Citrus tree fruit yield prediction method and system

Abstract

The invention discloses a citrus tree fruit yield prediction method and system, and particularly relates to the technical field of fruit yield prediction. An orchard is divided into a plurality of monitoring areas, local temperature shock speed and citrus tree moisture absorption time delay data are collected in real time, the data are used for dynamically adjusting weight assignment of a fruit yield prediction model, a prediction accuracy index is generated, the index and a preset gradient standard threshold value are compared and analyzed, and the fruit yield prediction accuracy is obtained. The accuracy of classification prediction results is improved; the abnormal degree of incomplete accurate prediction is evaluated, the prediction result is continuously corrected and optimized, and refined monitoring and dynamic adjustment are carried out, so that the problem of traditional meteorological data homogenization is solved, the accuracy of fruit yield prediction in a large-scale orchard is effectively improved, and management decision errors are avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of fruit yield prediction, and in particular to a method and system for predicting citrus fruit yield. Background Art

[0002] With the rapid development of global agriculture and the continuous advancement of modern planting technologies, fruit cultivation has become increasingly important in agricultural production. Citrus, as a highly profitable fruit crop, is widely cultivated worldwide. Citrus fruit yield is directly related to growers' economic returns, making accurate citrus yield prediction crucial for plantation management and market planning. Traditional fruit yield prediction methods rely heavily on manual experience. Growers roughly estimate yield based on years of intuition and observation, combined with factors such as the current year's climatic conditions, tree growth status, and pest and disease patterns. However, these empirically based estimates often lack scientific evidence and are susceptible to subjective judgment, leading to inaccurate predictions. Furthermore, with the continued expansion of citrus cultivation, manual estimation methods have proven inefficient, time-consuming, and labor-intensive in large-scale orchards, failing to meet the demands of efficient management in modern agriculture.

[0003] To improve the accuracy and efficiency of citrus fruit yield forecasting, modern technology-based forecasting methods have gained increasing attention in recent years. These methods typically combine advanced tools such as big data, artificial intelligence, sensor networks, remote sensing, and machine learning. By collecting and analyzing multi-dimensional information such as historical data, meteorological data, soil conditions, and tree growth conditions, they establish fruit yield forecasting models. This automated and intelligent forecasting system not only significantly improves forecast accuracy but also helps growers optimize production management processes, reduce operating costs, and improve fruit quality and yield.

[0004] The existing technology has the following shortcomings:

[0005] In large-scale cultivation, the planting environment and management measures of citrus trees often vary significantly. However, during the data collection process, meteorological data are usually based on data from regional weather stations, which usually cover a large geographical area and ignore the influence of microclimate. For example, the impact of local microclimate phenomena (such as local temperature differences or humidity changes in the orchard) on fruit yield may be very significant, but because homogenized meteorological data fails to capture these local differences, the fruit yield prediction model will misleadingly output inaccurate results. At the same time, because the model cannot accurately capture microclimate differences, citrus fruit yield predictions will face serious deviations, which may lead to management decision errors in large orchards. For example, improper irrigation and fertilization measures may cause excessive water or nutrient supply to fruit trees in local areas, resulting in reduced fruit yield or reduced quality. Summary of the Invention

[0006] The purpose of the present invention is to provide a citrus fruit yield prediction method and system to address the deficiencies in the background technology.

[0007] In order to achieve the above object, the present invention provides the following technical solution: a method for predicting citrus fruit yield, comprising the following steps:

[0008] S1: Divide the orchard into M monitoring areas. When the internal environment of the orchard changes, obtain the local temperature sudden change rate data and the citrus tree water absorption time delay data in real time in each monitoring area within several time periods;

[0009] S2: Determine the weighted values ​​for the accuracy of the output results of the fruit yield prediction model in each time period based on the fluctuation amplitude of the local temperature sudden change rate data and the lag degree of the citrus tree water absorption time delay data, and calculate the prediction accuracy index of the fruit yield prediction model by performing a weighted average calculation on the weighted values ​​for the accuracy of the output results of the fruit yield prediction model in each time period;

[0010] S3: Compare and analyze the prediction accuracy index of the fruit yield prediction model with the gradient standard threshold, and classify the accuracy of the fruit yield prediction results into accurate prediction, incompletely accurate prediction, and inaccurate prediction according to the analysis results;

[0011] S4: For incompletely accurate predictions, the degree of abnormality in the prediction accuracy of the fruit yield prediction model within a fixed time period is evaluated. If the degree of abnormality in accuracy is high, the fruit yield prediction results are continuously revised.

