Panda life cycle weight prediction method and system based on staged adaptive integration

Abstract

The invention provides a staged adaptive integrated giant panda life cycle weight prediction method and system, and relates to the technical field of big data mining, and the method comprises the steps: obtaining original data; carrying out digestion cycle dynamic modeling according to the original data, and obtaining a real-time gastric content mass estimation value by constructing a time-varying differential equation with gastric emptying dynamic characteristics; according to the real-time gastric content mass estimation value, carrying out fasting weight inversion to obtain a basic fasting weight sequence; performing growth stage adaptive division according to the basic empty stomach body weight sequence to obtain a stage standardized growth baseline; performing multi-model integrated prediction according to the stage standardized growth base line to obtain a weight prediction value; and life cycle uncertainty quantification is carried out according to the weight prediction value, and a life cycle weight prediction pedigree is generated. According to the method, the long-standing problems of poor stage adaptability, large data noise interference and high prediction uncertainty in panda weight prediction are solved.

Description

Technical Field

[0001] This invention relates to the field of big data mining technology, and more specifically, to a phased adaptive integrated method and system for predicting the life cycle weight of giant pandas. Background Technology

[0002] In modern giant panda ex-situ conservation and breeding research institutions, captive management has entered a new digital stage. Individual weight, as a core indicator for assessing the developmental and health status of giant pandas, directly impacts the timeliness of feeding strategies and medical interventions due to its predictive accuracy. Currently, numerous automated monitoring devices have been installed in the breeding enclosures for giant pandas ex-situ conservation and breeding, continuously recording the weight, behavior, and even local environmental parameters of individual pandas; the keepers' logs have also been digitized, detailing the type and precise weight of food given each time. Theoretically, this high-frequency, multi-dimensional data collection provides an unprecedented data foundation for a deeper understanding of giant panda individual development and the formulation of personalized feeding programs. However, faced with massive amounts of heterogeneous time-series data, keepers and researchers are caught in a typical dilemma of "rich data, but lacking insight."

[0003] The core of this dilemma stems from the data complexity arising from two intertwined but vastly different biological cycles in the giant panda's physiological activities. The first is the "life cycle," from birth to old age, where giant pandas experience cub, juvenile, sub-adult, and adult stages, each with fundamentally different growth drivers and patterns: cubs exhibit exponential growth primarily driven by nutrient intake; sub-adults show significant sexual dimorphism in growth rate peaks; and in adulthood, weight changes are more governed by regular fluctuations dominated by the reproductive cycle and seasonal changes, with even older pandas potentially experiencing atypical weight loss due to factors such as tooth wear. The second is the "digestive cycle," measured in hours or days—the continuous physiological process of food intake, digestion, and emptying. After each meal, the mass of stomach contents is added to the scale reading in real time, resulting in a mixed signal of actual body mass and instantaneous digestive system load. The superposition of these two cycles makes the original continuous weight monitoring sequence a complex curve that is heavily contaminated by high-frequency, large-amplitude "noise" (digestive cycle load) while also containing low-frequency, long-term "trend" (life cycle growth).

[0004] Traditional weight prediction methods mostly rely on static growth models or single machine learning algorithms. While they can achieve certain results within specific age groups, they suffer from three main shortcomings when considering the significant stage-specific growth characteristics throughout the giant panda's lifespan: First, the fixed-age segmentation strategy is inherently difficult to adapt to individual differences in developmental pace, particularly exhibiting low predictive reliability during data-sparse stages such as adulthood and old age. Second, at the data level, the interference of digestion on real-time weight measurement is not fully considered, and there is a lack of effective calibration mechanisms for fasting weight. Third, at the model architecture level, the rigid model integration method cannot dynamically adjust prediction weights according to growth stages, leading to systematic biases at physiological turning points. These shortcomings of existing technologies stem from inherent flaws in their methodological principles, and even conventional improvements such as parameter optimization or model replacement cannot fundamentally eliminate these limitations. Traditional methods or static models struggle to effectively separate two effects: ignoring the digestive cycle and directly using raw body weight for growth analysis severely overestimates daily fluctuations, making it impossible to accurately determine whether long-term growth and development are healthy or off track; using fixed-age segmentation models to handle the life cycle fails to adapt to individual differences in developmental pace (e.g., precocious or late-maturing individuals), especially in data-sparse late adulthood or old age, where predictions are often unreliable. Therefore, accurately separating the pure trend representing an individual's long-term growth and development from time-series data heavily influenced by daily physiological activity noise, and constructing a model that adapts to different life stages and quantifies predictive uncertainty, has become a critical technological bottleneck in the precise health management of captive giant pandas. This invention aims to systematically overcome this data analysis and prediction challenge caused by the "dual-cycle superposition effect."

[0005] Based on the shortcomings of the existing technologies, there is an urgent need for a phased adaptive integrated method and system for predicting the life cycle weight of giant pandas. Summary of the Invention

[0006] The purpose of this invention is to provide a phased adaptive integrated method and system for predicting the life-cycle weight of giant pandas, in order to improve the aforementioned problems. To achieve the above objective, the technical solution adopted by this invention is as follows: Firstly, this application provides a phased adaptive ensemble method for predicting the life-cycle weight of giant pandas, including: Obtain raw data, including continuous weight monitoring sequences, feeding behavior logs, seasonal climate indicators, and individual characteristics of giant pandas; Based on the raw data, a dynamic model of the digestion cycle is performed. By constructing a time-varying differential equation with gastric emptying dynamics, the mass decay process of food from intake to complete digestion is simulated, and the real-time gastric contents mass estimate is obtained. Fasting weight is inverted based on real-time gastric contents mass estimates. The basal fasting weight sequence is obtained by subtracting the digestive system load from the measured weight values ​​point by point. Based on the aforementioned basic fasting weight sequence, growth stages are adaptively divided. The weight sequence is separated into age-related growth trend components and stage fluctuation components using a variational mode decomposition algorithm to obtain a stage-standardized growth baseline. Based on the standardized growth baseline of the aforementioned stage, multi-model ensemble prediction is performed. By constructing a stage-adaptive weighted ensemble network, the prediction model weights for different growth stages are optimized to obtain ensemble-enhanced weight prediction values. Based on the predicted weight, the uncertainty of the life cycle is quantified, and a Bayesian probabilistic graphical model is used to integrate the statistical characteristics of historical observation data with multi-stage prediction results to generate a life cycle weight prediction spectrum with confidence intervals.

