CN-122022043-A - Livestock feed conversion ratio dynamic evaluation method and system based on multisource time sequence fusion

CN122022043ACN 122022043 ACN122022043 ACN 122022043ACN-122022043-A

Abstract

The invention provides a method and a system for dynamically evaluating the feed/meat ratio of livestock based on multi-source time sequence fusion, and relates to the technical field of livestock breeding data analysis. The method comprises the steps of establishing an individual file through ear marks, acquiring weight, feed intake and environmental data according to days, performing quality control, constructing a dynamic base line and a feature set, performing joint modeling on the weight and the feed intake based on a state space model, calculating a feed-meat ratio and a trusted interval thereof in a sliding time window, realizing abnormal identification through residual analysis and distribution detection, updating model parameters by combining incremental learning, generating feeding optimization suggestions, and realizing continuous, stable and self-adaptive evaluation on the feed-meat ratio of livestock.

Inventors

ZHANG JIE
TIAN XIANG
MA JUAN
LI HAO
KONG LINGZHUO
FANG YUE
YAO SHIQI
FENG BIN
YU XIUZHEN
An Shiguan
ZHAO CHAO
FU DONGQING
SU JIAN
WANG BAO

Assignees

新疆维吾尔自治区农业科学院

Dates

Publication Date: 20260512
Application Date: 20260131

Claims (10)