[0012] Preferably, in S2, a local temperature sudden change rate fluctuation index is generated according to the fluctuation amplitude of the local temperature sudden change rate data, and the method for obtaining the local temperature sudden change rate fluctuation index is:

[0013] The temperature sudden change rate data ΔT(t) within the Q time period is obtained from the monitoring area, and the calculation formula is: Where T(t) and T(t+1) are the temperatures at the current moment and the next moment, respectively, and Δt is the time interval. The temperature sudden change rate data is organized into a time series R = {ΔT(1), ΔT(2), …, ΔT(N)}; where N is the length of the time series. The embedding dimension m, that is, the length of the subsequence used to construct the delay vector, is determined. According to the embedding dimension m, the time series R of the temperature sudden change rate data is constructed into an m-dimensional embedding vector R m (i), which is of the form: R m (i) = {ΔT(i), ΔT(i+1), ..., ΔT(i+m-1)}, i = 1, 2, ..., N-m+1; for each embedding vector R m (i), calculate and embed the vector R m(j), the distance is expressed as: d[R m (i), R m (j)] = max(|ΔT(i + k) - ΔT(j + k)|), 0 ≤ k ≤ m - 1; The condition for judging whether two vectors are similar is: d[RX m (i), R m (j)] < r; Where r is the tolerance threshold, used to define the similarity of two trajectories, calculate the proportion of all vector pairs that meet the conditions, denoted as C m (i), and its formula is: Calculate the average value B of similar vector pairs for all i m , and the expression is: Similarly, raise the embedding dimension to m + 1, construct the m + 1 - dimensional embedding vector X m+1 (i), and calculate the proportion C m+1 (i) and B m+1 ; Calculate the local temperature abrupt change speed fluctuation index, and the expression is: Where SQ is the local temperature abrupt change speed fluctuation index.

[0014] Preferably, in S2, generate a water absorption lag index according to the lag degree of the citrus tree water absorption time delay data, and the method for obtaining the water absorption lag index is:

[0015] Construct the soil moisture time series X = {x(1), x(2), …, x(A)}, with a length of A, reflecting the soil moisture content at different time points; construct the citrus tree water absorption time series Y = {y(1), y(2), …, y(B)}, with a length of B, construct an A×B cumulative distance matrix D(i, j) to store the optimal matching distance between each point in the time series X and Y, and each element D(i, j) represents the alignment distance between the soil moisture sequence X(i) and the water absorption sequence Y(j). Initial condition: D(0, 0) = 0, indicating the initial positions of the two sequences at the start of the match;

[0016] Initialize D(i, 0) and D(0, j) to represent the boundary cases; for each pair of points x i and y j , calculate the local Euclidean distance d(x i , y j ), and the expression is: d(x i , y j ) = (x i - y j ) 2; The local distance measures the difference between soil moisture and water absorption at each time point; the dynamic programming algorithm is used to fill the cumulative distance matrix D(i,j), and the cumulative distance is calculated as: The formula shows that the cumulative distance of the current position D(i,j) is the local distance d(x i ,y j ) and the sum of the minimum cumulative distances among its three neighboring points; after the cumulative distance matrix D is filled, backtrack to find the optimal path P from D(A,B) back to D(0,0), which represents the optimal alignment of soil moisture data and water absorption data on the time axis. Starting from D(A,B), choose the direction of the minimum value to backtrack until reaching D(0,0), and record each matching point (i,j) on the path. The path represents the best match between soil moisture X(i) and water absorption Y(j); through the optimal alignment path, determine the lag relationship between soil moisture data and water absorption data. The time difference Δt(i) = ji of each pair of (i,j) matching points represents the lag time between soil moisture change and water absorption response. The water absorption lag index EK is defined as the average value of the time difference of all matching points, and the expression is: Where EK is the moisture absorption hysteresis index, and K is the total number of matching points on the optimal alignment path.

