Fish egg hatching rate prediction optimization method based on deep learning

Abstract

The invention discloses a fish egg hatching rate prediction optimization method based on deep learning, and belongs to the technical field of deep learning. Hatching environment parameters such as water temperature, dissolved oxygen and illumination intensity and roe physiological state parameters are collected, and a unified feature identifier is established; by introducing environment disturbance parameters such as temperature fluctuation, dissolved oxygen change and pollutant concentration and combining with a hatching rate index for correction, a dynamic response diagram under a real working condition is generated, and the robustness of a prediction model in a complex environment is improved; by constructing a multi-modal prediction strategy and a priority weighted multi-factor collaborative prediction mechanism, efficient collaboration and conflict arbitration among environmental factors are realized; and a real-time evaluation signal is generated in combination with sensor feedback, and the deep learning network is continuously updated, so that online optimization and adaptive adjustment of the prediction model are realized. The method can effectively improve the prediction precision of the roe hatching rate, improves the stability, energy efficiency and success rate of the hatching process, and has good agricultural applicability and expansibility.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of deep learning, in particular to a fish egg hatching rate prediction optimization method based on deep learning. BACKGROUND

[0002] With the continuous development of aquaculture technology, the fish egg hatching process has gradually become a key link to improve the efficiency and quality of fish breeding. The fish egg hatching process is affected by many environmental factors, including water temperature, dissolved oxygen, light cycle, etc. The coordinated control of these factors is crucial to the success rate of hatching. However, the hatching environment has high dynamicity and uncertainty, such as temperature fluctuations, dissolved oxygen changes, and pollutant invasions, which will significantly affect the development process and hatching rate of fish eggs. Existing researches mostly use traditional statistical prediction methods, including regression model-based hatching rate prediction methods, empirical formula methods, and fuzzy logic methods, etc. These methods have certain effect in dealing with single factor or stable environment, but their prediction accuracy and adaptability are often insufficient in the face of multi-factor coupling and dynamic changes in complex environments. In addition, existing methods generally rely on historical data and fixed models, while the hatching environment has high nonlinearity and time-varying nature, which is difficult to accurately describe by deterministic models.

[0003] However, the existing researches on directly applying deep learning to fish egg hatching rate prediction optimization are still limited, mainly focusing on offline analysis, lacking of system integration of environmental factor characteristic parameters and dynamic response mechanism design of multi-source disturbance factors. In terms of environmental factors, current methods often only consider single temperature or dissolved oxygen control, failing to achieve multi-factor collaborative prediction of multiple environmental factors, resulting in slow prediction adjustment speed, high energy consumption and insufficient accuracy under complex disturbances. Especially in the aspects of prediction strategy scheduling and priority weight introduction, traditional methods lack adaptive dynamic optimization mechanism, and cannot realize flexible adjustment for different disturbance intensities and coupling effects. Therefore, the existing technology still has obvious deficiencies in the real-time, collaboration and adaptability of hatching rate prediction. SUMMARY

[0004] The present application aims to provide a fish egg hatching rate prediction optimization method based on deep learning to solve the problems raised in the background art.

[0005] To solve the above technical problems, the present application provides the following technical solutions: The method comprises the following steps: S1, collecting basic performance parameters of different environmental factor units from a fish egg hatching control system backend, and constructing a multi-source state feature set; S2, presetting hatching environment disturbance parameters, constructing a dynamic disturbance label hatching response graph, constructing a hatching rate index matrix, matching the hatching rate index matrix, and generating a corrected dynamic disturbance label hatching response graph; S3, based on the corrected dynamic disturbance label hatching response graph, constructing an environmental factor action space set, introducing a multi-modal prediction strategy to the environmental factor action space set, constructing a prediction strategy scheduling matrix, and generating a multi-modal prediction strategy distribution graph; S4, based on the multi-modal prediction strategy distribution graph, mapping all environmental factor units in the corrected dynamic disturbance label hatching response graph and the corresponding prediction strategy items, constructing a multi-factor collaborative strategy mapping matrix, and introducing a regulation priority weight to generate a multi-factor collaborative regulation graph; S5, based on the multi-factor collaborative regulation graph, generating a final hatching optimization instruction sequence; the final hatching optimization instruction sequence is sent to the corresponding environmental regulation unit through the control system communication bus; an instant evaluation signal is generated; S6, based on the final hatching optimization instruction sequence, the current hatching environment disturbance parameters, the new physiological state parameters and the instant evaluation signal, an experience tuple is constructed; the prediction network and evaluation network parameters of the deep learning network are updated, and adaptive prediction and adjustment are fed back.

[0006] As a preferred scheme of the fish egg hatching rate prediction optimization method based on deep learning, the basic performance parameters of different environmental factor units are collected from the fish egg hatching control system backend, the environmental factor units include water temperature control unit, dissolved oxygen regulation unit and light management unit, the basic performance parameters of the water temperature control unit include heating power, temperature control accuracy and energy consumption parameters, the basic performance parameters of the dissolved oxygen regulation unit include oxygen supply rate, dissolved oxygen stability and response time, and the basic performance parameters of the light management unit include light intensity range, spectral composition and periodic adjustment parameters.