[0007] Secondly, this application also provides a phased adaptive integrated giant panda lifecycle weight prediction system, including: The acquisition module is used to acquire raw data, which includes the giant panda's continuous weight monitoring sequence, feeding behavior log, seasonal climate indicators and individual characteristics; The modeling module is used to perform dynamic modeling of the digestion cycle based on the raw data. By constructing a time-varying differential equation with gastric emptying dynamics, it simulates the mass decay process of food from intake to complete digestion and obtains a real-time estimate of the mass of gastric contents. The inversion module is used to invert fasting weight based on real-time gastric contents mass estimates. It obtains the baseline fasting weight sequence by subtracting the digestive system load point by point from the measured weight values. The segmentation module is used to adaptively segment the growth stages based on the basic fasting weight sequence. The weight sequence is separated into age-related growth trend components and stage fluctuation components through variational mode decomposition algorithm to obtain stage-standardized growth baselines. The prediction module is used to perform multi-model integrated prediction based on the stage-standardized growth baseline. By constructing a stage-adaptive weighted ensemble network, the prediction model weights for different growth stages are optimized to obtain an integrated and enhanced weight prediction value. The output module is used to quantify the life cycle uncertainty based on the predicted weight value, and to integrate the statistical characteristics of historical observation data with multi-stage prediction results using a Bayesian probabilistic graphical model to generate a life cycle weight prediction spectrum with confidence intervals.

[0008] The beneficial effects of this invention are as follows: This invention solves the long-standing problems of poor stage adaptability, large data noise interference, and high prediction uncertainty in giant panda weight prediction through the collaborative innovation of phased processing and adaptive integration, providing a scientific basis and technical support for the precise health management of ex-situ protected captive giant pandas. Attached Figure Description

[0009] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart illustrating a staged adaptive integrated method for predicting the life-cycle weight of giant pandas, as described in an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of a phased adaptive integrated giant panda life cycle weight prediction system as described in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a phased adaptive integrated giant panda life cycle weight prediction device as described in an embodiment of the present invention.

[0011] The diagram is labeled as follows: 800, a phased adaptive integrated giant panda life cycle weight prediction device; 801, processor; 802, memory; 803, multimedia component; 804, I / O interface; 805, communication component; 901, acquisition module; 902, modeling module; 903, inversion module; 904, partitioning module; 905, prediction module; 906, output module. Detailed Implementation

[0012] 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, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0013] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0014] Example 1:

[0015] This embodiment provides a phased adaptive integrated method for predicting the life cycle weight of giant pandas.

[0016] See Figure 1 The figure shows that the method includes steps S100 to S600.

[0017] Step S100: Obtain raw data, which includes the giant panda's continuous weight monitoring sequence, feeding behavior log, seasonal climate indicators, and individual characteristics; In actual captive management, giant panda weight monitoring sequences are often affected by momentary physiological activities such as feeding, while seasonal climate indicators (such as temperature, humidity, and daylight hours) and individual characteristics (such as age, sex, and health status) significantly influence metabolic rate and digestive efficiency, making it impossible for a single weight value to accurately reflect long-term growth trends. Therefore, relying solely on isolated weight measurements for health assessment has significant limitations. The systematic data acquisition scheme in step S100 aims to build a multi-dimensional, time-series data foundation capable of supporting subsequent refined modeling. Its purpose is to lay the foundation for adaptive growth stage segmentation and stage-specific model integration from the data source.

[0018] The specific data collection process is as follows: Continuous weight monitoring is typically achieved through automated weighing platforms installed in the enclosures or activity corridors. The weight is automatically recorded as the pandas pass through these platforms multiple times daily, generating high-frequency time-series data. Feeding behavior logs are meticulously recorded by keepers, including the time of each feeding, the type of food (e.g., bamboo shoots, bamboo leaves, concentrated feed supplements), and its specific weight, thus accurately depicting the temporal distribution and quality of ingested food. Seasonal climate indicators are primarily derived from meteorological station data at the location of the enclosure, including daily records of parameters such as ambient temperature, humidity, and daylight hours, used to quantify the impact of the external environment on the pandas' physiological rhythms. Individual characteristics are derived from historical health records, including resting metabolic rate assessed using techniques such as indirect calorimetry, individual digestive efficiency estimated by combining daily feed intake and weight changes, and basic information such as age, sex, and health status. The simultaneous collection and correlation of these multimodal data lays a solid foundation for establishing a dynamic calibration model that reflects individual differences and spatiotemporal variations.

[0019] Step S200: Perform dynamic modeling of the digestion cycle based on the original data. By constructing a time-varying differential equation with gastric emptying dynamics, simulate the mass decay process of food from intake to complete digestion and obtain a real-time estimate of the mass of gastric contents. Understandably, this step transforms physiological mechanisms into a computable mathematical model. By constructing a time-varying differential equation for gastric emptying dynamics, the abstract digestive process is visualized as a mass change trajectory that decays over time. The aim is to quantify the dynamic impact of food intake on real-time weight measurements, providing a theoretical basis and quantitative tool for reconstructing the fasting state from non-fasting measurements.

[0020] Step S300: Perform fasting weight inversion based on the real-time gastric contents mass estimate. Obtain the baseline fasting weight sequence by subtracting the digestive system load from the measured weight values ​​point by point. It should be noted that this step is based on the idea of ​​physical inversion, which treats the estimated value of gastric contents as a kind of "noise" or "interference". By designing a reverse calculation process, this digestive system load is accurately separated from the measured weight at each moment, so as to obtain a weight benchmark sequence that is closer to the true fasting state. This step directly determines the accuracy of fasting weight estimation.