[0017] Preferably, in S2, the local temperature sudden change rate fluctuation index and the water absorption hysteresis index are converted into a first eigenvector, and the first eigenvector is used as the input of the machine learning model. The machine learning model uses the weighted assignment label of the accuracy of the output result of the fruit yield prediction model in each time period predicted by each group of first eigenvectors as the prediction target, and takes minimizing the sum of the prediction errors of the weighted assignment labels of the accuracy of the output results of the fruit yield prediction model in all time periods as the training target. The machine learning model is trained until the sum of the prediction errors reaches convergence and the model training is stopped. The weighted assignment of the accuracy of the output result of the fruit yield prediction model in each time period is determined according to the model output result, wherein the machine learning model is a polynomial regression model, and the prediction accuracy index of the fruit yield prediction model is obtained by weighted average calculation of the weighted assignment of the accuracy of the output result of the fruit yield prediction model in each time period.

[0018] Preferably, in S3, the prediction accuracy index of the fruit yield prediction model is compared and analyzed with the gradient standard threshold, and the accuracy of the fruit yield prediction result is divided into accurate prediction, incompletely accurate prediction and inaccurate prediction according to the analysis result, specifically:

[0019] The obtained prediction accuracy index of the fruit yield prediction model is compared with the gradient standard threshold, the gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold, and the prediction accuracy index of the fruit yield prediction model is compared with the first standard threshold and the second standard threshold respectively;

[0020] If the prediction accuracy index of the fruit yield prediction model is greater than the second standard threshold, it indicates that the prediction accuracy of the fruit yield prediction model is high, and a high-accuracy prediction signal is generated. The accuracy of the fruit yield prediction result is classified as an accurate prediction, and the current management strategy is continued without additional adjustment.

[0021] If the prediction accuracy index of the fruit yield prediction model is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it means that the prediction accuracy of the fruit yield prediction model is medium. At this time, a medium accuracy prediction signal is generated, and the accuracy of the fruit yield prediction result is classified as an incompletely accurate prediction, and the current management strategy is dynamically adjusted;

[0022] If the prediction accuracy index of the fruit yield prediction model is less than the first standard threshold, it means that the prediction accuracy of the fruit yield prediction model is low. At this time, a low-accuracy prediction signal is generated, and the accuracy of the fruit yield prediction result is classified as an inaccurate prediction. Since the prediction result is inaccurate, the grower should immediately take emergency measures.

[0023] Preferably, in S4, for incompletely accurate prediction, within a fixed time period T, the model's predicted output is P pred (T), the actual output is P actual (T), then the prediction error E(T) is defined as: E(T) = |P actual (T)-P pred (T)|; Calculate the accuracy abnormality value MH of the fruit yield prediction model within a fixed time period T, the expression is: If the calculated accuracy anomaly value is greater than or equal to the preset accuracy anomaly value reference threshold, it means that the prediction error is too large and the prediction accuracy anomaly is high;

[0024] When it is detected that the prediction error in the T time period is abnormally high, the fruit yield prediction results need to be continuously revised; the initial prediction result is marked as P pred (T), using the prediction error and environmental variables to correct, the corrected predicted yield P 修正 (T) is expressed as: P 修正 (T)=P pred(T)-α·E(T-1)+β·FG(T); where FG(T) is the prediction accuracy index of the fruit yield prediction model in time period T; α is used to adjust the correction coefficient based on the error of the previous time period, and β is the correction weight of the prediction accuracy index of the fruit yield prediction model, and α>β>0.

[0025] The present invention also provides a citrus fruit yield prediction system, comprising a data acquisition module, a prediction accuracy index calculation module, a result classification module and a prediction result correction module;

[0026] Data acquisition module: The orchard is divided into M monitoring areas. When the internal environment of the orchard changes, the local temperature sudden change rate data and the citrus tree water absorption time delay data within several time periods in each monitoring area are obtained in real time;

[0027] Prediction Accuracy Index Calculation Module: This module determines the weighted values ​​for the accuracy of the fruit yield prediction model output results within each time period based on the fluctuation amplitude of the local temperature sudden change rate data and the lag degree of the citrus tree water absorption time delay data. The prediction accuracy index of the fruit yield prediction model is obtained by performing a weighted average calculation on the weighted values ​​for the accuracy of the fruit yield prediction model output results within each time period.

[0028] Result classification module: compares and analyzes the prediction accuracy index of the fruit yield prediction model with the gradient standard threshold, and classifies the accuracy of the fruit yield prediction results into accurate prediction, incompletely accurate prediction and inaccurate prediction according to the analysis results;

[0029] Forecast result correction module: For incompletely accurate predictions, the abnormal degree of prediction accuracy of the fruit yield prediction model within a fixed time period is evaluated. If the abnormal degree of accuracy is high, the fruit yield prediction results are continuously corrected.