[0007] As a preferred scheme of the fish egg hatching rate prediction optimization method based on deep learning, based on the basic performance parameters of the water temperature control unit, the basic performance parameters of the dissolved oxygen regulation unit and the basic performance parameters of the light management unit, a multi-source state feature set is constructed, specifically as follows: The uniform feature identification code is numbered, the uniform feature identification code includes a three-segment combined code using a unit category prefix, a device number and a performance serial number, and a corresponding relationship is established based on the physiological state parameters and control attribute parameters of the environmental factor unit, the physiological state parameters include the development stage of fish eggs, heart rate variability and egg membrane permeability, and the control attribute parameters include the regulation mode, available time period and action priority of the environmental factor unit. input the multi-source state feature set into a deep learning network, generate an initial hatching response graph according to the correlation relationship between the environmental factor units, the correlation relationship between the environmental factor units is calculated by normalizing and Euclidean distance to calculate the matching degree of the basic performance parameters and the control attribute parameters between the environmental factor units, and the distance weight factor of the preset physiological state parameters is combined for weighting, the synergy degree between different environmental factor units is calculated, and the specific process is as follows: ; wherein, denotes the synergy degree between the i th environmental factor unit and the j th environmental factor unit, denotes the basic performance parameter value of the i th environmental factor unit in the k th performance dimension, denotes the normalized value of the basic performance parameter of the i th environmental factor unit in the k th performance dimension, denotes the preset physiological state parameter weight factor, denotes the basic performance parameter value of the j th environmental factor unit in the k th performance dimension, denotes the normalized value of the basic performance parameter of the j th environmental factor unit in the k th performance dimension, and n denotes the total number of performance dimensions.

[0008] As a preferred scheme of the fish egg hatching rate prediction optimization method based on deep learning, preset hatching environment disturbance parameters are introduced, the hatching environment disturbance parameters include temperature fluctuation parameters, dissolved oxygen change parameters and pollutant concentration parameters; the hatching environment disturbance parameters are introduced into all environmental factor units in the initial hatching response graph, and the unified feature identification code is corresponded to construct a dynamic disturbance label hatching response graph; construct a hatching rate index matrix, the hatching rate index matrix includes hatching success rate, upper limit of deformity rate and upper limit of energy consumption; the dynamic disturbance label hatching response graph is matched with the hatching rate index matrix, the basic performance parameters of the environmental factor units are corrected by weighted matching, and the corrected dynamic disturbance label hatching response graph is generated, and the specific process is as follows: ; ; ; wherein, denotes the temperature fluctuation parameter of the i th environmental factor unit, denotes the dissolved oxygen change parameter of the i th environmental factor unit, denotes the pollutant concentration disturbance parameter of the i th environmental factor unit, denotes the preset environmental disturbance weight coefficient, denotes the hatching success rate, represents an upper limit of deformity rate, represents an upper limit of energy consumption, represents a disturbance correction coefficient of the i th environmental factor unit, represents a basic performance parameter value of the corrected i th environmental factor unit in the k th performance dimension, represents the degree of synergy between the corrected i th environmental factor unit and the j th environmental factor unit, represents a corrected dynamic disturbance label hatching response graph matrix, and m represents the total number of environmental factor units.

[0009] As a preferred scheme of the fish egg hatching rate prediction optimization method based on deep learning according to the present application, based on the corrected dynamic disturbance label hatching response graph, the temperature control precision parameter of the water temperature control unit and the periodic adjustment parameter of the light management unit are extracted; the action capability parameter is uniformly quantized, and an environmental factor action space set is constructed; A multi-modal prediction strategy is introduced to the environmental factor action space set, the multi-modal prediction strategy includes a main control prediction strategy, an auxiliary prediction strategy and an emergency correction strategy; and the multi-modal prediction strategy is bound with the unified feature identification code to construct a prediction strategy scheduling matrix; The prediction strategy scheduling matrix is input into a deep learning network, and a multi-modal prediction strategy distribution graph is generated according to the correlation between the physiological state parameters and the disturbance intensity parameters of the environmental factor units.

[0010] As a preferred scheme of the fish egg hatching rate prediction optimization method based on deep learning according to the present application, based on the multi-modal prediction strategy distribution graph, all environmental factor units in the corrected dynamic disturbance label hatching response graph are mapped with corresponding prediction strategy items to construct a multi-factor synergistic strategy mapping matrix, which is used to represent the controllability of different environmental factor units under the prediction strategy; A regulation priority weight is introduced to the multi-factor synergistic strategy mapping matrix, the regulation priority weight represents the control attribute parameter in the hatching rate index matrix, and the regulation priority weight and the multi-factor synergistic strategy mapping matrix are weighted and fused to construct a priority weighted prediction table, specifically as follows: ; ; Wherein, represents the response intensity of the i th environmental factor unit to the p th strategy, represents the regulation priority weight of the i th environmental factor unit, represents the response intensity of the weighted i th environmental factor unit to the p th strategy, This represents the degree of synergy between the i-th and j-th environmental factor units after modification. This represents the comprehensive control value of the i-th environmental factor unit, where m represents the total number of environmental factor units; The priority-weighted prediction table is input into a deep learning network, and a multi-factor collaborative regulation map is generated based on the correlation between environmental factor units, denoted as... .

