Livestock breeding precision feeding quantity dynamic optimization system based on individual growth curve
By using multi-source data monitoring and feature determination technology, the problem of distorted feed intake data in existing systems has been solved, enabling accurate correction of actual feed intake and dynamic optimization of individual growth curves, thereby improving the precision feeding effect in livestock farming.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- JIANGXI GRASSROOTS BEAR AGRICULTURAL DEVELOPMENT CO LTD
- Filing Date
- 2026-06-01
- Publication Date
- 2026-07-17
Smart Images
Figure CN122397632A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent livestock farming and precision feeding technology, specifically to a dynamic optimization system for precise feeding amount in livestock farming based on individual growth curves. Background Technology
[0002] With the development of large-scale and intelligent farming, precision feeding technology for livestock farming based on industrial big data analysis is gradually being applied to the growth management of livestock (pigs, cattle, etc.).
[0003] Existing technologies typically acquire livestock identity, weight, and historical feed intake data through electronic ear tags, weighing devices, automatic feeding devices, and growth curve models, and determine the target feeding amount accordingly to achieve dynamic feeding based on individual differences. However, in actual feeding, existing systems often directly record the decrease in trough weight as the actual feed intake, failing to adequately distinguish between actual feed intake and weight changes caused by non-feeding factors such as livestock rooting in the trough, feed spillage, trough vibration, or feed slippage, leading to distorted feed intake data. This distorted feed intake data, when further fed back to individual growth curves and feeding amount optimization models, can easily cause deviations in feeding amount calculations, errors in growth curve correction, and a decrease in the effectiveness of precise feeding.
[0004] Therefore, there is a need for a dynamic optimization system for precise feeding in livestock farming that can combine multi-source data to determine the source of feed trough weight loss and update individual growth curves based on the corrected actual feed intake. Summary of the Invention
[0005] The purpose of this invention is to provide a dynamic optimization system for precise feeding amount in livestock farming based on individual growth curves, so as to solve the shortcomings of the prior art.
[0006] To achieve the above objectives, the present invention provides the following technical solution: a dynamic optimization system for precise feeding amount in livestock farming based on individual growth curves, comprising: The feeding amount determination module identifies the target livestock entering the feeding position, obtains its individual profile, real-time weight and historical feeding data, and determines the target feeding amount based on the real-time weight and individual growth curve. The multi-source data acquisition module controls the automatic feeding device to feed in batches according to the target feeding amount. During the feeding process, it collects data on changes in trough weight, livestock mouth and nose contact status, head movement status, trough vibration status, and feed scattering status outside the trough. Based on the collected data, it generates feeding evidence features and disturbance evidence features respectively. The weight loss source determination module divides the feeding process into segments according to a preset time window, and determines the source of weight loss in each segment based on the feeding evidence characteristics and disturbance evidence characteristics in each segment. The true feed intake correction module includes weight loss determined to be true feed intake in the true feed intake, and removes weight loss determined to be caused by arching, overflow, vibration or slippage from the true feed intake. The dynamic feed amount optimization module updates the target livestock's feed intake records and individual growth curves based on the corrected actual feed intake, and dynamically adjusts the target feed amount for the next feeding accordingly.
[0007] Preferably, determining the target feeding amount includes the following steps: After identifying the target livestock that has entered the feeding area, the system retrieves the individual file, historical feeding data, and historical weight data corresponding to the target livestock. Real-time weight is continuously collected, and abnormal weight data caused by the target animal not fully entering the feeding position or being unstable are removed; The processed real-time weight is compared with the target weight for the corresponding age in the individual growth curve to determine the current growth status of the target livestock. The target feeding amount is determined based on the current growth status, historical feeding completion, and the allowable range for a single feeding.
[0008] Preferably, generating feeding evidence features and disturbance evidence features includes the following steps: After each batch of feed is completed, the weight impact data corresponding to the start and stop of the automatic feeding device and the moment the feed falls into the trough are shielded. The weight changes of the feed trough after shielding, the contact status of the livestock's mouth and nose, the head movement status, the vibration status of the feed trough, and the state of feed scattering outside the feed trough were time-aligned. Based on the time-aligned data, synchronous observation data corresponding to the current batch of material is generated; Based on synchronous observation data, feeding evidence features for characterizing actual feeding and disturbance evidence features for characterizing non-feeding disturbances were extracted respectively.
[0009] Preferably, the extraction of feeding evidence features includes the following steps: Continuous state segments were extracted from synchronous observation data, showing that the livestock's mouth and nose remained in the feeding area of the trough. Among these continuous state segments, segments in which the head movements remained stable, the weight of the trough decreased continuously, and the vibration of the trough did not reach a disturbed state were selected as feeding candidate segments. Determine whether there is a new trend of feed scattering outside the feed trough within the candidate feeding segment; When there is no new trend of scattering, the persistence of mouth-nose contact, head stability, and continuous decrease in the weight of the feeding trough are jointly identified as characteristics of feeding evidence.
[0010] Preferably, the extraction of perturbation evidence features includes the following steps: Identify sudden increases in trough vibration, rapid head pushing or swinging, step decreases in trough weight, and new instances of feed scattering outside the trough from synchronous observation data. When any of these states occur within a continuous observation period, the corresponding segment is marked as a candidate disturbance segment. The vibration mutation state, abnormal head movement state, sudden weight drop state, and scattered new state in the perturbation candidate segments are correlated, and the correlation state combination is determined as the perturbation evidence feature.