1. A method for dynamically assessing feed to meat ratio of livestock based on multi-source time series fusion, comprising: Establishing an individual file for target livestock based on ear tag identification, and acquiring weight data, feed intake data and environmental parameters of the target livestock daily to form an individual multisource daily sequence data set; Performing a consistent quality control process on the individual multisource daily sequence data sets to obtain quality controlled daily record data sets that are consistent in time granularity and used for joint modeling; Establishing a dynamic baseline according to the first column entry information and the staged health state of the target livestock, calculating a daily gain characteristic, a feeding fluctuation degree characteristic and a behavior stability characteristic based on the quality-controlled daily record data set, and carrying out steady normalization on the daily gain characteristic, the feeding fluctuation degree characteristic and the behavior stability characteristic to obtain a standardized characteristic set; Based on the quality-controlled daily record data set, the dynamic baseline and the standardized feature set, a state space model integrating a growth mechanism and statistical learning is constructed, the weight data and the feed intake data are subjected to joint modeling, a decomposition representation of a daily trend, short-term disturbance and periodic components is obtained, and a parameter set is determined based on the state space model, wherein the parameter set at least comprises an individuation parameter and a whole group sharing parameter; Calculating the individual feed-meat ratio of the target livestock based on the decomposition representation in a sliding time window, generating a trusted interval of the individual feed-meat ratio, monitoring based on residual error deviation of the state space model, and outputting abnormal indication information for early warning and decision by using chi-square residual error and mahalanobis distance; And updating the individuation parameters and the whole group sharing parameters by adopting incremental learning, generating feeding optimization suggestions and early warning information based on group comparison and distribution drift detection, and realizing dynamic evaluation and self-adaptive updating of the feed/meat ratio of the target livestock.
2. The method for dynamically estimating feed/meat ratio of livestock based on multi-source time series fusion according to claim 1, wherein daily acquisition of weight data, feed intake data and environmental parameters of the target livestock forms an individual multi-source daily sequence data set, comprising: writing a weighing result of the current day into a weight data daily record corresponding to the target livestock based on the ear tag identification; writing the current day feeding measurement result into a feeding amount data daily record corresponding to the target livestock based on the ear tag identification; and taking the acquisition date which is the same as the daily record of the weight data as the environmental parameter, and correlating the environmental parameter with the daily record of the weight data and the daily record of the feed intake data to obtain the individual multisource daily sequence data set.
3. The method of dynamic assessment of livestock feed meat ratio based on multisource time series fusion according to claim 1, wherein performing a consistent quality control process on the individual multisource daily sequence data sets to obtain quality controlled daily record data sets that are consistent in time granularity and used for joint modeling, comprises: Pairing the weight data, the feed intake data and the environmental parameters on the same date by taking the acquisition date as an index to generate a daily granularity record; identifying abnormal daily granularity records based on a preset physiological reasonable range constraint and an adjacent date change amplitude constraint, and marking the identified abnormal daily granularity records as abnormal records; Supplementing the missing daily granularity record based on the continuity constraint of adjacent dates, and keeping the consistency of the weight data and the feed intake data under the same acquisition date; And performing dimension unification conversion on the weight data, the feed intake data and the environmental parameters, and outputting the quality-controlled daily record data set.
4. The method for dynamically assessing the feed to meat ratio of livestock based on multi-source time series fusion according to claim 1, wherein establishing a dynamic baseline based on the first-time listing information and the staged health status of the target livestock comprises: Selecting the quality-controlled daily record data set of a preset baseline period after the listing date indicated by the first listing information, and determining the baseline weight and the baseline feed intake; When the staged health status of the target livestock is changed compared with the previous evaluation stage, the baseline weight and the baseline feed intake of the corresponding stage are redetermined according to the changed staged health status so as to form a dynamic baseline which can be updated along with the staged health status change; And enabling the dynamic baselines to be aligned and called with the quality-controlled daily record data set according to date for the effective date intervals corresponding to the dynamic baselines.
5. The method for dynamically assessing feed/meat ratio of livestock based on multi-source time series fusion of claim 1, wherein calculating daily gain characteristics, feeding volatility characteristics and behavioral stability characteristics based on the quality-controlled daily record data set, and performing robust normalization on the daily gain characteristics, feeding volatility characteristics and behavioral stability characteristics to obtain a standardized feature set, comprises: calculating the average daily gain based on the difference between the weight data of two adjacent days in a preset time window, and taking the average daily gain as the daily gain characteristic; Determining the feeding fluctuation degree characteristics based on the daytime fluctuation range of the feeding quantity data in a preset characteristic window; constructing a behavioral stability score based on the continuity of the daily gain characteristic and the stability of the feeding fluctuation characteristic, and taking the behavioral stability score as the behavioral stability characteristic; and normalizing the daily gain characteristic, the feeding fluctuation degree characteristic and the behavior stability characteristic by adopting a median reference and a quartile space scale, and outputting the standardized characteristic set.
6. The method for dynamically assessing feed/meat ratio of livestock based on multi-source time series fusion according to claim 1, wherein constructing a state space model for fusion of growth mechanism and statistical learning, and performing joint modeling on the weight data and the feed intake data to obtain a decomposition representation of daily trend, short-term disturbance and periodic components, comprises: Constructing a state transfer relation of representing a growth state and a feeding efficiency state by a hidden state, and enabling the hidden state to be recursively updated along with a date so as to embody a growth mechanism; Constructing an observation relation, enabling the weight data and the feed intake data to serve as joint observation of the hidden state, and introducing noise items to represent daily scale disturbance; and performing recursive estimation and smooth estimation on the hidden state to separate from the weight data and the feed intake data to obtain cross-day trend, short-term disturbance and periodic components and form the decomposition representation.
7. The method for dynamic assessment of livestock feed meat ratio based on multi-source time series fusion according to claim 1, wherein determining a set of parameters based on the state space model comprises: determining growth rate coefficients, feeding response coefficients, and disturbance sensitivity coefficients associated with the target livestock as the individualizing parameters; Determining the observed noise intensity, the state noise intensity and the period length of the period component as the whole group sharing parameter; And calculating initial statistics on the group level based on the quality-controlled daily record data set, taking the initial statistics as an initialization basis of the whole group sharing parameters, and taking initial daily records corresponding to the target livestock as an initialization basis of the individuation parameters.
8. The method for dynamically estimating feed meat ratio of livestock based on multi-source time series fusion according to claim 1, wherein calculating individual feed meat ratio of the target livestock based on the decomposed representation in a sliding time window and generating a trusted interval of the individual feed meat ratio, simultaneously monitoring based on residual error deviation of the state space model and outputting abnormality indication information for early warning and decision with chi-square residual and mahalanobis distance comprises: setting sliding time windows with the window length of 3 to 14 days continuously, and summarizing the feed intake data and the daily gain characteristic in each sliding time window to obtain window feed intake and window gain; Determining the individual feed conversion ratio according to the corresponding relation between the feed intake of the window and the weight gain of the window; generating a trusted interval of the individual feed to meat ratio based on the estimated uncertainty of the state space model for the decomposed representation; And the chi-square residual error is used for representing the overall deviation degree of the weight data and the feed intake data relative to the state space model, the mahalanobis distance is used for representing the multi-feature joint deviation degree of the standardized feature set, and the abnormal indication information comprising the abnormal grade and the abnormal occurrence date is output based on the threshold judgment result of the chi-square residual error and the mahalanobis distance.
9. The method for dynamically assessing the feed/meat ratio of livestock based on multi-source time series fusion according to claim 1, wherein the updating of the individualization parameters and the whole group sharing parameters by incremental learning and the generation of feeding optimization advice and early warning information based on group comparison and distribution drift detection comprises: After the quality-controlled day record data set of the new day is acquired, the quality-controlled day record of the new day is utilized to recursively update the personalized parameters; Determining a comparison reference interval based on individual feed conversion ratio distribution of the group level, and comparing the individual feed conversion ratio of the target livestock with the comparison reference interval to generate a group comparison conclusion; comparing the recent difference degree of the individual feed ratio distribution and the historical individual feed ratio distribution in a preset drift window, and judging that the distribution drift occurs when the difference degree exceeds a preset threshold value; And generating feeding optimization suggestions and early warning information comprising adjusting target feeding level, rechecking the staged health state and checking the fluctuation of environmental parameters according to the group comparison conclusion, the distribution drift judgment and the abnormality indication information.
10. A multi-source time series fusion based livestock feed meat ratio dynamic assessment system, comprising: The individual data acquisition and multisource sequence construction unit is used for establishing an individual file for target livestock based on the ear marks, acquiring weight data, feed intake data and environmental parameters of the target livestock daily, and forming an individual multisource daily sequence data set; A multi-source data consistency quality control unit for performing a consistency quality control process on the individual multi-source daily sequence data sets to obtain quality controlled daily record data sets with consistent time granularity for joint modeling; The dynamic baseline construction and feature extraction unit is used for establishing a dynamic baseline according to the first listing information and the staged health state of the target livestock, calculating daily gain features, feeding fluctuation degree features and behavior stability features based on the quality-controlled daily record data set, and carrying out steady normalization on the daily gain features, the feeding fluctuation degree features and the behavior stability features to obtain a standardized feature set; The state space joint modeling unit is used for constructing a state space model integrating a growth mechanism and statistical learning based on the quality-controlled daily record data set, the dynamic baseline and the standardized feature set, carrying out joint modeling on the weight data and the feed intake data to obtain a decomposition representation of daily trend, short-term disturbance and periodic components, and determining a parameter set based on the state space model, wherein the parameter set at least comprises an individuation parameter and a whole group sharing parameter; The feed conversion ratio evaluation and abnormality judgment unit is used for calculating the individual feed conversion ratio of the target livestock based on the decomposition representation in a sliding time window, generating a trusted interval of the individual feed conversion ratio, monitoring based on residual error deviation of the state space model, and outputting abnormality indication information for early warning and decision by using chi-square residual error and mahalanobis distance; And the parameter self-updating and feeding decision generating unit is used for updating the personalized parameters and the whole group sharing parameters by adopting incremental learning, generating feeding strategy optimization suggestions and early warning information based on group comparison and distribution drift detection, and realizing dynamic evaluation and self-adaptive updating of the target livestock feed/meat ratio.