[0011] As a preferred embodiment of the deep learning-based fish egg hatching rate prediction optimization method of the present invention, based on the multi-factor collaborative regulation graph, the collaborative prediction instruction set of all environmental factor units under the current hatching environment disturbance parameters is analyzed by utilizing the degree of collaboration and regulation priority weights between different environmental factor units. The collaborative prediction instruction set is compared and verified with the priority weighted prediction table. Instruction conflicts are identified by a rule-based conflict detection algorithm, and arbitration is performed according to the priority weights in the priority weighted prediction table to generate the final incubation optimization instruction sequence. The final incubation optimization instruction sequence is sent to the corresponding environmental control unit via the control system communication bus, as follows: The water temperature control unit fine-tunes the temperature, the dissolved oxygen regulation unit adjusts the oxygen supply rate, and the light management unit adjusts the light cycle. By deploying a sensor network in the incubator, new physiological state parameters of fish eggs are collected in real time after the command is executed, including actual developmental stage, actual heart rate and changes in egg membrane permeability. These parameters are then compared with the hatching rate index matrix. An instant evaluation signal is generated based on the hatching success rate, upper limit of deformity rate and upper limit of energy consumption. The instant evaluation signal is calculated by weighting the developmental deviation penalty, energy consumption penalty and health reward.

[0012] As a preferred embodiment of the deep learning-based fish egg hatching rate prediction and optimization method of the present invention, an experience tuple is constructed based on the final hatching optimization instruction sequence, the current hatching environment disturbance parameters, the new physiological state parameters and the real-time evaluation signal, and the experience tuple is stored in the experience replay pool of the deep learning network based on the first-in-first-out strategy. When the training cycle is triggered, a small batch of experience tuples is randomly sampled from the experience replay pool. Using a deep learning algorithm, the parameters of the prediction network and the evaluation network of the deep learning network are updated with the goal of minimizing the mean square error between the prediction evaluation and the target evaluation. The updated evaluation network parameters are then hot-updated online to the predictive generation engine deployed in the control system, providing feedback and enabling adaptive prediction and adjustment.

[0013] Compared with existing technologies, the beneficial effects achieved by this invention are as follows: The deep learning-based fish egg hatching rate prediction and optimization method provided by this invention collects basic performance parameters of environmental factors such as water temperature, dissolved oxygen, and light intensity, and constructs a multi-source state feature set. This achieves a quantitative description and unified encoding of the capabilities of environmental factors, enabling subsequent prediction training to have reliable input and shortening the convergence cycle. Subsequently, by introducing environmental disturbance parameters such as temperature fluctuations, dissolved oxygen changes, and pollutant concentrations, and combining them with hatching rate indicators such as hatching success rate, upper limit of deformity rate, and upper limit of energy consumption for matching and correction, it ensures that the generated dynamic disturbance label hatching response map can accurately reflect the real working conditions and safety constraints, thereby improving the robustness of the prediction model in extreme environments. Furthermore, by quantifying the action capability parameters and combining them with main control and auxiliary... A prediction strategy scheduling matrix is ​​constructed using three types of multimodal prediction strategies: prediction, emergency correction, and prediction. This enables layered responses to different disturbance levels, improving the system's fault tolerance and emergency response capabilities. Based on this, environmental factor units are mapped to prediction strategy entries, and priority weights are introduced to generate a multi-factor collaborative control graph. This effectively avoids conflicting commands and improves the coordination and execution efficiency of multi-factor collaboration. Subsequently, the prediction instruction set is analyzed based on the collaborative prediction graph and validated using a rule base to generate the final hatching optimization instruction sequence. Real-time evaluation signals are generated by a sensor network, achieving a closed-loop prediction process of instruction-execution-evaluation, thus improving the safety and transparency of prediction. Finally, by constructing experience tuples from the prediction results, disturbance, state, and evaluation information and updating the deep learning network, adaptive iterative optimization of the prediction model is achieved, enabling the system to continuously adapt to complex and changing hatching environments. Overall, this invention not only improves the safety, robustness, and energy efficiency of the fish egg hatching process but also possesses good scalability and engineering verifiability, thereby significantly enhancing the stable operation capability of the hatching system in real-world complex environments. Attached Figure Description

[0014] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0015] Figure 1 This is a schematic diagram illustrating the steps of the fish egg hatching rate prediction and optimization method based on deep learning in this invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Please see Figure 1 In this first embodiment: a deep learning-based method for predicting and optimizing fish egg hatching rates is provided, which includes the following steps: Step S1: Collect basic performance parameters of different environmental factor units from the back end of the fish egg hatching control system and construct a multi-source state feature set.

[0018] Specifically, basic performance parameters of different environmental factor units are collected from the back end of the fish egg hatching control system. These environmental factor units include a water temperature control unit, a dissolved oxygen regulation unit, and a light management unit. The basic performance parameters of the water temperature control unit include heating power, temperature control accuracy, and energy consumption parameters. The basic performance parameters of the dissolved oxygen regulation unit include oxygen supply rate, dissolved oxygen stability, and response time. The basic performance parameters of the light management unit include light intensity range, spectral composition, and periodic adjustment parameters.