[0011] Preferably, determining the source of weight loss includes the following steps: The feeding process is divided into multiple consecutive segments, and overlapping observation areas are preserved between adjacent segments; Read the feeding evidence features and disturbance evidence features in each segment; When the feeding evidence features remain consistent in terms of mouth-nose contact persistence, head stability, and continuous decrease in feed trough weight, the segment is assigned a feeding dominance marker. When the perturbation evidence features a series of vibrational abrupt changes, abnormal head movements, sudden weight drops, or newly added scattered states, the segment is assigned a perturbation-dominant marker.
[0012] Preferably, the weight loss source determination module further includes the following steps: When both feeding evidence features and disturbance evidence features are present in the same segment, or neither is sufficient to determine the source on its own, the segment is marked as pending conflict. Perform backtracking judgment on adjacent segments before and after the segment with pending conflict marker; When both adjacent segments are feeding-dominant markers and no disturbance-dominant markers appear, the segment with conflict pending markers is incorporated into the actual feeding process; When a perturbation-dominant marker exists in the preceding or following segment, the segment with the pending conflict marker is incorporated into the corresponding perturbation process.
[0013] Preferably, correcting the actual feed intake includes the following steps: Receive the final label of each segment and the corresponding weight loss of the feed trough, and form a sequence of weight loss segments in chronological order; Merge adjacent segments from the same source into the same weight decrease event; The duplicate weight loss values located in the overlapping observation area are deduplicated, and only the weight loss values that occurred earlier are retained. Write the event weight loss corresponding to the actual feed source into the actual feed intake, and write the event weight loss corresponding to the arching, spill, vibration or slippage source into the non-feed loss.
[0014] Preferably, updating the individual growth curve includes the following steps: Receive corrected feeding results that include actual feed intake and non-feed loss; Write the actual feed intake into the feed intake record of the target livestock, and store the non-feed loss amount as a feeding interference record by associating it with the corresponding feeding time; Data from feeding records with stable actual feed intake and non-feed loss not meeting the abnormality marking criteria were selected as data for updating growth curves. Data showing abnormalities such as arching, overflow, vibration, or slippage are saved as reference data and are not used directly to correct individual growth curves.
[0015] Preferably, dynamically adjusting the target feeding amount for the next feeding includes the following steps: The feed intake status and weight gain trend of the target livestock at the corresponding age were corrected based on the updated data from the growth curve. The direction of deviation of the target livestock from the target growth state is determined based on the revised individual growth curve; The feeding completion status of the target livestock is determined by combining the most recent actual feed intake, and the degree of non-feeding interference is determined by combining the most recent feeding interference record. The target feeding amount for the next feeding is adjusted based on deviation from the feeding direction, feeding completion status, and the degree of non-feeding interference, so that the decrease in non-feeding weight is not considered as effective intake and participates in the increase of continuous feeding amount.
[0016] The technical effects and advantages provided by the present invention in the above technical solution are as follows: In the automatic feeding process, this invention does not solely rely on the decrease in trough weight to determine whether livestock have consumed feed. Instead, it simultaneously collects data on changes in trough weight, the livestock's mouth and nose contact status, head movements, trough vibration, and feed spillage outside the trough, forming evidence of both feeding and disturbance. The feeding process is then segmented into segments using preset time windows, and the source of weight loss is determined by combining the dual-type evidence in each segment. This distinguishes weight loss caused by actual feeding from non-feeding weight loss due to trough arching, spillage, vibration, or slippage, avoiding the problem of inflated feed intake caused by directly equating trough weight reduction with actual feed intake in existing technologies.
[0017] This invention enables the inclusion of only weight loss corresponding to actual feed sources in the actual feed intake after feeding, while removing or recording weight loss from non-feed sources as non-feeding losses. This makes the data written into the target livestock's feed intake record closer to the actual intake. Furthermore, the corrected actual feed intake is used to update the individual growth curve and dynamically adjust the target feeding amount for the next feeding. This prevents abnormal events such as rooting, overflow, vibration, or slippage from contaminating the growth curve and feeding model, reduces continuous feeding deviation, and improves the accuracy of determining the feeding amount for a single animal, the stability of the feeding process, and the reliability of individual growth curve optimization. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.
[0019] Figure 1 This is a flowchart of a dynamic optimization system module for precise feeding amount in livestock farming based on individual growth curves, according to the present invention.
[0020] Figure 2 This is a flowchart illustrating the method for dynamically adjusting the target feeding amount for the next feeding cycle according to the present invention. Detailed Implementation
[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.
[0022] For examples, please refer to Figure 1 As shown in this embodiment, the livestock breeding precision feeding dynamic optimization system based on individual growth curves includes: The feeding amount determination module identifies the target livestock (which can be pigs or cattle) entering the feeding position, obtains its individual profile, real-time weight and historical feeding data, and determines the target feeding amount based on the real-time weight and individual growth curve.
[0023] In one embodiment, the feeding amount determination module is used to determine the target feeding amount for a livestock after it enters the feeding position. Specifically, when a livestock enters the feeding position, an identification unit located at the entrance of the feeding position or within the feeding passage reads the electronic ear tag information worn by the target livestock, or an image recognition unit collects the livestock's physical characteristics and identifies its individual number to determine the identity of the target livestock entering the feeding position. After identification is completed, the feeding amount determination module retrieves the corresponding individual file from the database based on the individual number. The individual file includes, but is not limited to, livestock breed, sex, date of birth, date of entry into the pen, current age, historical weight record, historical feed intake record, historical feeding amount record, health status record, and target slaughter weight.