[0021] Step S400: Based on the baseline fasting weight sequence, adaptive division of growth stages is performed. The weight sequence is separated into age-related growth trend components and stage fluctuation components by variational mode decomposition algorithm to obtain stage-standardized growth baseline. Understandably, the purpose of this step is to identify the non-stationary characteristics of weight changes in the giant panda's life cycle. The algorithm can adaptively capture key physiological turning points from cub stage, juvenile stage to adult stage, avoiding the rigid constraints of traditional fixed age segmentation, and thus more accurately reflecting the differences in individual developmental rhythms.

[0022] Step S500: Perform multi-model ensemble prediction based on the stage-standardized growth baseline. By constructing a stage-adaptive weighted ensemble network, optimize the prediction model weights for different growth stages to obtain the ensemble-enhanced weight prediction value. It should be noted that, in response to the stage-specific heterogeneity in giant panda weight changes, such as the rapid growth during the cub stage and weight fluctuations during adulthood, this step integrates the advantages of each stage-specific model through a dynamic weight allocation mechanism to ensure the smoothness and accuracy of the prediction results in the stage transition region.

[0023] Step S600: Quantify the life cycle uncertainty based on the predicted weight value, and use a Bayesian probabilistic graphical model to integrate the statistical characteristics of historical observation data with multi-stage prediction results to generate a life cycle weight prediction spectrum with confidence intervals.

[0024] It should be noted that this step aims to address the unevenness of data collection in captive environments, especially in the data-sparse adult and elderly stages, by quantifying the uncertainty of prediction through a probabilistic framework, and providing decision support for risk visualization in weight monitoring.

[0025] Further, step S200 includes steps S210 to S230.

[0026] Step S210: Extract eating event features based on the raw data. By constructing a multimodal data alignment model based on timestamp association, spatiotemporally match discrete eating behavior logs with continuous weight monitoring sequences to obtain eating event vectors with temporal tags. Step S220: Based on the eating event vector, perform gastric emptying dynamics modeling. By establishing a variable parameter differential equation that considers the coupling of food type and seasonal metabolic rate, simulate the gastric contents mass decay trajectory at different digestion stages and obtain the theoretical gastric contents mass curve. Step S230: Based on the theoretical gastric contents mass curve, real-time correction is performed. By introducing a feedback regulation mechanism based on the slope of body weight change, the emptying rate parameter is dynamically adjusted to obtain a real-time gastric contents mass estimate that matches the actual metabolic characteristics.

[0027] Specifically, step S210 first addresses the core issue of the mismatch between the discreteness of giant panda feeding behavior and the continuity of weight monitoring. By constructing a timestamp-linked multimodal data alignment model, it precisely matches the feeding behavior logs recorded by keepers (including feeding time and food weight) with the continuous weight monitoring sequences collected by automated equipment. This process essentially uses timestamps as a bridge to find the corresponding weight monitoring point on a continuous time axis for each discrete feeding event, thereby generating a feeding event vector with temporal markers. This step integrates the originally isolated and asynchronous data sources into a unified dataset with strict temporal correlation, laying an accurate input foundation for subsequent dynamic modeling. The formula for generating the feeding event vector is: ; In the formula, v i Let t be the feeding event vector; i w is the timestamp of the i-th feeding event; i Let b be the weight of the food in the i-th feeding event; i For time t i Weight values ​​obtained from a weight monitoring sequence through linear interpolation.

[0028] Based on this, step S220 constructs a variable-parameter differential equation, namely a time-varying differential equation with gastric emptying dynamics. The parameters of this equation are not fixed but simultaneously consider the coupled effects of food type (such as the physical properties and nutritional components of different foods like bamboo shoots and bamboo leaves) and seasonal metabolic rate (such as periodic physiological changes like slower metabolism in winter and more active metabolism in summer). Through this dynamic adjustment mechanism, the model can more realistically simulate the nonlinear trajectory of gastric contents decay over time from the start of feeding to complete digestion and emptying, thus outputting a theoretically more accurate curve of gastric contents mass change. Compared to simpler models with fixed parameters, this model is better adapted to the varying digestive physiology of giant pandas due to seasonal and dietary changes. The gastric contents mass decay model (time-varying differential equation with gastric emptying dynamics) is expressed as: ; In the formula, m(t) is the mass of stomach contents at the current time t; V is the set of feeding event vectors; For food type f i The base emptying rate; s(τ) is the seasonal metabolic rate factor used to simulate periodic fluctuations; τ is the historical time; d is the differential symbol.

[0029] Step S230 introduces a feedback adjustment mechanism based on the slope of weight change for real-time correction. This mechanism works by continuously monitoring the first difference (i.e., the instantaneous rate of change) of the actual weight sequence after a feeding event and comparing it in real-time with the weight change trend predicted by the theoretical model. When a systematic deviation occurs between the slope of the actual weight loss and the theoretical prediction, the feedback mechanism dynamically adjusts the emptying rate parameter in the differential equation. For example, if the actual weight loss rate is consistently slower than the prediction, the emptying rate parameter is lowered, slowing down the digestive process simulated by the model; conversely, the parameter is raised. This closed-loop feedback control allows the model to adapt to the actual metabolic characteristics of individual giant pandas at specific times (such as individual differences in digestive efficiency, the potential impact of disease or special physiological states on digestion), ultimately outputting a more reliable real-time estimate of gastric contents mass that matches the individual's real-time physiological state. The real-time calibration model for gastric contents mass is expressed as follows: ; In the formula, m calibrated (t) represents the estimated mass of gastric contents after real-time calibration; η is the calibration gain, used to control the intensity of calibration; This represents the slope of the actual weight change. This is the slope of the change in gastric contents mass predicted by the theoretical model, i.e., the derivative of m(t).

[0030] Further, step S300 includes steps S310 to S330.

[0031] Step S310: Quantify the digestive load based on the real-time gastric contents mass estimate. By establishing a load distribution model based on the time decay function, the gastric contents mass estimate is converted into the instantaneous digestive system load corresponding to the measured body weight time point, thus obtaining the load sequence. Step S320: Dynamically adjust the load according to the load sequence. By introducing seasonal metabolic coefficient and individual digestive efficiency factor, the standard load is adaptively adjusted based on physiological characteristics to obtain the individualized adjusted real-time load. Step S330: Generate a fasting weight sequence based on real-time load. By constructing a weight-optimized load stripping model, the dynamically changing digestive system load is deducted point by point from the measured weight value to obtain a continuous basal fasting weight sequence.