[0019] Furthermore, based on the basic performance parameters of the water temperature control unit, the dissolved oxygen regulation unit, and the light management unit, a multi-source state feature set is constructed, as follows: Numbering is performed according to a unified feature identification code, which includes a three-segment combination code consisting of a unit category prefix, device number, and performance serial number. A correspondence is established based on the physiological state parameters and control attribute parameters of the environmental factor unit. The physiological state parameters include the developmental stage of the fish egg, heart rate variability, and egg membrane permeability. The control attribute parameters include the regulation mode, available time period, and action priority of the environmental factor unit. The multi-source state feature set is input into a deep learning network. An initial incubation response map is generated based on the correlation between environmental factor units. The correlation between environmental factor units is calculated by normalization and Euclidean distance to determine the matching degree of basic performance parameters and control attribute parameters between environmental factor units. The degree of synergy between different environmental factor units is calculated by weighting the results with a preset distance weight factor for physiological state parameters, as follows: ; in, This represents the degree of synergy between the i-th environmental factor unit and the j-th environmental factor unit. This represents the basic performance parameter value of the i-th environmental factor unit in the k-th performance dimension. For example, when i represents the water temperature control unit, It can represent heating power. Indicates the accuracy of temperature control. This represents the normalized value of the basic performance parameter of the i-th environmental factor unit in the k-th performance dimension. This represents the preset physiological state parameter weighting factor, reflecting the influence of the fish egg's physiological state on the coordination degree calculation. Its value can be set according to the fish egg's developmental stage, such as a higher weight for early stages (e.g., 0.8) and a lower weight for later stages (e.g., 0.5). This represents the basic performance parameter value of the j-th environmental factor unit in the k-th performance dimension. This represents the normalized value of the basic performance parameter of the j-th environmental factor unit in the k-th performance dimension, and n represents the total number of environmental factor units.

[0020] It should be noted that by calculating the matching degree between environmental factors based on normalization and Euclidean distance and combining physiological state distance weighting factors, a quantitative representation of the degree of synergy between units is realized, thereby generating an initial incubation response map. This enables the prediction model to identify natural synergistic pairs and inefficient combinations in the initial stage, providing engineering priors for subsequent action space construction and strategy initialization. Furthermore, it can accelerate model convergence, reduce the exploration of ineffective prediction combinations, and reduce energy consumption and stability loss caused by misjudgment of synergy.

[0021] Step S2: Preset incubation environment disturbance parameters and construct a dynamic disturbance label incubation response map; construct an incubation rate index matrix and match it with the incubation rate index matrix to generate a corrected dynamic disturbance label incubation response map.

[0022] Specifically, preset incubation environment disturbance parameters are used, including temperature fluctuation parameters, dissolved oxygen change parameters, and pollutant concentration parameters. The incubation environment disturbance parameters are introduced into all environmental factor units in the initial incubation response map and mapped to the unified feature identifier code to construct a dynamic disturbance label incubation response map. A hatching rate index matrix is ​​constructed, comprising hatching success rate, upper limit of deformity rate, and upper limit of energy consumption. The hatching response map of the dynamic disturbance tag is matched with the hatching rate index matrix, and the basic performance parameters of the environmental factor unit are corrected through weighted matching to generate a corrected dynamic disturbance tag hatching response map, as detailed below: ; ; ; in, This represents the temperature fluctuation parameter of the i-th environmental factor unit (i.e., the temperature deviation amplitude of the i-th environmental factor unit under water temperature disturbance). This represents the dissolved oxygen change parameter of the i-th environmental factor unit (i.e., the change in dissolved oxygen concentration of the i-th environmental factor unit). This represents the pollutant concentration perturbation parameter of the i-th environmental factor unit (i.e., the pollution load of the i-th environmental factor unit). This represents the preset environmental disturbance weighting coefficient. Indicates the hatching success rate. This indicates the upper limit of the deformity rate. Indicates the upper limit of energy consumption. This represents the disturbance correction coefficient for the i-th environmental factor unit. This represents the modified basic performance parameter value of the i-th environmental factor unit in the k-th performance dimension. This represents the degree of synergy between the i-th and j-th environmental factor units after modification. represents the modified dynamic disturbance label incubation response map matrix, where m represents the total number of environmental factor units.

[0023] It should be noted that by pre-setting incubation disturbance parameters (temperature fluctuations, dissolved oxygen changes, and pollutant concentrations) and mapping the disturbances to the initial incubation response map to form a dynamic disturbance label map, and then using a weighted matching correction with a hatching rate index matrix (hatching success rate, upper limit of deformity rate, and upper limit of energy), coupled modeling of disturbance scenarios and safety constraints is achieved. Embedding engineering safety boundaries into the scenario generation and parameter correction process before training / simulation and online decision-making allows for the screening of action combinations that could lead to exceeding limits or excessive energy consumption before strategy generation, and feeds real-world constraints back into subsequent prediction generation. Ultimately, this significantly reduces the probability of exceeding limits (deformity or excessive energy consumption) events, improves the robustness and verifiability of the prediction model in extreme or unseen environments, and facilitates risk verification and compliance testing.

[0024] Step S3: Based on the modified dynamic disturbance label incubation response map, construct an environmental factor action space set; introduce a multimodal prediction strategy into the environmental factor action space set, construct a prediction strategy scheduling matrix, and generate a multimodal prediction strategy distribution map.

