Feed efficiency optimization system based on growth model

Abstract

The invention relates to the technical field of feed efficiency optimization, in particular to a feed efficiency optimization system based on a growth model, and the system comprises a data collection unit which is used for collecting and associating individual identities, individual weight data, time-phased feed intake data, environment data and environment forecast data; the model calibration unit is used for calibrating individual energy conversion efficiency parameters, generating an environment dynamic correction factor by using the environment data and generating a group classification label based on historical individual performance data; and the decision and execution unit operates a double-cycle cooperation mechanism based on pre-judgment and execution. By dynamically calibrating the individual energy conversion efficiency and combining with the environment forecast, real-time calculation and differentiated feeding regulation and control of the individual energy demand can be realized.

Description

Technical Field

[0001] This application relates to the technical field of feed efficiency optimization, and in particular to a feed efficiency optimization system based on a growth model. Background Technology

[0002] Optimizing feed efficiency is a crucial aspect of large-scale livestock and poultry farming, directly impacting breeding costs, animal health, and environmental sustainability. In the farming of economically important animals such as pigs, feed costs constitute a significant proportion of total costs, especially during the fattening stage, where even small improvements in feed conversion ratio can yield substantial economic benefits. Therefore, how to improve feed utilization efficiency through precise nutrient supply and feeding management has consistently been a focus of research and development in livestock technology and equipment.

[0003] Currently, common feed efficiency optimization systems are typically built upon animal growth models. These systems generally adjust feed formulations periodically by periodically collecting data on the average body weight and total feed intake of the herd, combined with static or semi-static nutritional requirement models. Their technical approach generally relies on historical herd average data to build predictive models, with model parameters often using fixed values ​​or undergoing manual calibration and updates over long periods. In ideal scenarios where environmental conditions are relatively stable and animal herd behavior is highly consistent, such methods can guide feed formulation to a certain extent.

[0004] However, existing technologies cannot respond in real-time and accurately to the differentiated feeding behaviors and energy demands of individual animals in dynamic environments. For example, in stressful production scenarios such as high temperature and humidity in summer, where indoor temperatures consistently exceed 30°C and relative humidity surpasses 60%, pigs exhibit typical temporal differentiation in feeding behavior: animals tend to concentrate their feeding during the relatively cooler morning and evening hours, while feeding is significantly inhibited during the midday heat. Simultaneously, there are significant differences in individual tolerance to heat stress; some individuals (sensitive types) experience a much larger decrease in feed intake than the group average, while others (tolerant types) show only a slight decrease. However, traditional systems relying on historical average group data have long data update cycles and cannot capture these individual-specific feeding dynamics that change on an hourly or even minute-by-minute basis. Due to the lack of synchronous perception and correlation analysis of multi-dimensional data across "individual-time-environment," the system cannot identify the true energy surplus or deficit status of individuals with different tolerance types. This leads to decisions based on inaccurate model outputs, resulting in a one-size-fits-all group adjustment strategy. The consequence is that sensitive individuals experience stunted growth due to insufficient energy intake, while tolerant individuals may waste feed due to relative energy surplus, leading to a decrease in overall feed conversion ratio. Furthermore, the entire chain, from data collection to model calibration, and then to feed formulation adjustment and execution, contains multiple delays, and the system's response speed is much slower than the rate of change in the environment and animal behavior, severely limiting the practicality and reliability of feed efficiency optimization systems in real, dynamic production environments. Summary of the Invention

[0005] In order to achieve real-time calculation of individual energy demand and differentiated feeding control by dynamically calibrating individual energy conversion efficiency and combining environmental forecasts, this application provides a feed efficiency optimization system based on a growth model.

[0006] This application provides a feed efficiency optimization system based on a growth model, which adopts the following technical solution: Optional, including: The data acquisition unit is used to collect and associate individual identification data, individual weight data, food intake data by time period, environmental data, and environmental forecast data. The model calibration unit, connected to the data acquisition unit, is used to calibrate the individual energy conversion efficiency parameters based on the individual weight data and the food intake data at different time periods, generate an environmental dynamic correction factor using the environmental data, and generate a group classification label based on historical individual performance data. The decision-making and execution unit connects the data acquisition unit and the model calibration unit, and operates based on a dual-cycle collaborative mechanism of prediction and execution, wherein: The prediction cycle generates pre-production instructions for differentiated feed formulations based on the environmental forecast data and the population classification tags. The execution cycle responds to the individual's identity identifier and, in conjunction with the environmental data, the calibrated individual energy conversion efficiency parameters, and the environmental dynamic correction factor, calculates the individual's real-time energy demand and controls differentiated energy delivery.