[0032] Specifically, steps S310 to S330 together constitute a refined processing chain from estimating the mass of gastric contents to generating a baseline fasting body weight sequence. The core task of step S310 is to address the distribution of gastric contents mass over time, achieved by establishing a load distribution model based on a time decay function. The basic principle of this model is to dynamically allocate the estimated gastric contents mass obtained in step S200 to subsequent time points according to a decay function describing the digestive system's processing rate, thereby calculating the instantaneous digestive system load contributed by the not-yet-completely-emptied gastric contents at each recorded body weight. This processing transforms the mass estimate at a single time point into a continuously varying load sequence over time, providing a time-aligned data basis for subsequent accurate deduction. Preferably, a non-homogeneous Poisson process is used to simulate the random distribution of load over time, avoiding the limitations of a simple decay function. The random load distribution model is expressed as: ; In the formula, L(t) is the instantaneous digestive system load at time t, representing the random accumulation of unemptied gastric contents; λ(t) is the intensity function used to describe the rate of digestive events at time t; λ(τ) is the intensity function used to describe the rate of digestive events at time τ. Step S320 then introduces physiological adaptive adjustments. Since the digestive efficiency of giant pandas is not constant but significantly influenced by seasonal metabolic fluctuations (such as slower metabolism in winter) and individual differences (such as age and health status), step S320 corrects the "standard" load sequence calculated by the general decay model by introducing two parameters: a seasonal metabolic coefficient and an individual digestive efficiency factor. The seasonal metabolic coefficient quantifies the impact of changes in the overall metabolic level of giant pandas under different seasonal environments on emptying rate, while the individual digestive efficiency factor characterizes the difference in digestive capacity of a specific individual relative to the average level. Through this correction, the load sequence can more accurately reflect the true physiological load state of a specific individual in a specific season, obtaining the individualized corrected real-time load. Preferably, this correction process is dynamically corrected using a state-space model, treating the load as a hidden state and the observed values ​​as physiological parameters. The state-space correction model is expressed as: Equations of state: ; Observation equation: ; In the formula, L adj (t) represents the corrected individualized real-time load, as a hidden state; A(t) is the state transition matrix, modeled as a time-varying function to reflect digestion kinetics; B(t) is the input matrix; E s (t) is the seasonal metabolic coefficient; E i denoted as the individual digestion efficiency factor; w(t) represents process noise, v(t) represents observation noise, both assumed to be Gaussian white noise; z(t) represents the observed value; and H represents the observation matrix.

[0033] The load stripping model in step S330 assigns a dynamic weight to the real-time load at each time point. This weight may be determined based on data quality (such as measurement error), model confidence at that time point, or other optimization criteria. Then, the model uses this weighted load to perform point-by-point, biased deduction from the corresponding measured load values. Preferably, this process defines load stripping as a constrained optimization problem, minimizing the expected value of the deduction error, and introducing a confidence constraint. The load stripping model is expressed as: ; Constraints: ; In the formula, B represents the expected value; baseline (t) is the baseline fasting body weight, used as the optimization target; B measured (t) represents the measured body weight; Ω(t) represents the dynamic weighting coefficient. Indicates variance; This represents the maximum permissible variance.

[0034] Further, step S400 includes steps S410 to S430.

[0035] Step S410: Detect growth stage boundaries based on the basic fasting weight sequence. By analyzing the curvature change characteristics of the weight sequence and combining it with the time points of key physiological events in giant pandas, dynamically identify the division boundaries of the cub stage, juvenile stage, sub-adult stage, and adult stage, and obtain the stage division threshold. Step S420: Perform stage-specific signal decomposition based on the stage division threshold. By constructing an adaptive variational mode decomposition model and setting a mode parameter adjustment mechanism based on the growth stage, the weight sequence is separated into age-related growth trend components and stage fluctuation components, and preliminary stage decomposition results are obtained. Step S430: Based on the preliminary stage decomposition results, perform stage consistency verification. By establishing a modal fusion model based on growth curve similarity, reorganize and optimize cross-stage components to obtain a stage-standardized growth baseline that conforms to the continuity of growth and development.

[0036] Specifically, step S410 analyzes the curvature change characteristics of the basic fasting weight sequence and combines it with key physiological events in giant pandas (such as weaning and sexual maturity) to dynamically identify growth stage boundaries. The curvature change characteristics are used to capture inflection points in the weight sequence, while the physiological event time points provide prior knowledge constraints, thus obtaining the stage division threshold. This processing addresses the non-stationarity of weight changes throughout the giant panda's life cycle, avoiding the limitations of traditional fixed-age segmentation and adaptively responding to individual developmental rhythm differences. Based on the stage division threshold, step S420 performs stage-specific signal decomposition using an adaptive variational mode decomposition model. This model separates the weight sequence into age-related growth trend components and stage-specific fluctuation components by setting a modal parameter adjustment mechanism based on the growth stage (e.g., increasing the number of modes in the cub stage to capture rapid changes and decreasing the number of modes in adulthood to highlight trends), obtaining preliminary staged decomposition results. Variational mode decomposition is a signal processing technique that adaptively separates different frequency components. In this embodiment, it adapts to the specific fluctuation patterns of each growth stage, ensuring that the decomposition results retain both global trends and capture local details. Step S430 performs stage consistency verification on the preliminary staged decomposition results. A modal fusion model based on growth curve similarity is established. This model uses the morphological similarity of historical growth curves as constraints to renormalize and optimize cross-stage components, eliminating potential boundary discontinuities or modal aliasing introduced by the decomposition. Ultimately, a stage-standardized growth baseline conforming to the continuity of growth and development is obtained. This verification mechanism ensures the physiological rationality of the analysis results, particularly addressing the smoothness of stage transition regions, thus providing reliable input for subsequent multi-model integrated prediction. The entire process embodies a progressive processing logic from boundary detection to signal decomposition and then to consistency optimization, progressively improving the accuracy and robustness of weight sequence analysis.

[0037] Further, step S500 includes steps S510 to S530.