[0024] Furthermore, a weighing unit is installed at the bottom of the feeding position or the bottom of the passageway. After the target livestock enters the feeding position, the weighing unit collects the real-time weight data of the target livestock. To improve the stability of the real-time weight data, the feeding amount determination module can perform outlier removal and smoothing on multiple weight values continuously collected by the weighing unit within a preset time, and use the processed weight value as the real-time weight of the target livestock. If it is detected that the livestock is unstable, enters the feeding position with only half of its body, or the weight data fluctuates beyond the preset range, the determination of the target feeding amount is temporarily suspended, and the real-time weight is collected again.
[0025] The feeding amount determination module calls the individual growth curve corresponding to the target livestock. This individual growth curve characterizes the standard weight change trend and target weight gain trend of the target livestock at different ages or growth stages. This individual growth curve can be pre-established based on the target livestock's historical weight data, historical feed intake data, and a group growth model of livestock of the same breed, and can be continuously updated with subsequent actual growth data of the target livestock; further details are omitted here. The feeding amount determination module matches the real-time weight with the target weight at the corresponding age or growth stage in the individual growth curve to obtain the current growth status of the target livestock. When the real-time weight is lower than the target growth level, it is determined that the target livestock has a need for compensatory growth; when the real-time weight is higher than the target growth level, it is determined that the target livestock has a need for controlled feeding or to reduce the risk of overfeeding.
[0026] Furthermore, the feeding amount determination module combines the target livestock's current age, real-time weight, historical feed intake, historical feeding completion rate, recent weight gain trend, and the target weight gain requirement corresponding to the individual growth curve to determine the basic feeding amount for the target livestock during this feeding process, and adjusts the basic feeding amount according to the target livestock's current growth status. Specifically, when the target livestock has a high recent feed completion rate and its real-time weight is lower than the target weight corresponding to the individual growth curve, the target feeding amount can be increased within the safe feeding range; when the target livestock has a lot of uneaten feed, a low feed completion rate, or a real-time weight higher than the target weight corresponding to the individual growth curve, the target feeding amount can be reduced or the batch feeding ratio can be adjusted.
[0027] In addition, the feeding amount determination module can limit the target feeding amount according to preset feeding rules, ensuring that it does not exceed the maximum single feeding amount allowed by the target livestock's current age, weight stage, and health status, while not falling below the minimum feeding amount required to maintain normal feeding induction. The feeding amount after limitation is taken as the target feeding amount and sent to the multi-source data acquisition module and automatic feeding device for subsequent batch feeding and monitoring of the feeding process.
[0028] It should be noted that this embodiment can determine the differentiated target feeding amount based on the individual identity, real-time weight, historical feeding data and individual growth curve of the target livestock before feeding, providing an accurate basis for subsequent correction of actual feed intake and dynamic optimization of feeding amount.
[0029] The multi-source data acquisition module controls the automatic feeding device to feed the livestock in batches according to the target feeding amount. During the feeding process, it collects data on changes in the weight of the feed trough, the contact status of the livestock's mouth and nose, the head movement status, the vibration status of the feed trough, and the state of feed scattering outside the feed trough. Based on the collected data, it generates feeding evidence features and disturbance evidence features respectively.
[0030] The automatic feeding device feeds the material in batches according to the target feeding amount. After each batch of feeding is completed, a preset observation period begins. The preset observation period is preferably 8 seconds, starting from the 1st second after the automatic feeding device stops feeding and continuing until the 9th second.
[0031] Because the automatic feeding device's start-up, stop-up, and the moment feed falls into the trough cause a short-term impact on the trough's weight, the data acquisition unit performs a feeding shielding process for each batch of feed to prevent this impact from being mistaken for livestock feeding. The feeding shielding process lasts from 0.3 seconds before the automatic feeding device starts to 1 second after feeding stops; the trough weight data obtained within this time range is not used for subsequent feeding judgments and is only saved as a record of equipment operation.
[0032] After the feeding shielding process was completed, time alignment was performed on the changes in trough weight, livestock snout-nose contact status, head movement status, trough vibration status, and feed scattering status outside the trough, using a uniform sampling interval of 0.2 s. The changes in trough weight were denoted as W(t), livestock snout-nose contact status as C(t), head movement status as A(t), trough vibration status as V(t), and feed scattering status outside the trough as S(t). When the sampling frequency of a certain type of data was higher than 0.2 s, the average of adjacent sampling points was taken; when the sampling frequency of a certain type of data was lower than 0.2 s, linear interpolation between adjacent points was used for completion. The linear interpolation was obtained as follows: when the time to be completed was between the previous and next sampling times, the value to be completed was equal to the previous sampling value plus the difference between the next and previous sampling values, multiplied by the time proportion of the time to be completed relative to the previous sampling time. After time alignment, synchronous observation data corresponding to the current batch of feeding was formed.