[0007] Optionally, the data acquisition unit divides the daily feeding time into multiple logical time periods, and within each logical time period, associates the individual's identity with its feeding behavior and corresponding feed intake data to obtain the time-segmented feed intake data.

[0008] Optionally, the model calibration unit calibrates the individual energy conversion efficiency parameter in the following ways: calculating the individual's actual weight gain based on the individual's weight data, and calculating the individual's actual metabolizable energy intake based on the time-segmented food intake data; based on the principle of energy balance, calculating the current calibration value of the individual's energy conversion efficiency parameter by comparing the difference between the individual's actual metabolizable energy intake and the metabolizable energy required to maintain the individual's actual weight gain; and using a data smoothing algorithm to fuse the current calibration value with the historical value of the individual's energy conversion efficiency parameter to achieve dynamic updating of the individual's energy conversion efficiency parameter.

[0009] Optionally, the model calibration unit generates the environmental dynamic correction factor by: identifying a specific environmental data interval based on the environmental data collected by the data acquisition unit, and calculating the actual average feed intake of the population within the specific environmental data interval; comparing the actual average feed intake with the expected average feed intake under predefined baseline environmental conditions, and generating the environmental dynamic correction factor based on the ratio.

[0010] Optionally, the execution cycle of the decision-making and execution unit includes: calculating the individual basic energy demand based on the calibrated individual energy conversion efficiency parameter; when the environmental data or environmental forecast data exceeds a preset environmental threshold, the environmental dynamic correction factor is invoked to adjust the individual basic energy demand to obtain the individual real-time energy demand for control of energy delivery.

[0011] Optionally, the step of the model calibration unit generating the group classification label includes: calculating the individual performance evaluation result corresponding to each individual identity based on the periodically collected individual weight data and the time-segmented food intake data; comparing the individual performance evaluation result with the group performance evaluation benchmark; and classifying individuals into different types according to the comparison result to form the dynamically updated group classification label.

[0012] Optionally, when the decision-making and execution unit calculates the real-time energy requirement of the individual during the execution cycle, it also matches different target feed energy concentrations for different types of individuals based on the individual type indicated in the group classification label and a predefined feeding strategy, so as to perform differentiated feeding.

[0013] Optionally, a feedback optimization unit may also be included, which is connected to the model calibration unit and the decision and execution unit, and is used for: Based on the group classification labels, the rate of change of the core performance indicators of the corresponding group types is periodically analyzed; The rate of change of the core performance indicators is compared with a preset improvement threshold; Based on the comparison results, the adjustment range of the target feed energy concentration in the predefined feeding strategy is automatically determined and updated.

[0014] Optionally, the feedback optimization unit is further configured to: The individual types classified based on the group classification labels and their associated group performance data are fed back to the model calibration unit; In response to the feedback, the model calibration unit applies differentiated parameter calibration weights to different types of individuals to update the individual energy conversion efficiency parameters, and periodically updates the environmental dynamic correction factor based on the feedback group performance data.

[0015] In summary, this application includes the following beneficial technical effects: This system achieves real-time calculation and differentiated feeding control of individual energy requirements by dynamically calibrating individual energy conversion efficiency parameters and combining them with environmental forecast data. It can accurately respond to the differentiated feeding behavior and energy demand changes of individuals in dynamic environments, effectively solving the feeding lag and one-size-fits-all problems caused by traditional systems that rely on population average data. This avoids the growth stunting of sensitive individuals due to insufficient energy and the feed waste caused by excessive energy in tolerant individuals, thus significantly improving feed utilization efficiency.

[0016] The system adopts a dual-cycle collaborative mechanism of prediction and execution. Based on environmental forecasts and population classification, it generates differentiated feed formulas in advance and combines real-time environmental data and calibration parameters to control precise delivery during the execution cycle. This significantly improves feed conversion rate and feeding response speed, reduces overall feed costs, and brings significant economic benefits to large-scale farming.

[0017] By periodically analyzing population performance data and autonomously adjusting feeding strategies and model parameters through feedback optimization units, the system achieves closed-loop optimization, continuously adapting to individual differences and environmental changes. This solves the problem of long-term accuracy decline caused by the lack of dynamic iteration in traditional systems, ensuring the reliability and practicality of the system in real production environments. Attached Figure Description

[0018] Figure 1 It is the overall closed-loop optimization logic diagram of the system. Detailed Implementation

[0019] The following is in conjunction with the appendix Figure 1 This application will be described in further detail.