[0038] Step S510: Train stage-specific models based on the stage-standardized growth baseline. Train gradient boosting tree prediction models for different growth stages, namely the cub stage, juvenile stage, sub-adult stage and adult stage, and optimize hyperparameters based on the data distribution characteristics of each stage to obtain a set of stage-specific prediction models. Step S520: Perform dynamic weight integration based on the set of stage-specific prediction models. By constructing a stage-adaptive attention network, the model weights are dynamically allocated according to the real-time growth stage identifier and prediction uncertainty to obtain the preliminary integrated prediction value. Step S530: Perform cross-stage consistency optimization based on the preliminary integrated prediction values. By introducing time series smoothing constraints and physiological rationality verification mechanisms, the stage transition region of the prediction results is corrected to obtain the integrated and enhanced weight prediction values.

[0039] Specifically, steps S510 to S530 together constitute a multi-stage adaptive ensemble framework for giant panda lifecycle weight prediction. Its core lies in overcoming the limitations of traditional methods in lifecycle prediction through staged modeling and dynamic weight adjustment. Step S510 first trains stage-specific models, training gradient boosting tree prediction models for different growth stages such as cub stage, juvenile stage, sub-adult stage, and adult stage. This training process optimizes hyperparameters based on the data distribution characteristics of each stage. For example, since cub stage data fluctuates significantly, a deeper tree structure is used to capture non-linear features, while adult stage data is relatively stable, focusing on regularization to prevent overfitting, thus obtaining a set of stage-specific prediction models. This design fully considers the growth and development characteristics of giant pandas at different physiological stages, and the stage-specific models can more accurately capture these specific patterns.

[0040] Step S520 builds upon this foundation by performing dynamic weight integration. This involves constructing a stage-adaptive attention network. This network dynamically allocates weights to each stage model based on real-time growth stage identification and prediction uncertainty. For example, when an individual is transitioning from childhood to pre-adulthood, the system simultaneously uses the prediction results from both the childhood and pre-adulthood models. It calculates weighting coefficients based on the matching degree between the current data point and the features of each stage, while also considering the model's historical prediction errors in that region, thus obtaining a preliminary integrated prediction value. This dynamic weighting mechanism effectively overcomes the lack of adaptability of fixed-weight integration in stage transition regions.

[0041] Step S530 concludes with cross-stage consistency optimization. By introducing time series smoothing constraints and physiological rationality verification mechanisms, the transitional regions of the initial integrated predictions are corrected. The time series smoothing constraints ensure that the predicted values ​​at adjacent time points do not exhibit drastic jumps, while the physiological rationality verification corrects abnormal fluctuations based on the general patterns of giant panda growth curves. This ultimately yields the integrated and enhanced weight prediction values. The entire processing flow, through a progressive design of stage division, dedicated modeling, dynamic integration, and consistency optimization, achieves accurate prediction of weight changes throughout the entire life cycle, demonstrating particularly good adaptability and robustness in the transitional regions where data distribution is uneven.

[0042] Further, step S600 includes steps S610 to S630.

[0043] Step S610: Perform multi-source uncertainty analysis based on the predicted weight value. By constructing a multi-factor variation model based on the seasonal metabolic fluctuations of giant pandas, individual developmental heterogeneity, and measurement errors, the main sources of uncertainty at different growth stages and their interactions are quantified to obtain a set of structured uncertainty components. Step S620: Construct a dynamic Bayesian network based on the set of structured uncertainty components. Integrate the statistical distribution characteristics of historical observation data with real-time prediction results by establishing a hierarchical probabilistic graphical model. Introduce time series dependencies and specific physiological constraints of giant pandas to obtain the posterior probability weight distribution. Step S630: Generate prediction intervals based on the posterior probability weight distribution. Calculate the weight prediction boundaries at different confidence levels using the quantile regression forest algorithm, and dynamically calibrate the interval width by incorporating the growth stage adaptive adjustment mechanism. Finally, generate a life-cycle weight prediction spectrum with confidence intervals.

[0044] Specifically, step S610 constructs a multi-factor variation model to systematically analyze three main sources of uncertainty in giant panda weight prediction: seasonal metabolic fluctuations (such as deviations in weight change patterns caused by slower metabolism in winter), individual developmental heterogeneity (such as significant differences in growth rates among individuals of the same age), and measurement errors (such as the precision limitations of automated weighing equipment). This model quantifies the contribution and interaction of each uncertainty source using variance decomposition technology, generating a structured set of uncertainty components, providing a clear input dimension for subsequent probabilistic modeling. Step S620 builds a dynamic Bayesian network based on this, employing a hierarchical probabilistic graphical structure to fuse the statistical distribution characteristics of historical observation data (such as mean, variance, and skewness) with real-time prediction results. The network structure specifically introduces time-series dependencies to capture the temporal correlation of weight changes, while embedding specific physiological constraints of giant pandas (such as the impact mechanism of the reproductive cycle on weight). The posterior probability distribution is calculated using a variational inference algorithm to obtain the probability weight allocation for each time point. This processing effectively solves the problem of insufficient prediction reliability of traditional methods in data-sparse stages (such as adulthood and old age), enabling the model to adaptively adjust the confidence level of information from different sources. Step S630 finally generates prediction intervals using the quantile regression forest algorithm. This algorithm adaptively weights the data based on the posterior probability weight distribution, calculates the weight prediction boundaries at different confidence levels, and innovatively introduces a growth stage adaptive adjustment mechanism to dynamically calibrate the interval width according to the individual's current physiological stage. The final output is a statistically significant and physiologically reasonable life-cycle weight prediction spectrum. The entire technology chain, through a progressive process of "uncertainty analysis - probability fusion - interval generation," achieves a quantitative assessment of the reliability of the prediction results, enabling accurate weight prediction for giant pandas throughout their entire life cycle from birth to old age, providing decision support for husbandry and management.

