execution device. The feeding execution device is used to modify the feeding plan according to the target feeding interval output by the control unit; the water exchange execution device is used to perform water exchange actions according to the target water exchange ratio; the oxygenation execution device is used to modify the oxygenation strategy according to the target oxygenation reserve; and the temperature control execution device is used to adjust the aquaculture water temperature according to the target temperature control setpoint. For the pond division adjustment, when the current stocking density is higher than the upper limit of the target stocking density corresponding to the candidate control scheme, and this state continuously reaches a preset number of control cycles, the current moment or the start of the next control cycle is determined as the pond division timing, and a pond division reminder or pond division control command is output. Optionally, two implementation methods can be adopted: first, the execution control unit is directly connected to the pond division execution mechanism and performs the pond division action when the pond division conditions are met; second, the execution control unit does not directly drive the pond division equipment, but outputs a pond division reminder to the manual review interaction layer, and the aquaculture personnel complete the pond division operation. Both of the above methods belong to the implementation forms of pond division adjustment.

[0176] In this embodiment, the manual verification interaction layer includes a manual verification alarm interface and an inspection terminal interface. The manual verification alarm interface is used to issue manual sampling and weighing, close-range inspection, or pre-pond verification tasks to the aquaculture personnel when the prediction confidence level is in the low confidence interval, the proportion of missing key data exceeds the limit, or environmental parameters exceed the stage safety boundary. The inspection terminal interface is used to receive the on-site verification results returned by the aquaculture personnel and write these results back to the aquaculture record unit as data input for the next control cycle. Through this feedback link, the manual verification results under the low confidence window can be re-involved in subsequent stage identification and prediction calculations, forming a closed-loop update.

[0177] In a preferred embodiment, the environmental sensing unit and the control unit communicate via RS485 / Modbus for on-site data acquisition, the aquaculture recording unit and the control unit communicate via MQTT message reporting, and the control unit and the execution and control unit communicate via relay control interface, industrial control I / O interface, or local control command interface to transmit execution commands. With this combined architecture, continuously sampled environmental data enters the edge control node through a stable bus, while event-based operation records and manual records enter the edge control node via messages. The edge control node performs calculations locally and issues control results, thus enabling the entire system to possess good real-time performance and engineering feasibility.

[0178] In this embodiment, the system executes cyclically according to a fixed control cycle during online operation. At each control cycle, the environmental sensing unit uploads environmental monitoring data for the current time slice to the control unit, and the aquaculture recording unit uploads the latest aquaculture operation records and growth sampling records to the control unit. The control unit sequentially calls the data preprocessing module, stage identification module, growth driving force calculation module, prediction confidence assessment module, and control scheme determination module to form the target control scheme for the current time slice. Subsequently, the execution control module converts the target control scheme into control instructions for the execution control unit, and outputs the corresponding task through the manual review interaction layer when manual review is required. This completes a system-level closed loop from data acquisition, indicator calculation, scheme optimization to execution output.