[0025] Specifically, based on the corrected dynamic disturbance tag hatching response map, the temperature control accuracy parameters of the water temperature control unit and the periodic adjustment parameters of the light management unit are extracted; the action capability parameters are uniformly quantified to construct an environmental factor action space set; A multimodal prediction strategy is introduced into the environmental factor action space set. The multimodal prediction strategy includes a master prediction strategy, an auxiliary prediction strategy, and an emergency correction strategy. The multimodal prediction strategy is then bound to the unified feature identifier encoding to construct a prediction strategy scheduling matrix. It should be noted that multimodal prediction strategies refer to strategy modes with different preset control target priorities for different environmental disturbance intensities and fish egg physiological states. These strategies themselves are based on a pre-set rule framework of domain knowledge, but their specific parameters and the conditions under which they are triggered are learned by deep learning networks: Main control prediction strategy: activated when environmental disturbances are small and fish egg state is stable, the strategy objective is to optimize energy efficiency, and finely adjust environmental parameters to achieve the lowest energy consumption while ensuring hatching rate; Auxiliary prediction strategy: activated when increased disturbances are detected (such as increased temperature fluctuations) or slight deviations in fish egg development indicators, the strategy objective shifts to stability priority, and the control frequency and amplitude are appropriately increased to quickly offset the impact of disturbances; Emergency correction strategy: activated when severe disturbances are detected (such as equipment failure, sudden increase in pollutant concentration) or fish egg key physiological indicators (such as abnormal heart rate) deviate significantly from the safety threshold, the strategy objective is the highest priority of safety and survival rate, taking the maximum range of correction actions regardless of energy consumption, and may trigger alarms. The policy selection is not based on fixed rules, but rather by a deep learning network (specifically a policy network) that evaluates and outputs in real time the overall system state represented by the currently corrected dynamic perturbation label hatching response map. The network outputs a policy distribution probability, and the system selects the policy with the highest probability as the current dominant mode. This achieves an upgrade from "rule-based policy selection" to "learning-based state-policy mapping".

[0026] The prediction strategy scheduling matrix is ​​input into the deep learning network, and a multimodal prediction strategy distribution map is generated based on the correlation between physiological state parameters and disturbance intensity parameters among environmental factor units.

[0027] It should be noted that by extracting key action capability parameters (such as temperature control accuracy and cycle adjustment parameters) from the corrected dynamic disturbance labeling map and uniformly quantifying the action capabilities, an environmental factor action space is constructed. Three types of multimodal prediction strategies—main control, auxiliary, and emergency correction—are introduced and bound to a unified code to generate a prediction strategy scheduling matrix and a multimodal strategy distribution map, achieving hierarchical organization of the action space and structured strategy backup. Layering prediction strategies according to function and urgency level facilitates the priority use of efficient main control under normal conditions and automatic switching to auxiliary or correction strategies in the event of equipment degradation or emergencies. Simultaneously, it provides clear strategy boundaries for simulation verification, playback review, and fault diagnosis. This step improves the response speed and reliability to sudden disturbances, reduces the probability of prediction interruption due to single-point failures, and makes the strategies easier to verify in engineering and formally review.

[0028] Step S4: Based on the multimodal prediction strategy distribution map, map all environmental factor units in the corrected dynamic disturbance label incubation response map to the corresponding prediction strategy entries, construct a multi-factor collaborative strategy mapping matrix, and introduce regulation priority weights to generate a multi-factor collaborative regulation map.

[0029] Specifically, based on the multimodal prediction strategy distribution map, all environmental factor units in the corrected dynamic disturbance label hatching response map are mapped to the corresponding prediction strategy entries to construct a multi-factor collaborative strategy mapping matrix. The multi-factor collaborative strategy mapping matrix is ​​used to characterize the controllability of different environmental factor units under the prediction strategy. A regulation priority weight is introduced into the multi-factor collaborative strategy mapping matrix. This regulation priority weight represents the control attribute parameters in the hatching rate index matrix (e.g., in an emergency, the inherent priority of dissolved oxygen regulation may be preset to be higher than that of light regulation; the regulation priority weight can also be dynamically adjusted by a deep learning network according to the requirements of the current hatching rate index matrix and the system state reflected in the real-time dynamic disturbance tag hatching response map; for example, when an increased risk of deformity is detected, the weight related to temperature stability will be temporarily increased by the network). The regulation priority weight is then weighted and fused with the multi-factor collaborative strategy mapping matrix to construct a priority-weighted prediction table, as detailed below: ; ; in, This represents the response strength of the i-th environmental factor unit to the p-th strategy (main control, auxiliary, emergency). This represents the regulatory priority weight of the i-th environmental factor unit. This represents the weighted response strength of the i-th environmental factor unit to the p-th strategy. This represents the degree of synergy between the i-th and j-th environmental factor units after modification. This represents the comprehensive control value of the i-th environmental factor unit, where m represents the total number of environmental factor units; The priority-weighted prediction table is input into a deep learning network, and a multi-factor collaborative regulation map is generated based on the correlation between environmental factor units, denoted as... .