[0033] In the synchronous observation data, the livestock's snout contact state C(t) is determined by the distance from the center point of the snout to the boundary of the feeding area of the trough. When the center point of the snout is located within the feeding area of the trough and the vertical distance from the feed surface does not exceed 3 cm, C(t) is set to 1; otherwise, it is set to 0. The head movement state A(t) is determined by the displacement of the head center point in two consecutive frames. A(t) is equal to the spatial distance between the head center point at the current moment and the head center point at the previous moment divided by 0.2 s. The trough vibration state V(t) is obtained by the vibration sensor installed at the bottom of the trough, and the average value of the absolute value of vibration acceleration within 0.2 s is taken. The feed scattering state S(t) outside the trough is determined by the area of newly added feed in the designated area outside the trough, in square centimeters. The decrease in trough weight between adjacent sampling points is denoted as D(t). D(t) is equal to the trough weight at the previous moment minus the trough weight at the current moment; when the calculation result is less than 0, D(t) is recorded as 0.
[0034] The feeding candidate fragments were obtained using a continuous state extraction method. The continuous state must simultaneously meet the following conditions: C(t) remains 1 for a continuous time of not less than 3 s; A(t) does not exceed 6 cm / s; W(t) shows a continuous decreasing state, and D(t) corresponding to any two adjacent sampling points is greater than 0; V(t) does not exceed 0.15 g, where g is the unit of gravitational acceleration.
[0035] Continuous time periods that meet the above conditions are marked as candidate feeding segments. This limitation can exclude weight fluctuations caused by rapid head swings, nose pushing against the feed trough, and falling feed impacts, ensuring that segments entering subsequent processing simultaneously have mouth-nose contact, head stability, continuous weight decrease, and low vibration characteristics.
[0036] Further analysis of feed dispersion outside the feeding trough is conducted within the candidate feeding segments. If the difference between S(t) at the end of the segment and S(t) at the beginning of the segment does not exceed 5 cm², it is considered that no new dispersion trend has emerged. In this case, the persistence of mouth-nose contact, head stability, and continuous decrease in trough weight within the segment are collectively encapsulated as feeding evidence features. Mouth-nose contact persistence is the proportion of sampling points where C(t) is 1 within the segment out of the total number of sampling points in the segment; head stability is 1 minus the ratio of the average A(t) within the segment to 6 cm / s, with a value of 0 when the result is less than 0; continuous decrease in trough weight is the proportion of sampling points where D(t) is greater than 0 within the segment out of the total number of sampling points in the segment. The resulting feeding evidence features reflect the temporal continuity and stability of the actual feeding action.
[0037] Disturbance candidate segments were independently extracted from synchronous observation data. A segment was marked as a disturbance candidate segment if any of the following states occurred continuously for at least 1 second: V(t) continuously exceeded 0.25 g; A(t) continuously exceeded 12 cm / s; the weight of the feed trough decreased by 8% of the current batch feed amount within 0.4 seconds; S(t) increased by more than 10 cm² within 1 second. These values were set based on equipment calibration data under normal fattening livestock feeding conditions. 0.25 g was used to distinguish between micro-vibrations caused by normal chewing and impact vibrations caused by pushing the feed trough; 12 cm / s was used to distinguish between stable feeding and rapid head-shaking movements; 8% was used to identify step-like weight drops; and 10 cm² was used to identify newly visible areas of scattered feed.
[0038] In the perturbation candidate segments, vibration abrupt change states, abnormal head movement states, sudden weight drop states, and newly added scattering states are collectively encapsulated as perturbation evidence features. Vibration abrupt change states are represented by the number of continuous sampling points where V(t) exceeds 0.25 g; abnormal head movement states are represented by the number of continuous sampling points where A(t) exceeds 12 cm / s; sudden weight drop states are represented by the proportion of the feed trough weight decrease within 0.4 s to the current batch feed amount; and newly added scattering states are represented by the area of S(t) increase within 1 s. Perturbation evidence features and feeding evidence features correspond to non-feeding factors and actual feeding factors, respectively. Subsequent determination of the source of weight drop can be directly based on a comparison of the two, thereby avoiding the miscounting of trough arching, overflow, vibration, or slippage as actual feed intake by relying solely on the feed trough weight drop.
[0039] The weight loss source determination module divides the feeding process into segments according to a preset time window, and determines the source of weight loss in each segment based on the feeding evidence characteristics and disturbance evidence characteristics in each segment.
[0040] The process for determining the source of weight loss is executed after each feeding begins. The feeding process starts when the automatic feeding device completes the first batch of feed and ends when the target livestock leaves the feeding position, the cumulative feed amount reaches the target feeding amount, or the stop feeding command is issued. The continuous time between the start and end points is divided into multiple preset time windows, each preferably 4 seconds long. A 1-second overlap observation zone is maintained between adjacent preset time windows, i.e., the first segment is 0 seconds to 4 seconds, the second segment is 3 seconds to 7 seconds, the third segment is 6 seconds to 10 seconds, and so on. By setting the overlap observation zone, the mouth and nose contact status, head movement status, trough weight change, and trough vibration status at the end of the previous segment can be carried over to the beginning of the next segment, avoiding misjudgments caused by single trough arching or brief feeding actions being truncated by time boundaries.