Claims

1. A confidence-driven regulation method for predicting growth rate in Hirudo medicinalis farming, characterized in that, include: S1. Collect environmental monitoring data, aquaculture operation data and growth sampling data of each aquaculture unit. The environmental monitoring data shall include at least water temperature, dissolved oxygen and pH. The aquaculture operation data shall include at least feeding records, water change records and stocking density. The growth sampling data shall include at least sampling and weighing results, feeding activity records and abnormal inspection records. S2. Perform time alignment, anomaly removal, and validity screening on the environmental monitoring data, aquaculture operation data, and growth sampling data, and identify the stage of the aquaculture object according to its juvenile leech stage, young leech stage, or commercial leech stage. S3. Based on the stage identification results, construct the corresponding stage breeding state sequence to form the basic breeding state representation of each breeding unit under the current stage; S4. Based on the target growth conditions at the current stage and combined with the basic aquaculture status characterization, calculate the growth driving force index characterizing the achievability of the current aquaculture unit's growth rate, and obtain the growth rate prediction result for the future target period accordingly. S5. Based on the target growth conditions of the current stage, the basic aquaculture status characterization, and the data quality status, calculate the prediction confidence index corresponding to the growth rate prediction result. S6. Based on the growth driving force index and the prediction confidence index, determine the target control scheme from the preset control candidate set, and control at least one of the following according to the target control scheme: feeding rhythm, water exchange intensity, oxygenation retention margin, temperature control target, pond separation timing and manual verification triggering conditions. S5 specifically includes: Extract the model prediction residuals used to characterize prediction stability, the sensor drift used to characterize sensor reliability, the recording delay used to characterize the timeliness of manual recording, and the missing proportion used to characterize data integrity; The model prediction residual, the sensor drift, the recording delay, and the missing proportion are normalized to form a data uncertainty load. Taking into account the current temperature suitability and physicochemical environment suitability, the prediction confidence index is calculated such that the greater the degree of environmental deviation and the higher the data uncertainty load, the lower the prediction confidence index. The prediction confidence index is calculated as follows: The basic environmental suitability value is determined based on the degree of deviation of the current water temperature from the target water temperature center of the current growth stage and the degree of deviation of the current pH from the target pH center of the current growth stage. The closer the current water temperature and current pH are to the target center value of the corresponding stage, the higher the basic environmental suitability value. Based on the relationship between the current dissolved oxygen level and the minimum dissolved oxygen requirement at the current growth stage, the oxygen supply reliability of the aforementioned environmental suitability baseline value is corrected. The higher the current dissolved oxygen level can meet the oxygen requirement of the corresponding stage, the higher the oxygen supply reliability. The model prediction residual, sensor drift, recording delay, and data missing ratio are then converted into data uncertainty load. The larger the model prediction residual, the more obvious the sensor drift, the longer the recording delay, and the higher the data missing ratio, the higher the data uncertainty load. Finally, using the baseline environmental suitability value and the oxygen supply reliability correction result, the data uncertainty load is subjected to inverse constraints to obtain the prediction confidence index, so that the more suitable the environmental state and the lower the data uncertainty, the higher the prediction confidence index.

2. The growth rate prediction confidence-driven regulation method for Hirudo medicinalis farming according to claim 1, characterized in that, S4 specifically includes: The target water temperature center, allowable deviation of water temperature, target pH center, allowable deviation of pH, minimum dissolved oxygen requirement, baseline growth rate, and coupling strength between density and feeding rhythm are determined for the juvenile leech stage, the young leech stage, and the commercial leech stage, respectively. The deviation between the current water temperature and the target water temperature center of the corresponding stage is used as a temperature suitability characterization; the deviation between the current pH and the target pH center of the corresponding stage, as well as the relationship between the current dissolved oxygen and the minimum dissolved oxygen requirement of the corresponding stage, are used as a physicochemical environment suitability characterization; and the combination relationship between the current stocking density and the current feeding rhythm is used as a crowding feeding coupling characterization. The growth driving force index is calculated based on the temperature suitability characterization, the physicochemical environment suitability characterization, and the crowding-feeding coupling characterization.

3. The growth rate prediction confidence-driven regulation method for Hirudo medicinalis farming according to claim 2, characterized in that, The growth driving force index is calculated as follows: The baseline growth rate corresponding to the current growth stage is used as the base value; The baseline value is attenuated and corrected based on the degree of deviation of the current water temperature from the target water temperature center of the current growth stage and the degree of deviation of the current pH from the target pH center of the current growth stage. The greater the deviation of the current water temperature or current pH from the target center value of the corresponding stage, the greater the attenuation of the baseline value. Based on the relationship between the current dissolved oxygen level and the minimum dissolved oxygen requirement at the current growth stage, the oxygen supply response is corrected after attenuation correction. The closer the current dissolved oxygen level is to or higher than the oxygen requirement level at the current growth stage, the stronger the oxygen supply response. Furthermore, crowding-feeding constraint correction is performed by combining the coupling relationship between the current stocking density and the current feeding interval. The higher the current stocking density and the longer the current feeding interval, the stronger the inhibitory effect on the result. The final result is a growth driving force index used to characterize the achievability of the current growth rate of the breeding unit.