[0030] It should be noted that by mapping all environmental factors to prediction strategy entries in the revised response graph, a multi-agent prediction strategy correspondence matrix is ​​constructed, and prediction priority weights based on the hatching rate index are introduced. These are then merged into a priority-weighted prediction table, generating a multi-agent collaborative prediction graph. This achieves a collaborative decision-making framework based on controllability-importance ranking. This provides a quantitative basis for conflict detection, arbitration, and task allocation when multiple environmental factors act in parallel. The priority weights ensure that key hatching rate actions are executed first, while incorporating engineering indicators such as energy consumption constraints and development timelines into the decision-making logic. This step reduces the frequency of issuing conflict commands simultaneously, improves collaborative efficiency and safety margins, and maintains overall hatching stability when resources are limited or some equipment degrades.

[0031] Step S5: Based on the multi-factor collaborative control diagram, generate the final hatching optimization instruction sequence; send the final hatching optimization instruction sequence to the corresponding environmental control unit via the control system communication bus; generate an instant evaluation signal.

[0032] Specifically, based on the multi-factor collaborative regulation diagram, the collaborative prediction instruction set of all environmental factor units under the current incubation environment disturbance parameters is analyzed by utilizing the degree of synergy and regulation priority weights among different environmental factor units. The collaborative prediction instruction set is compared and verified with the priority-weighted prediction table. Instruction conflicts are identified using a conflict detection algorithm based on a rule base, and arbitration is performed according to the priority weights in the priority-weighted prediction table to generate the final incubation optimization instruction sequence. The rule base is a predefined logical set based on domain knowledge and system security constraints. Rule base construction description: Physically infeasible rule: for example, prohibiting the simultaneous execution of instructions for large temperature increase and large temperature decrease; Resource competition rule: for example, the total power consumption must not exceed the system safety limit; Physiological safety rule: for example, in the early stage of fish egg development, the rate of temperature change must not exceed a certain safety threshold; Control target conflict rule: for example, under the "optimal energy efficiency" strategy, if an instruction that consumes a large amount of energy to achieve a small temperature increase will conflict with the target, it may be marked.

[0033] The final incubation optimization instruction sequence is sent to the corresponding environmental control unit via the control system communication bus, as follows: The water temperature control unit fine-tunes the temperature, the dissolved oxygen regulation unit adjusts the oxygen supply rate, and the light management unit adjusts the light cycle. A sensor network deployed in the incubator collects real-time physiological parameters of the fish eggs after command execution, including actual developmental stage, actual heart rate, and changes in egg membrane permeability. These parameters are compared with the hatching rate index matrix. Based on the hatching success rate, upper limit of deformity rate, and upper limit of energy consumption, an instant evaluation signal is generated. This instant evaluation signal is calculated by weighting a developmental deviation penalty, an energy consumption penalty, and a health reward, as detailed below: ; ; ; ; in, Indicates the target developmental stage indicators. This indicates the actual developmental stage indicators detected. This indicates a penalty for developmental deviation. This represents the actual energy consumption value. Indicates the maximum allowable energy consumption. This indicates an energy consumption penalty item. This indicates actual health indicators (such as heart rate stability and oocyte membrane integrity). This represents the theoretical maximum health value. This indicates a health reward item. This represents the weighting coefficients for rewards and punishments. This indicates an immediate assessment signal.

[0034] It should be noted that a collaborative prediction instruction set is obtained by parsing the multi-agent collaborative prediction graph. After verification and arbitration using a conflict detection algorithm based on a rule base against a priority-weighted prediction table, a final hatching optimization instruction sequence is generated and sent to the environmental control unit in real time via the communication bus. Simultaneously, the physiological state after execution is collected using the incubator's sensor network, and real-time evaluation signals are calculated based on developmental deviation, energy consumption, and health rewards, achieving a closed loop of execution-perception-evaluation. An engineered conflict check is added before strategy issuance to ensure that commands are executable and do not violate safety constraints. The execution results are instantly converted into deep learning feedback signals for online model optimization. This step avoids issuing unexecutable or dangerous commands, improves execution transparency and auditability, and guides the prediction model towards energy-efficient optimization and rapid increase in hatching rate through real-time evaluation.

[0035] Step S6: Based on the final incubation optimization instruction sequence, the current incubation environment perturbation parameters, the new physiological state parameters, and the instantaneous evaluation signal, construct an experience tuple; update the prediction network and evaluation network parameters of the deep learning network, and provide feedback for adaptive prediction and adjustment.

[0036] Specifically, based on the final incubation optimization instruction sequence, the current incubation environment perturbation parameters, the new physiological state parameters, and the real-time evaluation signal, an experience tuple is constructed, and the experience tuple is stored in the experience replay pool of the deep learning network based on a first-in-first-out strategy. When the training cycle is triggered, a small batch of experience tuples is randomly sampled from the experience replay pool. Using a deep learning algorithm, the parameters of the prediction network and the evaluation network of the deep learning network are updated with the goal of minimizing the mean square error between the prediction evaluation and the target evaluation. The updated evaluation network parameters are then hot-updated online to the predictive generation engine deployed in the control system, providing feedback and enabling adaptive prediction and adjustment.