[0041] For each segment, feeding evidence and disturbance evidence features were collected within the corresponding time range. Feeding evidence features included the continuity of mouth-nose contact, head stability, and continuous decrease in the weight of the feeding trough, denoted as P1, P2, and P3, respectively. Disturbance evidence features included abrupt vibration, abnormal head movements, sudden weight drops, and newly added scattered particles, denoted as Q1, Q2, Q3, and Q4, respectively. P1, P2, and P3 were normalized to between 0 and 1, with values closer to 1 indicating a closer match to a genuine feeding process. Q1, Q2, Q3, and Q4 were also normalized to between 0 and 1, with values closer to 1 indicating a closer match to a non-feeding disturbance process. The normalization method is to divide the measured value by the corresponding judgment limit and limit the maximum value to 1. Among them, the mouth and nose contact persistence is obtained by the proportion of mouth and nose contact sampling points within the segment; head stability is obtained by subtracting the ratio of the average head movement speed to 6 cm / s from 1; the continuous weight drop state of the feed trough is obtained by the proportion of continuous drop sampling points within the segment; the vibration change state is obtained by the proportion of continuous sampling points exceeding 0.25 g; the abnormal head movement state is obtained by the proportion of continuous sampling points exceeding 12 cm / s; the weight drop state is obtained by the proportion of the weight drop within 0.4 s to 8% of the current batch feed amount; and the new scattered state is obtained by the proportion of the newly scattered area within 1 s to 10 cm².
[0042] Based on the above characteristics, calculate the strength of evidence for food consumption F and the strength of evidence for disturbance R. The strength of evidence for food consumption F is equal to the sum of P1 multiplied by 0.4, P2 multiplied by 0.3, and P3 multiplied by 0.3; the strength of evidence for disturbance R is equal to the sum of Q1 multiplied by 0.25, Q2 multiplied by 0.25, Q3 multiplied by 0.25, and Q4 multiplied by 0.25.
[0043] The aforementioned weights are used to ensure that the persistence of mouth-nose contact plays a dominant role in determining actual feed intake, while maintaining equal contributions from vibration, abnormal head movements, sudden weight drops, and new littering in the determination of disturbance. Judgment limits are set based on normal feed intake calibration data. Preferably, the disturbance lateral limit is the mean of data collected from 20 animals of the same age over three consecutive days under conditions of no spillage and no rooting. The feed intake lateral limit is the mean of the same batch of data minus two standard deviations.
[0044] When F is not less than 0.75, and P1, P2, and P3 are all not less than 0.6, while R is less than 0.35, the segment is assigned a feeding-dominant label; when R is not less than 0.6, and at least one of Q1, Q2, Q3, and Q4 is not less than 0.8, the segment is assigned a disturbance-dominant label; when F is not less than 0.75 and R is not less than 0.35, or when F is less than 0.75 and R is less than 0.6, the segment is assigned a conflict-pending label. This determination method can avoid identifying feeding solely based on a decrease in trough weight, and can also prevent short-term slight vibrations from directly overriding genuine feeding segments.
[0045] For segments marked with pending conflict, a backtracking determination of adjacent segments is performed. The backtracking range is the segment before and the segment after the target segment; when the target segment is located at the beginning or end of the feeding process, only the adjacent segments on one side that exist are used.
[0046] If both adjacent segments are feeding-dominant markers, and Q1, Q2, Q3, and Q4 in the adjacent segments are all below 0.5, then the target segment is incorporated into the actual feeding process. If the preceding or following segment is a perturbation-dominant marker, and any one of Q1, Q2, Q3, and Q4 in the target segment reaches 0.5, then the target segment is incorporated into the corresponding perturbation process. If the adjacent segments are feeding-dominant and perturbation-dominant markers respectively, then the characteristic continuity of the target segment is compared with that of the segments on both sides. The continuity is determined by the absolute values of the F difference and the R difference between the adjacent segments, and the target segment is incorporated into the process on the side with the lower sum of the absolute values of the differences.
[0047] After the retrospective determination is completed, the source of weight loss is determined based on the final segment label. For feeding-dominant segments and conflict-pending segments incorporated into the actual feeding process, the weight loss in the feed trough within these segments is determined to be the actual feeding source. For disturbance-dominant segments and conflict-pending segments incorporated into the corresponding disturbance process, the weight loss in the feed trough within these segments is determined to be a source of arching, overflow, vibration, or slippage. If the same weight loss event spans two overlapping observation areas, only the determination result of the segment with the earlier time start is retained, and the weight loss repeatedly covered in the later segment is not counted again. This ensures that the determination results of the weight loss source are temporally continuous and without duplicate statistics.
[0048] The true feed intake correction module includes weight loss determined as true feed intake in the true feed intake, and removes weight loss determined as caused by arching, overflow, vibration or slippage from the true feed intake.
[0049] The actual feed intake correction process is executed after receiving the weight loss source determination result. Each segment records the segment start time, segment end time, segment final marker, and feed trough weight loss. The segment final marker includes the actual feed intake source and sources such as arching, overflow, vibration, or slippage. Segments are arranged from earliest to latest according to their start time, forming a weight loss segment sequence, denoted as A1, A2, ..., An. The feed trough weight loss corresponding to the i-th segment is denoted as Di. Di is obtained by summing the weight loss of adjacent sampling points; that is, when the weight of a subsequent sampling point is lower than the weight of a previous sampling point, the difference is taken as the weight loss for that sampling interval; when the weight of a subsequent sampling point is not lower than the weight of a previous sampling point, the weight loss for that sampling interval is recorded as 0. The sum of the weight loss of all sampling intervals within a segment is Di.