[0020] This application discloses a feed efficiency optimization system based on a growth model. For example... Figure 1 As shown, a feed efficiency optimization system based on a growth model includes a data acquisition unit, a model calibration unit, a decision-making and execution unit, and a feedback optimization unit. These four units work collaboratively to achieve real-time calculation of individual energy requirements and differentiated feeding control. Specifically, this application includes the following steps: S1 performs multi-source data acquisition and correlation. The data acquisition unit collects individual identification data, individual weight data, food intake data at different times, environmental data, and environmental forecast data, and establishes the correlation between the various data to provide accurate and complete basic data sources for the subsequent model calibration unit.

[0021] S11 collects individual identification and weight data. The system equips each pig with a unique high-frequency RFID ear tag, ensuring the uniqueness and stability of individual identification and adapting to the needs of individual traceability and data association in large-scale farming scenarios. Automatic weighing stations are installed at high-frequency passageways such as watering points and rest area exits in the pigsty. The weight sensors at these stations have an accuracy of ±0.1kg, meeting the accuracy requirements of weighing equipment in the livestock industry. The system sets 05:00-07:00 daily as the standard weighing window. Pilot data from 300 fattening pigs shows that during this period, pigs are generally in a fasting state, with weight data fluctuations within ±0.2kg, and an effective weighing rate of over 95%, ensuring that each pig receives at least one effective weight data (in kg) daily. When a pig passes through the automatic weighing station, the RFID reader automatically reads the identification information from the individual's ear tag, and the automatic weighing station simultaneously records the individual's weight data for the day. ( (Represents the collection date, unit: kg). The system will upload the individual's identity information and corresponding weight data to the cloud platform in real time. The cloud platform will then store the weight data according to individual identity identifiers, forming an individual weight database.

[0022] S12 collects feed intake data by time period. The system divides daily feeding times based on the feeding behavior patterns of individual pigs, specifically into three logical time slots: 06:00-09:00 in the morning, 11:00-14:00 in the afternoon, and 17:00-20:00 in the evening. A pilot experiment with 300 fattening pigs in a high-temperature summer environment showed that this time slot division accurately captures individual feeding differences under different temperatures, with a data capture rate exceeding 92% for each time slot. To handle feeding behavior outside the designated time slots, the system automatically assigns feeding records to the most recent logical time slot. The system deploys smart feed troughs in the feeding area corresponding to each logical time slot. The built-in weight sensors in the smart feed troughs have an accuracy of ±1g, meeting the requirement for accurate feed intake measurement (unit: kg). When an individual animal enters the smart feed trough feeding area, the RFID reader in the smart feed trough reads the individual's identification tag, and the weight sensor records the weight of the feed in the trough before feeding and the weight of the remaining feed in the trough after feeding. The system calculates the difference between the two weights to obtain the individual's feed intake data for the current time period, which is recorded as morning feed intake, midday feed intake, and evening feed intake. The system automatically accumulates the individual's feed intake for each time period of the day to obtain the individual's total feed intake for the day. All feed intake data for each time period and the corresponding individual identification tag are uploaded to the cloud platform in real time. The cloud platform associates and stores the feed intake data for each time period with the individual identification tag, enabling precise tracking of the individual's feeding behavior.

[0023] S13 collects environmental data and environmental forecast data. The system deploys temperature and humidity transmitters evenly throughout the pigsty at a density of one unit per 50 square meters, ensuring comprehensive and representative environmental data collection. Each transmitter collects real-time temperature (°C) and relative humidity (%) data per minute. The temperature measurement range is -20°C to 60°C with an accuracy of ±0.5°C, and the relative humidity measurement range is 0% to 100% with an accuracy of ±3%. This accuracy meets the needs of assessing stress levels in the breeding environment. Pilot data from a 300-head fattening pigsty shows that this collection frequency effectively captures temperature and humidity fluctuations, preventing the omission of environmental changes due to excessively long collection intervals. The system connects to the China Meteorological Data Network via API. The environmental forecast data provided by this platform complies with national meteorological data service standards. The system acquires hourly environmental forecast data for the next 24 hours daily at 18:00, focusing on extracting temperature forecast data with an accuracy to one hour. This data acquisition at this time point provides forward-looking support for feed production decisions for the next forecast period, avoiding decision-making delays caused by late forecast data acquisition.