[0045] Example 2:

[0046] like Figure 2 As shown, this embodiment provides a phased adaptive integrated giant panda lifecycle weight prediction system, the system including: The acquisition module 901 is used to acquire raw data, including continuous weight monitoring sequences, feeding behavior logs, seasonal climate indicators, and individual characteristics of giant pandas. Modeling module 902 is used to perform dynamic modeling of the digestion cycle based on raw data. By constructing a time-varying differential equation with gastric emptying dynamics, it simulates the mass decay process of food from intake to complete digestion and obtains a real-time estimate of the mass of gastric contents. The inversion module 903 is used to invert fasting weight based on the real-time gastric contents mass estimate. By subtracting the digestive system load point by point from the measured weight value, the basic fasting weight sequence is obtained. The segmentation module 904 is used to adaptively segment the growth stages based on the baseline fasting weight sequence. The weight sequence is separated into age-related growth trend components and stage fluctuation components through variational mode decomposition algorithm to obtain the stage-standardized growth baseline. The prediction module 905 is used to perform multi-model integrated prediction based on the stage-standardized growth baseline. By constructing a stage-adaptive weighted ensemble network, the prediction model weights for different growth stages are optimized to obtain the integrated and enhanced weight prediction value. Output module 906 is used to quantify life cycle uncertainty based on weight prediction values. It integrates the statistical characteristics of historical observation data with multi-stage prediction results using a Bayesian probabilistic graphical model to generate a life cycle weight prediction spectrum with confidence intervals.

[0047] In one specific embodiment of this application, the modeling module 902 includes: The first modeling unit is used to extract eating event features from the raw data. By constructing a multimodal data alignment model based on timestamp association, it spatiotemporally matches discrete eating behavior logs with continuous weight monitoring sequences to obtain eating event vectors with temporal tags. The second modeling unit is used to model the gastric emptying dynamics based on the eating event vector. By establishing a variable parameter differential equation that considers the coupling between food type and seasonal metabolic rate, it simulates the gastric contents mass decay trajectory at different digestion stages and obtains the theoretical gastric contents mass curve. The third modeling unit is used to make real-time corrections based on the theoretical gastric contents mass curve. By introducing a feedback regulation mechanism based on the slope of weight change, the emptying rate parameter is dynamically adjusted to obtain a real-time gastric contents mass estimate that matches the actual metabolic characteristics.

[0048] In one specific embodiment of this application, the inversion module 903 includes: The first inversion unit is used to quantify the digestive load based on the real-time gastric contents mass estimate. By establishing a load distribution model based on the time decay function, the gastric contents mass estimate is converted into the instantaneous digestive system load corresponding to the measured body weight time point, thus obtaining the load sequence. The second inversion unit is used to dynamically adjust the load based on the load sequence. By introducing seasonal metabolic coefficients and individual digestive efficiency factors, the standard load is adaptively adjusted based on physiological characteristics to obtain the individualized adjusted real-time load. The third inversion unit is used to generate a fasting weight sequence based on real-time load. By constructing a weight-optimized load stripping model, the dynamically changing digestive system load is deducted point by point from the measured weight value to obtain a continuous basal fasting weight sequence.

[0049] In one specific embodiment of this application, the partitioning module 904 includes: The first segmentation unit is used to detect the growth stage boundaries based on the basic fasting weight sequence. By analyzing the curvature change characteristics of the weight sequence and combining the key physiological events of the giant panda, the segmentation boundaries of the cub, juvenile, sub-adult, and adult growth stages are dynamically identified, and the stage segmentation thresholds are obtained. The second segmentation unit is used to perform stage-specific signal decomposition based on the stage segmentation threshold. By constructing an adaptive variational mode decomposition model and setting a mode parameter adjustment mechanism based on the growth stage, the weight sequence is separated into age-related growth trend components and stage fluctuation components, and preliminary stage decomposition results are obtained. The third division unit is used to verify the consistency of stages based on the preliminary stage decomposition results. By establishing a modal fusion model based on the similarity of growth curves, the cross-stage components are reorganized and optimized to obtain a stage-standardized growth baseline that conforms to the continuity of growth and development.

[0050] In one specific embodiment of this application, the prediction module 905 includes: The first prediction unit is used to train stage-specific models based on the stage-standardized growth baseline. It trains gradient boosting tree prediction models for different growth stages, namely the cub stage, juvenile stage, sub-adult stage and adult stage, and optimizes hyperparameters based on the data distribution characteristics of each stage to obtain a set of stage-specific prediction models. The second prediction unit is used to perform dynamic weight integration based on the set of stage-specific prediction models. By constructing a stage-adaptive attention network, the model weights are dynamically allocated according to the real-time growth stage identifier and prediction uncertainty to obtain the initial integrated prediction value. The third prediction unit is used to perform cross-stage consistency optimization based on the preliminary integrated prediction values. By introducing time series smoothing constraints and physiological rationality verification mechanisms, the phase transition region of the prediction results is corrected to obtain integrated and enhanced weight prediction values.

[0051] In one specific embodiment of this application, the output module 906 includes: The first output unit is used to perform multi-source uncertainty analysis based on the predicted weight value. By constructing a multi-factor variation model based on the seasonal metabolic fluctuations, individual developmental heterogeneity and measurement error of giant pandas, the main sources of uncertainty at different growth stages and their interaction relationships are quantified to obtain a set of structured uncertainty components. The second output unit is used to construct a dynamic Bayesian network based on the set of structured uncertainty components. It integrates the statistical distribution characteristics of historical observation data with real-time prediction results by establishing a hierarchical probabilistic graphical model, and introduces time series dependencies and specific physiological constraints of giant pandas to obtain the posterior probability weight distribution. The third output unit is used to generate prediction intervals based on the posterior probability weight distribution. It calculates the weight prediction boundary at different confidence levels by using the quantile regression forest algorithm and dynamically calibrates the interval width by incorporating the growth stage adaptive adjustment mechanism, and finally generates a life cycle weight prediction spectrum with confidence intervals.

[0052] Example 3:

[0053] Corresponding to the above method embodiments, this embodiment also provides a phased adaptive integration giant panda life cycle weight prediction device. The phased adaptive integration giant panda life cycle weight prediction device described below and the phased adaptive integration giant panda life cycle weight prediction method described above can be referred to in correspondence.