[0037] It should be noted that by forming an empirical tuple from the final instruction sequence, current perturbation, new physiological state, and immediate evaluation, and storing it in the empirical replay pool according to a first-in-first-out strategy, periodically sampling small batches of data to minimize the mean squared error and updating the prediction and evaluation network, and then hot-updating it online to the deployment engine and feeding it back to step S3, a closed loop of online-offline hybrid training and adaptive deployment is achieved. This process maintains the continuity of on-site prediction while completing model learning and parameter updates, enabling the system to continuously adapt in non-stationary environments, equipment aging, or changing operating conditions, reducing reliance on manual parameter retuning. This step improves long-term robustness and adaptability, shortens recovery time from distress states, reduces maintenance and debugging costs, and achieves rapid iteration and validation of model quality through the hot-update mechanism.

[0038] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0039] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A deep learning-based method for predicting and optimizing fish egg hatching rates, characterized in that, The method includes the following steps: Step S1: Collect basic performance parameters of different environmental factor units from the back end of the fish egg hatching control system and construct a multi-source state feature set; Step S2: Preset incubation environment disturbance parameters and construct a dynamic disturbance label incubation response map; construct an incubation rate index matrix and match it with the incubation rate index matrix to generate a corrected dynamic disturbance label incubation response map; Step S3: Based on the corrected dynamic disturbance label incubation response map, construct an environmental factor action space set; introduce a multimodal prediction strategy into the environmental factor action space set, construct a prediction strategy scheduling matrix, and generate a multimodal prediction strategy distribution map; Step S4: Based on the multimodal prediction strategy distribution map, map all environmental factor units in the corrected dynamic disturbance label incubation response map to the corresponding prediction strategy entries, construct a multi-factor collaborative strategy mapping matrix, and introduce regulation priority weights to generate a multi-factor collaborative regulation map. Step S5: Based on the multi-factor collaborative control diagram, generate the final hatching optimization instruction sequence; send the final hatching optimization instruction sequence to the corresponding environmental control unit via the control system communication bus; generate an instant evaluation signal; Step S6: Based on the final incubation optimization instruction sequence, the current incubation environment perturbation parameters, the new physiological state parameters, and the instantaneous evaluation signal, construct an experience tuple; update the prediction network and evaluation network parameters of the deep learning network, and provide feedback for adaptive prediction and adjustment.

2. The deep learning-based fish egg hatching rate prediction and optimization method according to claim 1, characterized in that, The specific implementation process of step S1 includes: Basic performance parameters of different environmental factor units are collected from the back end of the fish egg hatching control system. The environmental factor units include a water temperature control unit, a dissolved oxygen regulation unit, and a light management unit. The basic performance parameters of the water temperature control unit include heating power, temperature control accuracy, and energy consumption parameters. The basic performance parameters of the dissolved oxygen regulation unit include oxygen supply rate, dissolved oxygen stability, and response time. The basic performance parameters of the light management unit include light intensity range, spectral composition, and periodic adjustment parameters.

3. The deep learning-based fish egg hatching rate prediction and optimization method according to claim 2, characterized in that, The specific implementation process of step S1 also includes: Based on the basic performance parameters of the water temperature control unit, the dissolved oxygen regulation unit, and the light management unit, a multi-source state feature set is constructed, as follows: Numbering is performed according to a unified feature identification code, which includes a three-segment combination code consisting of a unit category prefix, device number, and performance serial number. A correspondence is established based on the physiological state parameters and control attribute parameters of the environmental factor unit. The physiological state parameters include the developmental stage of the fish egg, heart rate variability, and egg membrane permeability. The control attribute parameters include the regulation mode, available time period, and action priority of the environmental factor unit. The multi-source state feature set is input into a deep learning network. An initial incubation response map is generated based on the correlation between environmental factor units. The correlation between environmental factor units is calculated by normalization and Euclidean distance to determine the matching degree of basic performance parameters and control attribute parameters between environmental factor units. The degree of synergy between different environmental factor units is calculated by weighting the results with a preset distance weight factor for physiological state parameters, as follows: ; in, This represents the degree of synergy between the i-th environmental factor unit and the j-th environmental factor unit. This represents the basic performance parameter value of the i-th environmental factor unit in the k-th performance dimension. This represents the normalized value of the basic performance parameter of the i-th environmental factor unit in the k-th performance dimension. This represents the preset physiological state parameter weighting factor. This represents the basic performance parameter value of the j-th environmental factor unit in the k-th performance dimension. This represents the normalized value of the basic performance parameter of the j-th environmental factor unit on the k-th performance dimension, where n represents the total number of performance dimensions.

4. The deep learning-based fish egg hatching rate prediction and optimization method according to claim 3, characterized in that, The specific implementation process of step S2 includes: Preset incubation environment disturbance parameters, including temperature fluctuation parameters, dissolved oxygen change parameters, and pollutant concentration parameters; introduce the incubation environment disturbance parameters into all environmental factor units in the initial incubation response map, and correspond them with the unified feature identifier code to construct a dynamic disturbance label incubation response map; A hatching rate index matrix is ​​constructed, comprising hatching success rate, upper limit of deformity rate, and upper limit of energy consumption. The hatching response map of the dynamic disturbance tag is matched with the hatching rate index matrix, and the basic performance parameters of the environmental factor unit are corrected through weighted matching to generate a corrected dynamic disturbance tag hatching response map, as detailed below: ; ; ; in, This represents the temperature fluctuation parameter of the i-th environmental factor unit. This represents the dissolved oxygen change parameter for the i-th environmental factor unit. This represents the pollutant concentration perturbation parameter of the i-th environmental factor unit. This indicates the preset environmental disturbance weighting coefficient. Indicates the hatching success rate. This indicates the upper limit of the deformity rate. Indicates the upper limit of energy consumption. This represents the disturbance correction coefficient for the i-th environmental factor unit. This represents the modified basic performance parameter value of the i-th environmental factor unit in the k-th performance dimension. This represents the degree of synergy between the i-th and j-th environmental factor units after modification. represents the modified dynamic disturbance label incubation response map matrix, where m represents the total number of environmental factor units.