[0030] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0031] 1. The present invention divides the orchard into multiple monitoring areas and collects data on the local temperature change rate and the time delay of water absorption by citrus trees in real time to solve the problem that existing technologies cannot accurately capture microclimate differences. By converting these data into fluctuation indexes and hysteresis indexes, and further combining them with machine learning models to dynamically adjust the weight assignments of the prediction model in each time period, the yield prediction results are ensured to be more accurate, and a prediction accuracy index is generated. This index is classified according to the gradient standard threshold, which helps to distinguish between "accurate predictions", "incompletely accurate predictions" and "inaccurate predictions". In the case of incompletely accurate predictions, the prediction results are continuously corrected through the assessment of the degree of abnormality to achieve efficient management.

[0032] 2. This invention significantly improves the accuracy of citrus fruit yield prediction, particularly in large-scale orchards. By capturing local climate variations, dynamically adjusting model weights, and promptly revising management decisions when predictions are inaccurate, it effectively avoids yield losses caused by improper irrigation and fertilization. Furthermore, through continuous optimization of the machine learning model, prediction errors gradually converge, enabling growers to adjust management measures in real time, thereby improving production efficiency and maximizing yield returns. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0034] Figure 1 Flow chart of the method of the present invention.

[0035] Figure 2 It is a system module diagram of the present invention. DETAILED DESCRIPTION

[0036] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0037] Example 1, please refer to Figure 1 and Figure 2 As shown, the citrus fruit yield prediction method described in this embodiment includes the following steps:

[0038] S1: Divide the orchard into M monitoring areas. When the internal environment of the orchard changes, obtain the local temperature sudden change rate data and the citrus tree water absorption time delay data in real time in each monitoring area within several time periods;

[0039] S2: Determine the weighted values ​​for the accuracy of the output results of the fruit yield prediction model in each time period based on the fluctuation amplitude of the local temperature sudden change rate data and the lag degree of the citrus tree water absorption time delay data, and calculate the prediction accuracy index of the fruit yield prediction model by performing a weighted average calculation on the weighted values ​​for the accuracy of the output results of the fruit yield prediction model in each time period;

[0040] S3: Compare and analyze the prediction accuracy index of the fruit yield prediction model with the gradient standard threshold, and classify the accuracy of the fruit yield prediction results into accurate prediction, incompletely accurate prediction, and inaccurate prediction according to the analysis results;

[0041] S4: For incompletely accurate predictions, the degree of abnormality in the prediction accuracy of the fruit yield prediction model within a fixed time period is evaluated. If the degree of abnormality in accuracy is high, the fruit yield prediction results are continuously revised.

[0042] In S1, the orchard is divided into M monitoring areas. When the internal environment of the orchard changes, the local temperature sudden change rate data and the citrus tree water absorption time delay data within several time periods in each monitoring area are obtained in real time. Specifically:

[0043] First, conduct a detailed analysis of the orchard's topography, including slope, elevation differences, soil type, and drainage. Topographical variations can affect local temperature, humidity, and water distribution. Determine other factors that influence the local microclimate, such as tree density, vegetation cover, the presence of obstructing structures (e.g., buildings, tree belts), and wind direction.

[0044] The orchard is divided into multiple monitoring areas (M) based on the terrain differences, climate change characteristics, and the area of ​​the planting area. Each area is divided based on the following criteria:

[0045] Topographic similarity: Try to ensure that the topographic conditions within a region are basically consistent, such as dividing high slopes, low-lying areas, and flat areas into different areas.

[0046] Microclimate conditions: The same microclimate conditions should be divided into a region. For example, some areas may be more affected by wind, while other areas may be more closed, resulting in different temperature or humidity fluctuations.

[0047] Operation controllability: Zoning should ensure that each area can be managed independently (such as irrigation and fertilization), and the convenience of equipment deployment also needs to be considered.

[0048] Temperature, humidity, and soil moisture sensors are deployed within each monitoring area. The density of sensor distribution depends on the area and terrain complexity. Each area should have at least one sensor network node to collect environmental data, including: Temperature and humidity sensors: used to monitor real-time changes in local temperature and humidity. Soil moisture sensors: used to monitor soil moisture content. Root zone monitoring devices: installed underground to detect changes in water absorption by citrus tree roots.