[0054] Figure 3 This is a block diagram illustrating a phased adaptive integrated giant panda lifecycle weight prediction device 800 according to an exemplary embodiment. Figure 3 As shown, the phased adaptive integrated giant panda lifecycle weight prediction device 800 may include: a processor 801 and a memory 802. The phased adaptive integrated giant panda lifecycle weight prediction device 800 may also include one or more of the following: a multimedia component 803, an I / O interface 804, and a communication component 805.

[0055] The processor 801 controls the overall operation of the phased adaptive integrated giant panda lifecycle weight prediction device 800 to complete all or part of the steps in the aforementioned phased adaptive integrated giant panda lifecycle weight prediction method. The memory 802 stores various types of data to support the operation of the phased adaptive integrated giant panda lifecycle weight prediction device 800. This data may include, for example, instructions for any application or method operating on the phased adaptive integrated giant panda lifecycle weight prediction device 800, as well as application-related data such as contact data, sent and received messages, images, audio, and video. The memory 802 can be implemented using any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The multimedia component 803 may include a screen and an audio component. The screen may be, for example, a touchscreen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signals may be further stored in the memory 802 or transmitted via the communication component 805. The audio component also includes at least one speaker for outputting audio signals. I / O interface 804 provides an interface between processor 801 and other interface modules, such as a keyboard, mouse, and buttons. These buttons can be virtual or physical. Communication component 805 is used for wired or wireless communication between the phased adaptive integrated giant panda lifecycle weight prediction device 800 and other devices. Wireless communication includes Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, or 4G, or a combination thereof. Therefore, the corresponding communication component 805 may include a Wi-Fi module, a Bluetooth module, and an NFC module.

[0056] In an exemplary embodiment, a phased adaptive integrated giant panda life cycle weight prediction device 800 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the aforementioned phased adaptive integrated giant panda life cycle weight prediction method.

[0057] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, which, when executed by a processor, implement the steps of the aforementioned staged adaptive integrated giant panda lifecycle weight prediction method. For example, the computer-readable storage medium may be the aforementioned memory 802 including program instructions, which may be executed by a processor 801 of a staged adaptive integrated giant panda lifecycle weight prediction device 800 to complete the aforementioned staged adaptive integrated giant panda lifecycle weight prediction method.

[0058] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention 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 the present invention should be included within the scope of protection of the present invention.

Claims

1. A phased adaptive ensemble method for predicting the life-cycle weight of giant pandas, characterized in that, include: Obtain raw data, including continuous weight monitoring sequences, feeding behavior logs, seasonal climate indicators, and individual characteristics of giant pandas; Based on the raw data, a dynamic model of the digestion cycle is performed. By constructing a time-varying differential equation with gastric emptying dynamics, the mass decay process of food from intake to complete digestion is simulated, and the real-time gastric contents mass estimate is obtained. Fasting weight is inverted based on real-time gastric contents mass estimates. The basal fasting weight sequence is obtained by subtracting the digestive system load from the measured weight values ​​point by point. Based on the aforementioned basic fasting weight sequence, growth stages are adaptively divided. The weight sequence is separated into age-related growth trend components and stage fluctuation components using a variational mode decomposition algorithm to obtain a stage-standardized growth baseline. Based on the standardized growth baseline of the aforementioned stage, multi-model ensemble prediction is performed. By constructing a stage-adaptive weighted ensemble network, the prediction model weights for different growth stages are optimized to obtain ensemble-enhanced weight prediction values. Based on the predicted weight, the uncertainty of the life cycle is quantified, and a Bayesian probabilistic graphical model is used to integrate the statistical characteristics of historical observation data with multi-stage prediction results to generate a life cycle weight prediction spectrum with confidence intervals.

2. The staged adaptive integrated method for predicting giant panda lifecycle weight according to claim 1, characterized in that, Dynamic modeling of the digestion cycle is performed based on the raw data, including: Based on the raw data, eating event features are extracted. By constructing a multimodal data alignment model based on timestamp association, discrete eating behavior logs are spatiotemporally matched with continuous weight monitoring sequences to obtain eating event vectors with temporal tags. Based on the feeding event vector, gastric emptying dynamics modeling is performed. By establishing a variable parameter differential equation that considers the coupling of food type and seasonal metabolic rate, the gastric contents mass decay trajectory at different digestion stages is simulated to obtain the theoretical gastric contents mass curve. Based on the theoretical gastric contents mass curve, real-time corrections are made, and the emptying rate parameter is dynamically adjusted by introducing a feedback regulation mechanism based on the slope of body weight change, so as to obtain a real-time gastric contents mass estimate that matches the actual metabolic characteristics.

3. The staged adaptive integrated method for predicting giant panda lifecycle weight according to claim 1, characterized in that, Fasting body weight is inverted based on real-time gastric contents mass estimates, including: The digestive load is quantified based on the real-time gastric contents mass estimate. By establishing a load distribution model based on the time decay function, the gastric contents mass estimate is converted into the instantaneous digestive system load corresponding to the measured body weight time point, thus obtaining the load sequence. The load is dynamically adjusted based on the load sequence. By introducing seasonal metabolic coefficients and individual digestive efficiency factors, the standard load is adaptively adjusted based on physiological characteristics to obtain the individualized real-time load. Based on the real-time load, a fasting weight sequence is generated. By constructing a weight-optimized load stripping model, the dynamically changing digestive system load is deducted point by point from the measured weight value to obtain a continuous basal fasting weight sequence.

4. The staged adaptive integrated method for predicting giant panda lifecycle weight according to claim 1, characterized in that, Adaptive segmentation of growth stages based on the aforementioned baseline fasting body weight sequence includes: Based on the aforementioned fasting weight sequence, growth stage boundary detection is performed. By analyzing the curvature change characteristics of the weight sequence and combining it with the time points of key physiological events in giant pandas, the division boundaries of the cub, juvenile, sub-adult, and adult growth stages are dynamically identified, and the stage division thresholds are obtained. Based on the stage division threshold, stage-specific signal decomposition is performed. By constructing an adaptive variational mode decomposition model and setting a mode parameter adjustment mechanism based on growth stage, the weight sequence is separated into age-related growth trend components and stage fluctuation components, and preliminary stage decomposition results are obtained. Based on the preliminary staged decomposition results, stage consistency verification is performed. By establishing a modal fusion model based on growth curve similarity, cross-stage components are reorganized and optimized to obtain a stage-standardized growth baseline that conforms to the continuity of growth and development.