[0024] S14 Integrating Multi-Dimensional Data The system utilizes a relational database (such as MySQL) built on a cloud platform. This mature technology offers stable data storage and efficient relational queries, meeting the needs of multi-dimensional data integration. Using millisecond-level timestamps as indexes, the system correlates individual identification data, individual weight data, time-segmented food intake data, real-time environmental data, and environmental forecast data, effectively avoiding matching errors caused by scattered data storage in traditional systems. The system stores the correlated data in a structured manner, ensuring traceability and correlation, providing a comprehensive and accurate data source for subsequent model calibration units, and supporting the accuracy of model parameter calibration and population classification label generation.

[0025] S2 Execution Parameter Calibration and Label Generation The model calibration unit is connected to the data acquisition unit. Relying on the multi-dimensional data collected by S1, it dynamically calibrates individual energy conversion efficiency parameters, dynamic correction factors of the generation environment, and group classification labels, providing accurate model support for subsequent decision-making.

[0026] S21 calibrates individual energy conversion efficiency parameters The system first calculates the individual's actual weight gain for the day. It extracts the individual's weight data for the day from S11. Weight the day before ,use minus Obtain the individual's actual weight gain on that day .

[0027] The system then calculates the individual's actual metabolizable energy intake. It obtains the food intake data by time period from S12, first adding up the food intake in the morning, noon, and evening to obtain the total food intake for the day. (unit: kg), combined with the basal metabolizable energy concentration of the feed. (Default is 3200kcal / kg), using Multiply Obtain the individual's actual daily metabolizable energy intake (Unit: kcal / day). This calculation is based on the universally accepted principle in feed nutrition: "Metabolic energy intake = feed intake × feed metabolizable energy concentration". In practical applications, the calculation error is less than 5%.

[0028] The system then calibrates individual energy conversion efficiency parameters based on the energy balance principle. The energy balance principle states that an individual's metabolic energy intake (…) A portion is used to maintain basic life activities. The remaining portion (i.e., the net energy used for growth) multiplied by the individual energy conversion efficiency parameter ( After that, it equals the energy actually deposited as body weight ( Based on this, establish the following relation: Among them, maintaining metabolic constant Values ; To calculate the average weight for that day, use ( Dividing by 2 (in kg) can balance the impact of daily weight fluctuations on metabolic maintenance. Actual energy deposited. (Unit: kcal / day), of which Multiplying by 1000 converts the weight gain unit from kg to g to match the weight gain energy coefficient. (The value is 5.5 kcal / g, which is a nutritionally accepted constant representing the energy required to gain 1g of body weight.) The dimensionless energy conversion efficiency parameter for the individual to be calibrated reflects the efficiency with which the individual converts the net energy used for growth into body weight deposition.

[0029] The system retrieves the current calibration value by rearranging the above relationships, transforming the formula into: This yields the current calibrated value (dimensionless) of the individual energy conversion efficiency parameter. To balance data stability and real-time performance, the system employs an exponentially weighted moving average algorithm for smooth updates, as shown in the following equation: .in The weighting coefficient is 0.3 (this value is an exemplary value determined based on pilot data and can be adjusted for actual applications). This value has been verified by simulation of 100 sets of fattening pig data, which can both preserve the stability of historical parameters and make the current calibration value reasonable. This is the historical value (parameter from the previous round); This is the updated parameter value.

[0030] S22 Generation Environment Dynamic Correction Factor The system first divides specific environmental data intervals. It obtains real-time environmental data from S13 and divides the environmental data into several intervals (such as temperature 30-32℃ and humidity 60%-70%), with temperature every 2℃ and humidity every 10%.

[0031] Next, the system calculates the actual average feed intake of the population within a specific environmental data interval. During the continuous monitoring period of each environmental interval, the system extracts the feed intake data of all individuals within that interval (from the time-segmented feed intake summation results of S12), and calculates the average to obtain the actual average feed intake of the population. (Unit: kg). This statistical method can eliminate the interference of individual occasional differences in feeding on environmental impact assessment, making the results more representative.

[0032] The system then sets baseline environmental conditions and expected average feed intake. A temperature of 20-25℃ and a relative humidity of 50%-60% are predefined as baseline environmental conditions. Under these conditions, the feed intake of fattening pigs is stable and is recognized in the industry as a suitable growth environment. The average feed intake of 300 fattening pigs under these conditions fluctuates by less than ±3%. Therefore, the system sets the group average feed intake under these conditions as the expected average feed intake. (Unit: kg)

[0033] Finally, the system generates an environmental dynamic correction factor, using the actual average feed intake of the population. Divide by the expected average feed intake Environmental dynamic correction factors were obtained. (Dimensionless). When When the value is less than 1, it indicates that the population's feed intake is lower than the baseline environment under the current conditions. The correction factor can quantify the degree of inhibition of feed intake by the environment and provide data support for subsequent energy demand adjustments.