[0049] The temperature sensor in each monitoring area needs to set a data collection frequency. It is recommended to collect data every 5 minutes. This frequency can capture rapid changes in local temperature in a timely manner. In each monitoring area, the instantaneous rate of temperature change is calculated by continuously collecting temperature data. If the temperature change rate is greater than the set threshold (such as ≥2°C / hour), it is marked as a temperature sudden change. All sensor data is transmitted wirelessly to the central server. Through the automatic data processing program, the temperature sudden change in each area is identified and recorded in the data table for subsequent analysis.

[0050] Soil moisture and tree physiological data collection, including soil moisture sensors, are deployed around the roots of fruit trees to monitor soil moisture changes in real time. Data collection can be set to occur every 10 minutes. Trunk circumference measuring devices installed on tree trunks monitor subtle changes in the trunk in real time, indirectly reflecting water absorption. Trunks expand when they absorb water and contract during droughts. Data collection is performed daily, combined with analysis of diurnal climate changes.

[0051] Citrus trees typically experience a delay in water absorption; increases in soil moisture do not immediately result in water absorption. Water absorption delay data is calculated by comparing changes in soil moisture with changes in tree trunk circumference. This collected water absorption delay data is aggregated to a central server, where the system analyzes the water absorption delay in each area and generates a corresponding delay model.

[0052] In order to effectively capture microclimate differences, the time period division should cover different climatic conditions and fruit tree growth stages. The following is a recommended time period division:

[0053] Daily segmentation: According to the temperature variation pattern during the day, a day can be divided into four main periods (early morning, morning, afternoon, and night), and the temperature, humidity, and moisture data of each period are collected separately.

[0054] Growth cycle segmentation: Based on the growth cycle of citrus trees, the time is divided into flowering period, fruit development period, and fruit ripening period. The microclimate of each stage has different effects on the yield and needs to be collected and analyzed separately.

[0055] S2: According to the fluctuation amplitude of the local temperature sudden change rate data and the lag degree of the citrus tree water absorption time delay data, the weight assignment of the accuracy of the fruit yield prediction model output results in each time period is determined, and the prediction accuracy index of the fruit yield prediction model is obtained by performing weighted average calculation on the weight assignment of the accuracy of the fruit yield prediction model output results in each time period.

[0056] The local temperature sudden change rate fluctuation index is generated according to the fluctuation amplitude of the local temperature sudden change rate data. The method for obtaining the local temperature sudden change rate fluctuation index is:

[0057] Obtain the temperature abrupt change rate data ΔT(t) within the Q time period from the monitoring area, and its calculation formula is: In the formula, T(t) and T(t + 1): represent the temperatures at the current moment and the next moment respectively, and Δt is the time interval (for example, sampling once every 5 minutes). Organize these temperature abrupt change rate data into a time series R = {ΔT(1), ΔT(2), …, ΔT(N)}; where N is the length of the time series; determine the embedding dimension m, that is, the length of the subsequence used to construct the delay vector, usually choose m = 2 or m = 3; according to the embedding dimension m, construct the time series R of the temperature abrupt change rate data into an m-dimensional embedding vector R m (i), and its form is: R m (i) = {ΔT(i), ΔT(i + 1), …, ΔT(i + m - 1)}, i = 1, 2, …, N - m + 1; for each embedding vector R m (i), calculate the distance between it and other embedding vectors R m (j) (where j ≠ i). The definition of the distance can adopt the maximum distance (Chebyshev distance), and the expression is: d[R m (i), R m (j)] = max(|ΔT(i + k) - ΔT(j + k)|), 0 ≤ k ≤ m - 1; the condition for judging whether two vectors are similar is: d[RX m (i), R m (j)] < r; in the formula, r is the tolerance threshold, used to define the similarity of two trajectories, usually taken as 20% of the standard deviation of the time series; calculate the proportion of all vector pairs that meet the conditions, denoted as C m (i), and its formula is: Calculate the average value B of the similar vector pairs for all i m , and the expression is: Similarly, raise the embedding dimension to m + 1, construct an m + 1-dimensional embedding vector X m+1 (i), and calculate the proportion C m+1 (i) and B m+1 ; calculate the local temperature abrupt change rate fluctuation index, and the expression is: In the formula, SQ is the local temperature abrupt change rate fluctuation index.