[0050] After the weight loss segment sequence is formed, adjacent segments with the same source are merged. If the final labels of segments Ai and Ai+1 are both from the actual feeding source, and the interval between the end time of Ai and the start time of Ai+1 does not exceed 0.2 s, then they are merged into the same weight loss event. If the final labels of segments Ai and Ai+1 are both from the same source—arch source, overflow source, vibration source, or slippage source—and the time interval does not exceed 0.2 s, they are also merged into the same weight loss event. The merged weight loss event records the start time, end time, source type, and the set of sampling points covered. The 0.2 s interval is consistent with the aforementioned unified sampling interval, used to ensure that adjacent segments that are temporally continuous can be considered the same event, while avoiding the erroneous merging of two independent actions.
[0051] Because there are overlapping observation areas between adjacent segments, the same sampling interval may fall into two segments simultaneously. To avoid duplicate statistics, deduplication is performed on each weight decrease event. Using the sampling interval as the smallest statistical unit, a set of already counted sampling points, B, is first established, initially an empty set. Weight decrease events are processed from earliest to latest according to their start time. If a sampling interval covered by the current weight decrease event is not included in B, the weight decrease corresponding to that sampling interval is included in the current weight decrease event; if the sampling interval is already included in B, it is not included in the current weight decrease event. After the current weight decrease event is processed, the included sampling intervals are added to B. The deduplicated weight decrease of the current weight decrease event is denoted as Gk, which is equal to the sum of the weight decreases of all sampling intervals in this event that were not counted by previous events. Through this process, the same weight decrease within the overlapping observation area is only retained in the weight decrease event that occurred earlier.
[0052] After deduplication, the weight loss events are categorized according to their source type. If the source of the weight loss event is actual feed intake, Gk is recorded in the actual feed intake account. If the source of the weight loss event is trough arching, overflow, vibration, or slippage, Gk is recorded in the non-feeding loss account, and the cause of the non-feeding loss is also recorded. The cumulative value of the actual feed intake account is denoted as I, which equals the sum of weight loss events from all actual feed intake events. The cumulative value of the non-feeding loss account is denoted as L, which equals the sum of weight loss events from all non-feeding sources. In cases where both actual and non-feeding sources exist simultaneously during the same feeding process, they are accumulated separately, and the total weight loss of the trough is not directly used to replace the actual feed intake.
[0053] At the end of this feeding period, a corrected feeding result is generated based on the actual feed intake I, non-feeding loss L, cumulative feed intake M, initial weight of the feed trough W0 before feeding, and remaining weight of the feed trough We at the end of feeding. To verify statistical consistency, the weight balance difference E is first calculated as follows: E = W0 + M − We − I − L; where W0 is the initial weight of the feed trough before feeding, M is the cumulative feed intake, We is the remaining weight of the feed trough at the end of feeding, I is the actual feed intake, and L is the non-feeding loss.
[0054] If the absolute value of E does not exceed 2% of the cumulative feed amount M, the corrected feeding result is valid; if the absolute value of E exceeds 2% of M, the feeding is marked as an abnormal weight balance, and I is retained as a priority feeding record, while E is written into the abnormal difference. The value of 2% is determined by the calibration error of the weighing sensor and the short-term fluctuation error after the feed falls into the trough, preferably twice the upper limit of the full-scale error of the weighing sensor.
[0055] The revised feed intake results include the target livestock number, feeding time, cumulative feed intake M, actual feed intake I, non-feeding loss L, remaining weight We, the type of each weight loss event, and the weight balance difference E. These revised feed intake results serve as the data source for the target livestock's current feed intake record. The actual feed intake I is used to subsequently update the individual growth curve, and the non-feeding loss L characterizes the impact of rooting, spillage, vibration, or slippage during the feeding process. Therefore, the portion of the trough weight loss caused by non-feeding reasons will not be included in the actual feed intake, reducing the impact of inflated feed intake records on the subsequent revision of the target feed intake.
[0056] Please see Figure 2 As shown, the dynamic feed amount optimization module updates the target livestock's feed intake record and individual growth curve based on the corrected actual feed intake, and dynamically adjusts the target feed amount for the next feeding accordingly.
[0057] In this invention, the dynamic optimization process for feeding amount is executed after the current feeding is completed and a corrected feeding result is generated. The corrected feeding result includes the target livestock number, the feeding time, the cumulative feed amount M, the actual feed intake I, the non-feeding loss L, the remaining weight We, and the weight balance difference E. Based on the target livestock number, I is written into the corresponding feeding record of the target livestock, and L is associated with the current feeding time and the type of weight loss source and saved as the interference record for this feeding. The feeding records are saved in the order of feeding time, and each feeding record includes at least the feeding time, target feeding amount, cumulative feed amount, actual feed intake, non-feeding loss, and feeding completion status, wherein the feeding completion status is obtained by dividing I by the target feeding amount.