[0034] S23 generates group classification labels The system first calculates the individual's weekly performance evaluation results, with a 7-day evaluation cycle. Daily weight data is obtained from S11 within the cycle, and daily food intake data by time period is obtained from S12. The total daily food intake within the cycle is accumulated to obtain the individual's total weekly food intake. (Unit: kg); Subtract the initial weight from the final weight at the end of the cycle to obtain the individual's total weekly weight gain. (Unit: kg); then use Divide by Calculate the individual weekly feed conversion ratio (Dimensionless, reflecting feed utilization efficiency), using Divide by 7 to calculate the individual's average daily weight gain per week. (Unit: kg / day, reflecting growth rate) These two indicators are the core standards for evaluating the performance of farmed individuals in the industry.

[0035] Next, the system calculates the population performance evaluation benchmark, and statistically analyzes the weekly feed conversion ratio and weekly average daily weight gain of all individuals in the same batch. The average of these values ​​is then used to obtain the population's weekly average feed conversion ratio. (Dimensionless) and the average daily weight gain per week of the population (Unit: kg / day). This benchmark represents the general performance level of the same batch of populations, avoiding the impact of individual individual data bias on classification results.

[0036] Finally, the system generates population classification labels, compares individual performance evaluation results with population performance evaluation benchmarks, and classifies individual types: sensitive individuals have a weekly feed conversion ratio greater than [missing value]. +0.3 and the weekly average daily weight gain is less than These individuals show a significant decrease in feed intake under environmental stress and are prone to insufficient energy requirements; the weekly feed conversion ratio of tolerant individuals is less than that of... -0.2 and the weekly average daily increase is greater than or equal to These individuals exhibit stable feed intake and low energy requirements under environmental stress; the remaining individuals are of the stable type. The classification threshold was determined based on statistical analysis of data from a pilot project of 300 fattening pigs (for illustrative purposes only). After classification, the performance differences between different types of individuals were significant, with a classification accuracy exceeding 85%. The system updates the group classification labels every 7 days based on the latest periodic data to ensure that the labels can quickly respond to dynamic changes in individual performance; while more macro-level feeding strategy optimization is carried out over a longer period (see S41) to ensure the stability and statistical significance of the adjustments.

[0037] S3 executes dual-cycle collaborative decision-making and deployment. The decision-making and execution unit connects the data acquisition unit and the model calibration unit. Relying on the environmental data and environmental forecast data collected by S1 and the individual energy conversion efficiency parameters, environmental dynamic correction factors, and population classification labels calibrated by S2, it realizes the pre-production of differentiated feed formulas and precise individual feeding through a dual-cycle collaborative mechanism of prediction and execution. This solves the problems of lagging feed adjustment and "one-size-fits-all" group feeding in traditional systems, and improves feeding accuracy and response speed.

[0038] S31 predicts the cycle and generates differentiated feed formulation pre-production instructions. The system reads hourly temperature forecast data for the next 24 hours from S13 at 18:00 daily, identifying high-temperature warning periods (28-32℃) and extreme high-temperature periods (≥32℃). Pilot data from 300 fattening pigs shows that individual feed intake decreases by 12%-15% and 18%-22% respectively during these two periods, requiring targeted adjustments to feed energy concentration to compensate for the energy deficit.

[0039] The system synchronously calculates the proportion of sensitive, tolerant, and stable individuals in the population classification tags generated by S23. For example, in a certain batch of population, sensitive individuals account for 30%, tolerant individuals account for 20%, and stable individuals account for 50%. This proportion data directly determines the production demand of different formula feeds, avoiding feed stockpiling or shortages caused by blind production.

[0040] The system matches target feed energy concentrations based on predefined feeding strategies. The strategy formulation is based on a nutritional requirement experiment with 300 fattening pigs: During periods of extreme high temperatures, sensitive pigs, due to the most significant decrease in feed intake, require a feed energy concentration of 3500 kcal / kg; tolerant pigs, with less of a decrease in feed intake, only need 3300 kcal / kg; and stable pigs require 3400 kcal / kg to balance energy supply and waste. During periods of high-temperature warnings, the target feed energy concentrations for each type of pig are reduced by 100 kcal / kg: 3400 kcal / kg for sensitive pigs, 3200 kcal / kg for tolerant pigs, and 3300 kcal / kg for stable pigs. This adjustment range adapts to differences in the degree of environmental stress.