[0058] A high local temperature fluctuation index indicates that the orchard's local temperature fluctuates dramatically, with frequent and unstable climatic conditions. These drastic temperature changes can significantly impact citrus tree growth and fruit development, particularly affecting water absorption, photosynthetic efficiency, and fruit expansion. If the prediction model fails to adequately capture these dramatic fluctuations and make corresponding adjustments, its prediction accuracy will decline significantly. Fruit yield may be overestimated or underestimated, leading to inaccurate management decisions such as improper irrigation or fertilization, ultimately impacting fruit yield and quality. Therefore, the larger the fluctuation index, the lower the prediction accuracy will tend to be if the model fails to make timely corrections.

[0059] On the contrary, when the local temperature sudden change rate fluctuation index is small, it means that the local climate of the orchard is relatively stable, the temperature fluctuation range is small, and the environmental conditions are relatively stable. In this stable climate, the growth of citrus trees is usually in a relatively stable state, and physiological activities (such as water absorption and photosynthesis) are not strongly affected by sudden climate changes. In this case, the fruit yield prediction model is more likely to capture the main environmental factors and its prediction accuracy will be higher. Therefore, the smaller the fluctuation index, the smaller the environmental fluctuation. Under such conditions, the prediction accuracy of the model is usually higher, and management decisions are relatively reliable.

[0060] The water absorption hysteresis index is generated according to the hysteresis degree of the citrus tree water absorption time delay data. The method for obtaining the water absorption hysteresis index is:

[0061] Construct a soil moisture time series X = {x(1), x(2), ..., x(A)} with a length of A, reflecting the soil moisture content at different time points. Construct a citrus tree water absorption time series Y = {y(1), y(2), ..., y(B)} with a length of B, usually reflected by changes in trunk diameter or other physiological signals. Construct an A×B cumulative distance matrix D(i, j) to store the optimal matching distance between each point in the time series X and Y. Each element D(i, j) represents the alignment distance between the soil moisture series X(i) and the water absorption series Y(j). The initial condition is: D(0, 0) = 0, indicating the initial position of the two sequences when matching begins. Initialize D(i, 0) and D(0, j) (usually set to infinity) to represent the boundary conditions. For each pair of points x in the two time series i and y j , calculate the local Euclidean distance d(x i ,y j ), the expression is: d(x i ,y j )=(x i -y j ) 2; This local distance measures the difference between soil moisture and water absorption at each time point. The cumulative distance matrix D(i,j) is filled using a dynamic programming algorithm. The cumulative distance is calculated as: The formula shows that the cumulative distance of the current position D(i,j) is the local distance d(x i ,y j ) and the sum of the minimum cumulative distances between it and its three neighboring points. After the cumulative distance matrix D is filled, backtrack to find the optimal path P from D(A,B) back to D(0,0). This path represents the optimal alignment of soil moisture data and water absorption data on the time axis. The specific steps are as follows:

[0062] Starting from D(A,B), choose the direction of the minimum value and backtrack (up, left, or left-up) until reaching D(0,0). Record each matching point (i,j) on the path, which represents the best match between soil moisture X(i) and water absorption Y(j). Through the optimal alignment path, the lag relationship between soil moisture data and water absorption data can be determined. The time difference Δt(i) = ji for each pair of (i,j) matching points represents the lag time between the soil moisture change and the water absorption response. The water absorption lag index EK is defined as the average of the time differences of all matching points, and the expression is: Where EK is the moisture absorption hysteresis index, and K is the total number of matching points on the optimal alignment path.

[0063] When the water absorption hysteresis index is large, it means that the citrus tree's water absorption response to changes in soil moisture has a longer lag time. This large lag usually indicates that the citrus tree responds slowly to environmental changes, which may be caused by factors such as tree age, variety, and root health. If the fruit yield prediction model fails to fully capture this delayed effect, it may lead to incorrect management decisions, such as excessively high or low irrigation and fertilization frequencies, which in turn affect the normal growth and development of the fruit. Therefore, the larger the water absorption hysteresis index, if the model does not effectively account for the lag effect, the prediction accuracy may be significantly reduced, especially under dynamic environmental changes.