5. The staged adaptive ensemble method for predicting giant panda lifecycle weight according to claim 1, characterized in that, Multi-model ensemble prediction based on the standardized growth baseline of the aforementioned stage includes: Based on the standardized growth baseline of the stage, stage-specific models are trained. Gradient boosting tree prediction models are trained for different growth stages, namely the cub stage, juvenile stage, sub-adult stage and adult stage. Hyperparameters are optimized based on the data distribution characteristics of each stage to obtain a set of stage-specific prediction models. Dynamic weight integration is performed based on the set of stage-specific prediction models. By constructing a stage-adaptive attention network, model weights are dynamically allocated according to the real-time growth stage identifier and prediction uncertainty to obtain preliminary integrated prediction values. Based on the preliminary integrated prediction values, cross-stage consistency optimization is performed. By introducing time series smoothing constraints and physiological rationality verification mechanisms, the stage transition regions of the prediction results are corrected to obtain integrated and enhanced weight prediction values.

6. A phased adaptive integrated giant panda lifecycle weight prediction system, characterized in that, include: The acquisition module is used to acquire raw data, which includes the giant panda's continuous weight monitoring sequence, feeding behavior log, seasonal climate indicators and individual characteristics; The modeling module is used to perform dynamic modeling of the digestion cycle based on the raw data. By constructing a time-varying differential equation with gastric emptying dynamics, it simulates the mass decay process of food from intake to complete digestion and obtains a real-time estimate of the mass of gastric contents. The inversion module is used to invert fasting weight based on real-time gastric contents mass estimates. It obtains the baseline fasting weight sequence by subtracting the digestive system load point by point from the measured weight values. The segmentation module is used to adaptively segment the growth stages based on the basic fasting weight sequence. The weight sequence is separated into age-related growth trend components and stage fluctuation components through variational mode decomposition algorithm to obtain stage-standardized growth baselines. The prediction module is used to perform multi-model integrated prediction based on the stage-standardized growth baseline. By constructing a stage-adaptive weighted ensemble network, the prediction model weights for different growth stages are optimized to obtain an integrated and enhanced weight prediction value. The output module is used to quantify the life cycle uncertainty based on the predicted weight value, and to integrate the statistical characteristics of historical observation data with multi-stage prediction results using a Bayesian probabilistic graphical model to generate a life cycle weight prediction spectrum with confidence intervals.

7. The staged adaptive integrated giant panda life cycle weight prediction system according to claim 6, characterized in that, The modeling module includes: The first modeling unit is used to extract eating event features based on the raw data. By constructing a multimodal data alignment model based on timestamp association, it performs spatiotemporal matching between discrete eating behavior logs and continuous weight monitoring sequences to obtain eating event vectors with temporal tags. The second modeling unit is used to perform gastric emptying dynamics modeling based on the eating event vector. By establishing a variable parameter differential equation that considers the coupling between food type and seasonal metabolic rate, it simulates the gastric contents mass decay trajectory at different digestion stages and obtains the theoretical gastric contents mass curve. The third modeling unit is used to make real-time corrections based on the theoretical gastric contents mass curve. By introducing a feedback regulation mechanism based on the slope of weight change, the emptying rate parameter is dynamically adjusted to obtain a real-time gastric contents mass estimate that matches the actual metabolic characteristics.

8. The staged adaptive integrated giant panda life cycle weight prediction system according to claim 6, characterized in that, The inversion module includes: The first inversion unit is used to quantify the digestive load based on the real-time gastric contents mass estimate. By establishing a load distribution model based on the time decay function, the gastric contents mass estimate is converted into the instantaneous digestive system load corresponding to the measured body weight time point, and a load sequence is obtained. The second inversion unit is used to dynamically correct the load according to the load sequence. By introducing seasonal metabolic coefficients and individual digestive efficiency factors, the standard load is adaptively adjusted based on physiological characteristics to obtain the individualized corrected real-time load. The third inversion unit is used to generate a fasting weight sequence based on the real-time load. By constructing a weight-optimized load stripping model, the dynamically changing digestive system load is deducted point by point from the measured weight value to obtain a continuous basic fasting weight sequence.

9. The staged adaptive integrated giant panda life cycle weight prediction system according to claim 6, characterized in that, The partitioning module includes: The first segmentation unit is used to detect the growth stage boundaries based on the basic fasting weight sequence. By analyzing the curvature change characteristics of the weight sequence and combining the key physiological events of the giant panda, the segmentation boundaries of the cub stage, juvenile stage, sub-adult stage and adult stage are dynamically identified, and the stage segmentation threshold is obtained. The second segmentation unit is used to perform stage-specific signal decomposition based on the stage segmentation threshold. By constructing an adaptive variational mode decomposition model and setting a mode parameter adjustment mechanism based on the growth stage, the weight sequence is separated into age-related growth trend components and stage fluctuation components to obtain preliminary stage decomposition results. The third division unit is used to perform stage consistency verification based on the preliminary stage decomposition results. By establishing a modal fusion model based on growth curve similarity, the cross-stage components are reorganized and optimized to obtain a stage-standardized growth baseline that conforms to the continuity of growth and development.

10. The staged adaptive integrated giant panda life cycle weight prediction system according to claim 6, characterized in that, The prediction module includes: The first prediction unit is used to train stage-specific models based on the stage-standardized growth baseline. It trains gradient boosting tree prediction models for different growth stages, namely the cub stage, juvenile stage, sub-adult stage and adult stage, and optimizes hyperparameters based on the data distribution characteristics of each stage to obtain a set of stage-specific prediction models. The second prediction unit is used to perform dynamic weight integration based on the set of stage-specific prediction models. By constructing a stage-adaptive attention network, the model weights are dynamically allocated according to the real-time growth stage identifier and prediction uncertainty to obtain the preliminary integrated prediction value. The third prediction unit is used to perform cross-stage consistency optimization based on the preliminary integrated prediction value. By introducing time series smoothing constraints and physiological rationality verification mechanisms, the phase transition region of the prediction results is corrected to obtain the integrated and enhanced weight prediction value.

Images