[0041] The system combines the proportion of individual feed intake with the average daily feed consumption per head (pilot data shows that the average daily feed consumption of fattening pigs is approximately 2.5-3.0 kg) to calculate the total production volume of each feed formula, generating differentiated feed formula pre-production instructions and sending them to the feed production unit. These instructions ensure that feed production is completed ahead of schedule and delivered to the intelligent feed silos in the pigsties before 4:00 AM the following day. The feed silos are divided into zones according to formula type to avoid mixing feeds of different energy concentrations, preparing for precise delivery during the execution cycle. Compared to the 1-2 day adjustment cycle of traditional systems, this process reduces feed preparation lag time to less than 10 hours.

[0042] S32 executes the cycle to calculate individual real-time energy needs and distribute energy accordingly. The system first calculates the individual's basic energy requirement, and based on the individual's energy conversion efficiency parameters calibrated by S21, establishes a calculation relationship according to the principle of energy balance: .in, To maintain metabolic constant ( ), The individual's weight (in kg) collected for S11 on the same day; The target daily weight gain was set at 600g / day, a value based on reasonable targets for the growth stage of fattening pigs. In the pilot project, this target resulted in the optimal combination of individual growth rate and feed efficiency. The weight gain energy coefficient (5.5 kcal / g, a nutritionally accepted constant) is multiplied by the product of the two, and the result is in kcal / day, which is consistent with the dimension of energy required to maintain metabolism. This is the calibrated individual energy conversion efficiency parameter. The individual basic energy requirement (unit: kcal / day) calculated using this relationship directly reflects the individual's energy requirement baseline under suitable conditions.

[0043] The system monitors real-time temperature and relative humidity data collected by S13. When the real-time temperature is ≥28℃ or the relative humidity is ≥60%, an environmental correction mechanism is triggered. These two thresholds were determined based on a stress experiment with 300 fattening pigs. Under these environmental conditions, the average feed intake of individuals decreased by more than 15%, requiring compensation for the energy deficit through correction factors. The system calls upon the dynamic environmental correction factor generated by S22. (generally <1, representing feeding inhibition), through the relational formula Adjustments yield the individual's real-time energy requirement (unit: kcal / day). This formula implies that when feeding is suppressed ( When <1), energy supply needs to be increased to compensate for insufficient intake.

[0044] The system reads the individual type tag generated by S23 and matches the corresponding target feed energy concentration to different types of individuals, ensuring that the energy supply is adapted to the individual's tolerance characteristics. When an individual passes through the smart feed trough area, the RFID reader in the feed trough reads the individual's identification tag. The system queries the individual's real-time energy requirement and target feed energy concentration, and combines this with the daily feeding frequency of 3 times (set according to the individual's feeding behavior pattern; in the pilot test, this feeding frequency can avoid overfeeding or underfeeding at any one time). The system calculates the single feeding amount (in kg) using the relationship "feeding amount = individual real-time energy requirement / target feed energy concentration / feeding frequency". For example, when the real-time energy requirement is 8400 kcal / day and the target feed energy concentration is 3500 kcal / kg, the single feeding amount is 0.8 kg.

[0045] The system controls the smart feed trough to open the valve of the corresponding formula compartment and dispenses the calculated amount. The built-in weight sensor in the feed trough ensures that the dispensing error is ≤5g. After dispensing, the system records the data of "individual ID-dispensing time-formula energy concentration-dispensing amount" and uploads it to the cloud platform in real time, enabling traceability of the feeding process.

[0046] S4 performs closed-loop optimization and parameter updates. like Figure 1As shown, the feedback optimization unit connects the model calibration unit and the decision and execution unit. Based on the feeding data recorded during the S3 decision execution process, the group classification labels generated by S2, and the associated performance data, it periodically optimizes the predefined feeding strategy and model calibration parameters to ensure that the system continuously adapts to individual differences and environmental changes, thus solving the problem of long-term accuracy decline caused by the lack of dynamic iteration in traditional systems.

[0047] S41 Adjust the predefined feeding strategy The system is set to an optimization cycle of 30 days. This cycle length has been verified by a pilot project with 300 fattening pigs. It can fully cover the performance change trend of individual growth stages, avoid frequent adjustments or data fluctuations caused by too short a cycle, and prevent the strategy from lagging behind actual breeding needs due to too long a cycle.