[0058] After the feeding records are written, the feeding records formed in the last 7 days for the target livestock are screened. Let the actual feed intake of the j-th feeding record be Ij, the non-feeding loss be Lj, and the cumulative feed input be Mj. First, calculate the proportion of non-feeding loss Kj, which is equal to Lj divided by Mj; then calculate the proportion of weight balance deviation Hj, which is equal to the absolute value of E divided by Mj. If Kj does not exceed 5%, Hj does not exceed 2%, and the coefficient of variation of the actual feed intake of this feeding record with the two preceding and following feeding records does not exceed 8%, then this feeding record is recorded as growth curve update data. The calculation method of the coefficient of variation of actual feed intake is as follows: first, calculate the average of the actual feed intake in this feeding record and the adjacent feeding records; then calculate the average of the squares of the differences between each actual feed intake and the average, and take the square root to obtain the standard deviation; divide the standard deviation by the average to obtain the coefficient of variation. If Kj exceeds 5%, Hj exceeds 2%, or the coefficient of variation exceeds 8%, then this feeding record is recorded as reference data. This reference data is retained in the feeding file but is not directly used to correct the individual growth curve. The 5% value is determined by the upper limit of the weight corresponding to the area of scattered feed outside the feed trough under normal feeding conditions as a percentage of the cumulative feed amount; the 2% value is consistent with the weighing error verification; and the 8% value is determined by the statistical analysis of the fluctuation within a consecutive normal feeding day for the same livestock.
[0059] After obtaining the updated growth curve data, update the feed intake completion status and weight gain trend of the target livestock at the corresponding age. Let Wd be the real-time weight of the target livestock on day d, and Wd-1 be the real-time weight on day d-1. Then, the daily weight gain Gd is equal to Wd minus Wd-1. Let Iday be the sum of the actual feed intake of all updated growth curve data within day d, and Pday be the sum of the total target feed intake within day d. Then, the feed intake completion status Fd is equal to Iday divided by Pday. If multiple weighings occur within the same age group, the average weight after removing the maximum and minimum values is taken as Wd; if only one weighing occurs, the result of that weighing is taken directly as Wd.
[0060] In this invention, the individual growth curve is corrected using a smooth update method. Let the body weight on day d before correction be Cd, and the body weight on day d after correction be Cd', then Cd' = 0.7 × Cd + 0.3 × Wd. Let the target weight gain on day d before correction be Ad, and the target weight gain after correction be Ad', then Ad' = 0.6 × Ad + 0.4 × Gd. These weights are used to prevent daily body weight fluctuations from directly altering the individual growth curve, while allowing continuous and effective feeding data to gradually adjust subsequent feeding guidelines. When only reference data is available for a certain age without updated growth curve data, Cd and Ad remain unchanged, and only the feeding interference at that age is recorded.
[0061] The target feeding amount for the next feeding is dynamically adjusted based on the revised individual growth curve. Let the basal feeding amount for the next feeding be P0, the current real-time weight be Wc, the target weight corresponding to the current age on the revised individual growth curve be Cc, and the weight deviation B = Cc - Wc. If B is positive, it indicates that the target livestock is below the target growth state; if B is negative, it indicates that the target livestock is above the target growth state.
[0062] Let Ilast be the most recent actual feed intake, Plast be the most recent target feed intake, and Rlast be the feed intake completion rate (Rlast = Ilast / Plast). The next target feed intake (Pnext) is equal to P0 plus the weight deviation correction plus the feed intake completion correction minus the interference deduction. The weight deviation correction is the result of P0 multiplied by B and divided by Cc, and is limited to ±10% of P0. The feed intake completion correction is 5% of P0 when Rlast is not less than 0.95, negative 5% of P0 when Rlast is less than 0.85, and 0 otherwise. The interference deduction is 5% of P0 when the most recent non-feeding loss (L) accounts for more than 5% of the cumulative feed intake (M), and 0 otherwise. The Pnext calculated above is then limited to the allowable single feed intake for the current age and used as the next target feed intake.
[0063] It should be noted that the actual feed intake is used to correct the individual growth curve, while non-feed loss is only used for interference deduction. Weight loss caused by troughing, overflow, vibration or slippage will not be mistakenly regarded as effective intake, thereby reducing continuous feeding deviation.
[0064] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.
Claims
1. A dynamic optimization system for precise feeding amount in livestock farming based on individual growth curves, characterized in that, include: The feeding amount determination module identifies the target livestock entering the feeding position, obtains its individual profile, real-time weight and historical feeding data, and determines the target feeding amount based on the real-time weight and individual growth curve. The multi-source data acquisition module controls the automatic feeding device to feed in batches according to the target feeding amount. During the feeding process, it collects data on changes in trough weight, livestock mouth and nose contact status, head movement status, trough vibration status, and feed scattering status outside the trough. Based on the collected data, it generates feeding evidence features and disturbance evidence features respectively. The weight loss source determination module divides the feeding process into segments according to a preset time window, and determines the source of weight loss in each segment based on the feeding evidence characteristics and disturbance evidence characteristics in each segment. The true feed intake correction module includes weight loss determined to be true feed intake in the true feed intake, and removes weight loss determined to be caused by arching, overflow, vibration or slippage from the true feed intake. The dynamic feed amount optimization module updates the target livestock's feed intake records and individual growth curves based on the corrected actual feed intake, and dynamically adjusts the target feed amount for the next feeding accordingly.
2. The livestock breeding precision feeding dynamic optimization system based on individual growth curves according to claim 1, characterized in that, Determining the target feeding amount involves the following steps: After identifying the target livestock that has entered the feeding area, the system retrieves the individual file, historical feeding data, and historical weight data corresponding to the target livestock. Real-time weight is continuously collected, and abnormal weight data caused by the target animal not fully entering the feeding position or being unstable are removed; The processed real-time weight is compared with the target weight for the corresponding age in the individual growth curve to determine the current growth status of the target livestock. The target feeding amount is determined based on the current growth status, historical feeding completion, and the allowable range for a single feeding.