4. The growth rate prediction confidence-driven regulation method for Hirudo medicinalis farming according to claim 3, characterized in that, S6 specifically includes: A set of regulatory candidates is constructed according to the current growth stage. The set of regulatory candidates includes multiple candidate regulatory schemes. Each candidate regulatory scheme includes at least the target feeding interval, the target water exchange ratio, the target oxygenation retention margin, the target temperature control setpoint, and the target stocking density upper limit. When the prediction confidence index reaches the high confidence interval, an active fine-tuning plan is allowed to be implemented to adjust the feeding rhythm, water exchange intensity, oxygenation retention margin and temperature control target in a coordinated manner. When the prediction confidence index is in the middle confidence interval, only the limited adjustment scheme is allowed to be implemented, and the feeding rhythm and environmental control parameters are corrected in small steps. When the prediction confidence index is in the low confidence interval, a conservative maintenance plan is implemented, and at least one of the following manual verification actions is triggered: manual sampling and weighing, close-range inspection, or pre-pooling verification.

5. The growth rate prediction confidence-driven regulation method for Hirudo medicinalis farming according to claim 4, characterized in that, The target control scheme is determined in the following manner: For the current growth stage, candidate regulatory schemes are retrieved one by one from the corresponding regulatory candidate set; For each candidate regulation scheme, the degree to which the candidate regulation scheme promotes the growth rate improvement in the future target period is evaluated in conjunction with the growth driving force index, and the feasibility and reliability of the candidate regulation scheme are evaluated in conjunction with the prediction confidence index. When evaluating the degree of promotion, the matching degree between the target temperature control setpoint corresponding to the candidate control scheme and the target water temperature center at the current growth stage, the degree of improvement of the dissolved oxygen guarantee capacity by the target oxygenation retention margin corresponding to the candidate control scheme, and the degree of coupling constraint between the target stocking density upper limit and the target feeding interval corresponding to the candidate control scheme are also considered. When assessing the feasibility and reliability of the proposed scheme, the deviation of the target water exchange ratio corresponding to the candidate control scheme from the conservative water exchange ratio at the current growth stage is also considered, as well as the deviation of the upper limit of the target stocking density corresponding to the candidate control scheme from the target stocking density at the current growth stage. Considering the combined benefits of growth promotion and the execution risks associated with deviating from conservative operating conditions, the candidate control scheme with the best overall evaluation is determined as the target control scheme.

6. A growth rate prediction confidence-driven control system for Hirudo medicinalis farming, characterized in that, For performing the method according to any one of claims 1-5, comprising: An environmental sensing unit is used to collect environmental monitoring data from each aquaculture unit. The environmental monitoring data includes at least water temperature, dissolved oxygen, and pH. The breeding record unit is used to acquire breeding operation data and growth sampling data of each breeding unit. The breeding operation data includes at least feeding records, water change records and stocking density. The growth sampling data includes at least sampling and weighing results, feeding activity records and abnormal inspection records. The execution control unit is used to perform at least one of the following control actions on each aquaculture unit: feeding regulation, water exchange regulation, oxygenation regulation, temperature control regulation, pond division regulation, and manual verification triggering; The control unit is communicatively connected to the environmental sensing unit, the aquaculture recording unit, and the execution control unit, respectively. The control unit includes: The data preprocessing module is used to perform time alignment, anomaly removal, and validity screening on the environmental monitoring data, aquaculture operation data, and growth sampling data; The stage identification module is used to identify the stage of the leech according to whether the leech is in the juvenile leech stage, the young leech stage, or the commercial leech stage, and to construct the corresponding breeding status sequence for the stage. The growth driving force calculation module is used to calculate the growth driving force index that represents the achievability of the current growth rate of the current breeding unit based on the target growth conditions and basic breeding status of the current stage, and to obtain the growth rate prediction results for the future target period. The prediction confidence assessment module is used to calculate the prediction confidence index corresponding to the growth rate prediction result based on the target growth conditions, basic aquaculture status, and data quality status at the current stage. The regulation scheme determination module is used to determine the target regulation scheme from a preset regulation candidate set based on the growth driving force index and the prediction confidence index. The execution control module is used to control the execution control unit according to the target control scheme, and to regulate at least one of the following: feeding rhythm, water exchange intensity, oxygenation retention, temperature control target, pond separation timing, and manual verification trigger conditions.

Images