[0048] Based on the population classification labels generated by S23, the system extracts the core performance data of sensitive, tolerant, and stable populations in the current and previous periods: For sensitive populations, the weekly feed conversion ratio change rate is calculated using ΔFCR=(average FCR of this period - average FCR of the previous period) / average FCR of the previous period, which directly reflects the improvement in their feed utilization efficiency; for tolerant populations, the weekly average daily weight gain change rate is calculated using ΔADG=(average ADG of this period - average ADG of the previous period) / average ADG of the previous period, which reflects the optimization effect of their growth rate; for stable populations, the combined fluctuations of the two types of indicators are monitored simultaneously to ensure no significant performance decline.

[0049] The system sets preset improvement thresholds: for sensitive groups, ΔFCR ≤ -3% (i.e., feed conversion rate decreases by more than 3% compared to the previous period, indicating that feed efficiency improvement has been achieved); for tolerant groups, ΔADG ≥ 2% (i.e., average daily weight gain increases by more than 2% compared to the previous period, indicating that growth rate optimization has been achieved). These thresholds are determined based on statistical analysis of a multi-cycle pilot project involving 300 fattening pigs (these are exemplary values), and they align with reasonable improvement targets for feed efficiency and growth rate in large-scale farming, balancing performance improvement with cost control.

[0050] If the sensitive population's ΔFCR > -3%, it indicates that the current feed energy concentration during extreme high-temperature periods is insufficient to fully compensate for their energy deficit. The system will increase the target feed energy concentration for this type of individual during extreme high-temperature periods by 50 kcal / kg (this adjustment is just an example). If the tolerant population's ΔADG < 2%, it indicates that their energy supply is excessive during high-temperature warning periods. The system will decrease the target feed energy concentration for this type of individual during high-temperature warning periods by 30 kcal / kg (this adjustment is just an example). The adjusted feeding strategy is synchronously updated to the predefined rule base of the decision-making and execution unit, providing updated parameter basis for the next round of S3 prediction and execution cycle.

[0051] S42 Update Model Calibration Parameters The system feeds back the individual types classified by S23 and their associated group performance data to the model calibration unit. The group performance data includes the periodic average FCR, average ADG, feed intake variation patterns at different time periods, and feed response characteristics in different environmental ranges for each type of group. This data can provide real-world support for adjusting model parameters and prevent parameters from deviating from the actual breeding conditions.

[0052] The model calibration unit responds to feedback and applies differentiated parameter calibration weights to different types of individuals: for sensitive individuals, whose energy demands are more affected by environmental fluctuations, the system adjusts the weighting coefficients of the energy conversion efficiency parameter in S21. The value was adjusted from 0.3 to 0.4, increasing the proportion of the current calibration value in parameter updates, allowing individual energy conversion efficiency parameters to adapt more quickly to changes in their energy requirements. A pilot study of 300 fattening pigs showed that this adjustment improved the update response speed of energy conversion efficiency parameters for sensitive individuals by more than 35%, better reflecting their energy demand fluctuations under environmental stress; for tolerant and stable individuals, The value remains unchanged at 0.3 to balance the stability of historical data with the real-time nature of current data.

[0053] Meanwhile, based on the feedback population performance data, the model calibration unit recalculates the actual average feed intake of the population within each environmental data interval every 14 days, and recalculates the environmental dynamic correction factor by combining it with the expected average feed intake under the baseline environment. The 14-day update cycle, verified through pilot testing, effectively reflects the changing trends of environmental influences on foraging while avoiding the pitfalls of frequent updates. Value fluctuations ensure that S22 is generated The factors remain consistently accurate, providing reliable support for adjusting the real-time energy requirements of individuals in S32.

[0054] The implementation principle of a feed efficiency optimization system based on a growth model in this application embodiment is as follows: The system collects and correlates individual identity, weight, feed intake at different times, environmental data, and environmental forecast data in real time through a data acquisition unit. The model calibration unit dynamically calibrates individual energy conversion efficiency parameters, generates dynamic environmental correction factors, and population classification labels based on this data. The decision-making and execution unit employs a dual-cycle collaborative mechanism of prediction and execution—in the prediction cycle, differentiated feed formulations are generated in advance based on environmental forecasts and population classification; in the execution cycle, real-time environmental data, calibrated individual parameters, and correction factors are combined to calculate individual real-time energy requirements and control precise feed delivery. This enables real-time calculation of individual energy requirements and differentiated feeding control. The implementation principle directly addresses the problem that traditional systems cannot respond in real time to changes in individual feeding behavior and energy requirements in dynamic environments. Through dynamic calibration of individual-level parameters and environmental correction, the system can accurately identify and respond to the true energy status of individuals with different tolerance types. This avoids the "one-size-fits-all" feeding strategy caused by relying on population average data, ensuring that sensitive individuals are spared from growth stagnation due to insufficient energy, and that tolerant individuals are spared from feed waste caused by excess energy. Ultimately, this significantly improves the overall feed conversion rate and feeding response speed in a dynamically changing production environment.