3. The livestock breeding precision feeding dynamic optimization system based on individual growth curves according to claim 1, characterized in that, Generating feeding evidence features and disturbance evidence features includes the following steps: After each batch of feed is completed, the weight impact data corresponding to the start and stop of the automatic feeding device and the moment the feed falls into the trough are shielded. The weight changes of the feed trough after shielding, the contact status of the livestock's mouth and nose, the head movement status, the vibration status of the feed trough, and the state of feed scattering outside the feed trough were time-aligned. Based on the time-aligned data, synchronous observation data corresponding to the current batch of material is generated; Based on synchronous observation data, feeding evidence features for characterizing actual feeding and disturbance evidence features for characterizing non-feeding disturbances were extracted respectively.
4. The livestock breeding precision feeding dynamic optimization system based on individual growth curves according to claim 3, characterized in that, The extraction of evidence features of feeding includes the following steps: Continuous state segments were extracted from synchronous observation data, showing that the livestock's mouth and nose remained in the feeding area of the trough. Among these continuous state segments, segments in which the head movements remained stable, the weight of the trough decreased continuously, and the vibration of the trough did not reach a disturbed state were selected as feeding candidate segments. Determine whether there is a new trend of feed scattering outside the feed trough within the candidate feeding segment; When there is no new trend of scattering, the persistence of mouth-nose contact, head stability, and continuous decrease in the weight of the feeding trough are jointly identified as characteristics of feeding evidence.
5. The livestock breeding precision feeding dynamic optimization system based on individual growth curves according to claim 3, characterized in that, The extraction of features from perturbation evidence includes the following steps: Identify sudden increases in trough vibration, rapid head pushing or swinging, step decreases in trough weight, and new instances of feed scattering outside the trough from synchronous observation data. When any of these states occur within a continuous observation period, the corresponding segment is marked as a candidate disturbance segment. The vibration mutation state, abnormal head movement state, sudden weight drop state, and scattered new state in the perturbation candidate segments are correlated, and the correlation state combination is determined as the perturbation evidence feature.
6. The livestock breeding precision feeding dynamic optimization system based on individual growth curves according to claim 1, characterized in that, Determining the source of weight loss includes the following steps: The feeding process is divided into multiple consecutive segments, and overlapping observation areas are preserved between adjacent segments; Read the feeding evidence features and disturbance evidence features in each segment; When the feeding evidence features remain consistent in terms of mouth-nose contact persistence, head stability, and continuous decrease in feed trough weight, the segment is assigned a feeding dominance marker. When the perturbation evidence features a series of vibrational abrupt changes, abnormal head movements, sudden weight drops, or newly added scattered states, the segment is assigned a perturbation-dominant marker.
7. The livestock breeding precision feeding dynamic optimization system based on individual growth curves according to claim 6, characterized in that, The weight loss source determination module also includes the following steps: When both feeding evidence features and disturbance evidence features are present in the same segment, or neither is sufficient to determine the source on its own, the segment is marked as pending conflict. Perform backtracking judgment on adjacent segments before and after the segment with pending conflict marker; When both adjacent segments are feeding-dominant markers and no disturbance-dominant markers appear, the segment with conflict pending markers is incorporated into the actual feeding process; When a perturbation-dominant marker exists in the preceding or following segment, the segment with the pending conflict marker is incorporated into the corresponding perturbation process.
8. The livestock breeding precision feeding dynamic optimization system based on individual growth curves according to claim 7, characterized in that, Correcting the actual feed intake includes the following steps: Receive the final label of each segment and the corresponding weight loss of the feed trough, and form a sequence of weight loss segments in chronological order; Merge adjacent segments from the same source into the same weight decrease event; The duplicate weight loss values located in the overlapping observation area are deduplicated, and only the weight loss values that occurred earlier are retained. Write the event weight loss corresponding to the actual feed source into the actual feed intake, and write the event weight loss corresponding to the arching, spill, vibration or slippage source into the non-feed loss.
9. The livestock breeding precision feeding dynamic optimization system based on individual growth curves according to claim 8, characterized in that, Updating an individual's growth curve includes the following steps: Receive corrected feeding results that include actual feed intake and non-feed loss; Write the actual feed intake into the feed intake record of the target livestock, and store the non-feed loss amount as a feeding interference record by associating it with the corresponding feeding time; Data from feeding records with stable actual feed intake and non-feed loss not meeting the abnormality marking criteria were selected as data for updating growth curves. Data showing abnormalities such as arching, overflow, vibration, or slippage are saved as reference data and are not used directly to correct individual growth curves.
10. The livestock breeding precision feeding dynamic optimization system based on individual growth curves according to claim 9, characterized in that, Dynamically adjust the target feeding amount for the next feeding cycle, including the following steps: The feed intake status and weight gain trend of the target livestock at the corresponding age were corrected based on the updated data from the growth curve. The direction of deviation of the target livestock from the target growth state is determined based on the revised individual growth curve; The feeding completion status of the target livestock is determined by combining the most recent actual feed intake, and the degree of non-feeding interference is determined by combining the most recent feeding interference record. The target feeding amount for the next feeding is adjusted based on deviation from the feeding direction, feeding completion status, and the degree of non-feeding interference, so that the decrease in non-feeding weight is not considered as effective intake and participates in the increase of continuous feeding amount.