Description

Livestock feed conversion ratio dynamic evaluation method and system based on multisource time sequence fusion Technical Field The invention relates to the technical field of livestock breeding data analysis, in particular to a method and a system for dynamically evaluating the feed/meat ratio of livestock based on multi-source time sequence fusion. Background The feed conversion ratio is used for representing the relationship between feed consumption and weight gain of livestock in a certain period, and is a key index for feeding management and production benefit accounting. Common practice in the production at the present stage is to weigh weekly or stage by stage, and to calculate statistically in combination with the accumulated feed intake at the stage, or to evaluate with average data at the group level. With the gradual application of means such as ear tag identification, automatic feeding metering, automatic weighing and environmental monitoring, the individual weight and the feeding amount can form daily scale continuous records, and a foundation is provided for finer granularity evaluation. Cultivation management gradually goes from stage statistics to individualized, continuous and refined dynamic evaluation. The actual demand not only comprises the acquisition of a feed conversion ratio value at a certain stage, but also needs to stably describe the change trend of the feed conversion ratio under the conditions of growth stage change, feeding fluctuation, environmental disturbance and the like, and the evaluation result is used for feeding strategy adjustment, abnormal early warning and historical tracing, so that the aim of carrying out accurate management on individuals to target analysis on groups is realized. In the existing method, the weight and the feed intake are directly combined according to time periods to obtain the ratio, so that the evaluation stability is difficult to maintain when noise, missing measurement and short-term fluctuation exist in daily scale data, and the influence of long-term growth trend and short-term disturbance on the feed conversion ratio is difficult to distinguish. In addition, the lack of individual base lines updated with initial and staged health status changes in the fence makes the same individual have insufficient comparability in different growth stages, and the lack of self-adaptive mechanism capable of being simultaneously applicable to individual difference and group commonality changes leads to easy distortion of evaluation results when group structure changes or feeding conditions change, thereby affecting timeliness and accuracy of feeding decisions. Disclosure of Invention In order to overcome the defects of the prior art, the invention aims to provide the dynamic evaluation method and the system for the feed-to-meat ratio of the livestock based on the multi-source time sequence fusion, and the stable calculation and the change depiction of the feed-to-meat ratio of the livestock in the continuous growth process are realized by constructing a dynamic evaluation mechanism based on the multi-source time sequence, so that the accuracy and the reliability of the feed-to-meat ratio evaluation under the complex feeding condition are effectively improved. In order to achieve the above object, the present invention provides the following solutions: a method for dynamic assessment of livestock feed conversion ratio based on multi-source time series fusion, comprising: Establishing an individual file for target livestock based on ear tag identification, and acquiring weight data, feed intake data and environmental parameters of the target livestock daily to form an individual multisource daily sequence data set; Performing a consistent quality control process on the individual multisource daily sequence data sets to obtain quality controlled daily record data sets that are consistent in time granularity and used for joint modeling; Establishing a dynamic baseline according to the first column entry information and the staged health state of the target livestock, calculating a daily gain characteristic, a feeding fluctuation degree characteristic and a behavior stability characteristic based on the quality-controlled daily record data set, and carrying out steady normalization on the daily gain characteristic, the feeding fluctuation degree characteristic and the behavior stability characteristic to obtain a standardized characteristic set; Based on the quality-controlled daily record data set, the dynamic baseline and the standardized feature set, a state space model integrating a growth mechanism and statistical learning is constructed, the weight data and the feed intake data are subjected to joint modeling, a decomposition representation of a daily trend, short-term disturbance and periodic components is obtained, and a parameter set is determined based on the state space model, wherein the parameter set at least comprises an individuation parameter a