[0055] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A feed efficiency optimization system based on a growth model, characterized in that, include: The data acquisition unit is used to collect and associate individual identification data, individual weight data, food intake data by time period, environmental data, and environmental forecast data. The model calibration unit, connected to the data acquisition unit, is used to calibrate the individual energy conversion efficiency parameters based on the individual weight data and the food intake data at different time periods, generate an environmental dynamic correction factor using the environmental data, and generate a group classification label based on historical individual performance data. The decision-making and execution unit connects the data acquisition unit and the model calibration unit, and operates based on a dual-cycle collaborative mechanism of prediction and execution, wherein: The prediction cycle generates pre-production instructions for differentiated feed formulations based on the environmental forecast data and the population classification tags. The execution cycle responds to the individual's identity identifier and, in conjunction with the environmental data, the calibrated individual energy conversion efficiency parameters, and the environmental dynamic correction factor, calculates the individual's real-time energy demand and controls differentiated energy delivery.

2. The system according to claim 1, characterized in that, The data acquisition unit divides the daily feeding time into multiple logical time periods, and within each logical time period, associates the individual's identity with its feeding behavior and corresponding feed intake data to obtain the time-segmented feed intake data.

3. The system according to claim 2, characterized in that, The model calibration unit calibrates the individual energy conversion efficiency parameter in the following ways: it calculates the individual's actual weight gain based on the individual's weight data and calculates the individual's actual metabolizable energy intake based on the time-segmented food intake data; based on the principle of energy balance, it calculates the current calibration value of the individual's energy conversion efficiency parameter by comparing the difference between the individual's actual metabolizable energy intake and the metabolizable energy required to maintain the individual's actual weight gain; and it uses a data smoothing algorithm to fuse the current calibration value with the historical value of the individual's energy conversion efficiency parameter to achieve dynamic updating of the individual's energy conversion efficiency parameter.

4. The system according to claim 1, characterized in that, The model calibration unit generates the environmental dynamic correction factor by: identifying a specific environmental data interval based on the environmental data collected by the data acquisition unit, and calculating the actual average feed intake of the population within the specific environmental data interval; comparing the actual average feed intake with the expected average feed intake under predefined baseline environmental conditions, and generating the environmental dynamic correction factor based on the ratio.

5. The system according to claim 1, characterized in that, The execution cycle of the decision-making and execution unit includes: calculating the basic energy demand of an individual based on the calibrated individual energy conversion efficiency parameter; when the environmental data or environmental forecast data exceeds a preset environmental threshold, the environmental dynamic correction factor is invoked to adjust the basic energy demand of the individual to obtain the real-time energy demand of the individual for control of energy delivery.

6. The system according to claim 5, characterized in that, The steps of the model calibration unit in generating the group classification label include: calculating the individual performance evaluation result corresponding to each individual identity based on the periodically collected individual weight data and the time-segmented food intake data; comparing the individual performance evaluation result with the group performance evaluation benchmark; and classifying individuals into different types according to the comparison result to form the dynamically updated group classification label.

7. The system according to claim 6, characterized in that, When the decision-making and execution unit calculates the real-time energy requirement of the individual during the execution cycle, it also matches different target feed energy concentrations for different types of individuals based on the individual type indicated in the group classification label and a predefined feeding strategy, so as to perform differentiated feeding.

8. The system according to claim 7, characterized in that, It also includes a feedback optimization unit, which is connected to the model calibration unit and the decision and execution unit, and is used for: Based on the group classification labels, the rate of change of the core performance indicators of the corresponding group types is periodically analyzed; The rate of change of the core performance indicators is compared with a preset improvement threshold; Based on the comparison results, the adjustment range of the target feed energy concentration in the predefined feeding strategy is automatically determined and updated.

9. The system according to claim 8, characterized in that, The feedback optimization unit is also used for: The individual types classified based on the group classification labels and their associated group performance data are fed back to the model calibration unit; In response to the feedback, the model calibration unit applies differentiated parameter calibration weights to different types of individuals to update the individual energy conversion efficiency parameters, and periodically updates the environmental dynamic correction factor based on the feedback group performance data.

Images