5. The deep learning-based method for predicting and optimizing fish egg hatching rate according to claim 4, characterized in that, The specific implementation process of step S3 includes: Based on the corrected dynamic disturbance tag hatching response map, the temperature control accuracy parameters of the water temperature control unit and the periodic adjustment parameters of the light management unit are extracted; the action capability parameters are uniformly quantified to construct an environmental factor action space set. A multimodal prediction strategy is introduced into the environmental factor action space set. The multimodal prediction strategy includes a master prediction strategy, an auxiliary prediction strategy, and an emergency correction strategy. The multimodal prediction strategy is then bound to the unified feature identifier encoding to construct a prediction strategy scheduling matrix. The prediction strategy scheduling matrix is ​​input into the deep learning network, and a multimodal prediction strategy distribution map is generated based on the correlation between physiological state parameters and disturbance intensity parameters among environmental factor units.

6. The deep learning-based method for predicting and optimizing fish egg hatching rate according to claim 5, characterized in that, The specific implementation process of step S4 includes: Based on the multimodal prediction strategy distribution map, all environmental factor units in the corrected dynamic disturbance label hatching response map are mapped to the corresponding prediction strategy entries to construct a multi-factor collaborative strategy mapping matrix. The multi-factor collaborative strategy mapping matrix is ​​used to characterize the controllability of different environmental factor units under the prediction strategy. A regulatory priority weight is introduced into the multi-factor collaborative strategy mapping matrix. The regulatory priority weight represents the control attribute parameter in the hatching rate index matrix. The regulatory priority weight is then weighted and fused with the multi-factor collaborative strategy mapping matrix to construct a priority-weighted prediction table, as follows: ; ; in, This represents the response strength of the i-th environmental factor unit to the p-th strategy. This represents the regulatory priority weight of the i-th environmental factor unit. This represents the weighted response strength of the i-th environmental factor unit to the p-th strategy. This represents the degree of synergy between the i-th and j-th environmental factor units after modification. This represents the comprehensive control value of the i-th environmental factor unit, where m represents the total number of environmental factor units; The priority-weighted prediction table is input into a deep learning network, and a multi-factor collaborative regulation map is generated based on the correlation between environmental factor units, denoted as... .

7. The deep learning-based fish egg hatching rate prediction and optimization method according to claim 6, characterized in that, The specific implementation process of step S5 includes: Based on the multi-factor collaborative regulation diagram, the collaborative prediction instruction set of all environmental factor units under the current incubation environment disturbance parameters is analyzed by utilizing the degree of synergy and regulation priority weights among different environmental factor units. The collaborative prediction instruction set is compared and verified with the priority weighted prediction table. Instruction conflicts are identified by a rule-based conflict detection algorithm, and arbitration is performed according to the priority weights in the priority weighted prediction table to generate the final incubation optimization instruction sequence. The final incubation optimization instruction sequence is sent to the corresponding environmental control unit via the control system communication bus, as follows: The water temperature control unit fine-tunes the temperature, the dissolved oxygen regulation unit adjusts the oxygen supply rate, and the light management unit adjusts the light cycle. By deploying a sensor network in the incubator, new physiological state parameters of fish eggs are collected in real time after the command is executed, including actual developmental stage, actual heart rate and changes in egg membrane permeability. These parameters are then compared with the hatching rate index matrix. An instant evaluation signal is generated based on the hatching success rate, upper limit of deformity rate and upper limit of energy consumption. The instant evaluation signal is calculated by weighting the developmental deviation penalty, energy consumption penalty and health reward.

8. The deep learning-based method for predicting and optimizing fish egg hatching rate according to claim 7, characterized in that, The specific implementation process of step S6 includes: Based on the final incubation optimization instruction sequence, the current incubation environment perturbation parameters, the new physiological state parameters, and the real-time evaluation signal, an experience tuple is constructed, and the experience tuple is stored in the experience replay pool of the deep learning network based on a first-in-first-out strategy. When the training cycle is triggered, a small batch of experience tuples is randomly sampled from the experience replay pool. Using a deep learning algorithm, the parameters of the prediction network and the evaluation network of the deep learning network are updated with the goal of minimizing the mean square error between the prediction evaluation and the target evaluation. The updated evaluation network parameters are then hot-updated online to the predictive generation engine deployed in the control system and fed back to step S3 for adaptive prediction and adjustment.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the deep learning-based fish egg hatching rate prediction and optimization method as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the deep learning-based fish egg hatching rate prediction and optimization method as described in any one of claims 1 to 8.

Images