[0064] Conversely, a smaller water uptake hysteresis index indicates that citrus trees respond more quickly to changes in soil moisture, absorbing water promptly and using it for fruit growth and development. In this scenario, environmental conditions have a more direct impact on citrus trees, and the correlation between fruit yield and management measures is stronger. If a forecasting model can capture this rapid response, its yield forecast will be more accurate. Therefore, a smaller water uptake hysteresis index indicates more timely interaction between trees and the environment, and the model's forecast accuracy is more likely to improve, as water management decisions can quickly impact fruit yield.

[0065] The local temperature sudden change rate fluctuation index and the water absorption hysteresis index are converted into the first eigenvector, and the first eigenvector is used as the input of the machine learning model. The machine learning model uses the weighted assignment label of the accuracy of the fruit yield prediction model output results in each time period predicted by each group of first eigenvectors as the prediction target, and takes minimizing the sum of the prediction errors of the weighted assignment labels of the accuracy of the fruit yield prediction model output results in all time periods as the training target. The machine learning model is trained until the sum of the prediction errors reaches convergence and the model training is stopped. The weighted assignment of the accuracy of the fruit yield prediction model output results in each time period is determined according to the model output results, wherein the machine learning model is a polynomial regression model, and the prediction accuracy index of the fruit yield prediction model is obtained by weighted average calculation of the weighted assignment of the accuracy of the fruit yield prediction model output results in each time period.

[0066] The method for obtaining the weighted value of the accuracy of the output results of the fruit yield prediction model in each time period is as follows: from the first eigenvector training data of the trained machine learning model, the corresponding function expression is obtained: CN = F(SQ, EK); where F is the output function of the model, SQ is the local temperature sudden change rate fluctuation index, EK is the water absorption hysteresis index, and CN is the weighted value of the accuracy of the output results of the fruit yield prediction model in each time period.

[0067] S3: The prediction accuracy index of the fruit yield prediction model is compared and analyzed with the gradient standard threshold, and the accuracy of the fruit yield prediction results is divided into accurate prediction, incompletely accurate prediction and inaccurate prediction according to the analysis results.

[0068] The obtained prediction accuracy index of the fruit yield prediction model is compared with the gradient standard threshold, the gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold, and the prediction accuracy index of the fruit yield prediction model is compared with the first standard threshold and the second standard threshold respectively;

[0069] If the prediction accuracy index of the fruit yield prediction model is greater than the second standard threshold, it indicates that the prediction accuracy of the fruit yield prediction model is high, and a high-accuracy prediction signal is generated. The accuracy of the fruit yield prediction result is classified as an accurate prediction, and the current management strategy is continued without additional adjustment.

[0070] If the prediction accuracy index of the fruit yield prediction model is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it means that the prediction accuracy of the fruit yield prediction model is medium. At this time, a medium accuracy prediction signal is generated, and the accuracy of the fruit yield prediction result is classified as an incompletely accurate prediction, and the current management strategy is dynamically adjusted;

[0071] If the prediction accuracy index of the fruit yield prediction model is less than the first standard threshold, it means that the prediction accuracy of the fruit yield prediction model is low. At this time, a low-accuracy prediction signal is generated, and the accuracy of the fruit yield prediction result is classified as an inaccurate prediction. Due to the inaccurate prediction result, the grower should immediately take emergency plans, such as adjusting the irrigation frequency, strengthening pest and disease control, etc.

[0072] S4: For incompletely accurate predictions, the degree of abnormality in the prediction accuracy of the fruit yield prediction model within a fixed time period is evaluated. If the degree of abnormality in accuracy is high, the fruit yield prediction results are continuously revised.

[0073] For incomplete predictions, within a fixed time period T, the model's predicted output is P pred (T), the actual output is P actual (T), then the prediction error E(T) is defined as: E(T) = |P actual (T)-P pred (T)|; Calculate the accuracy abnormality value MH of the fruit yield prediction model within a fixed time period T, the expression is: If the calculated accuracy abnormality value is greater than or equal to the preset accuracy abnormality value reference threshold, it means that the prediction error is too large and the prediction accuracy abnormality is high; when the prediction error abnormality is high within the T time period, the fruit yield prediction result needs to be continuously corrected. The initial prediction result is marked as P pred (T), continuous correction can be made based on historical errors and the prediction accuracy index of the fruit yield prediction model. Using the prediction error and environmental variables to make corrections, the corrected predicted yield P 修正 (T) is expressed as: P 修正 (T)=P pred (T) - α·E(T-1) + β·FG(T); where FG(T) is the prediction accuracy index of the fruit yield prediction model in time period T; α is used to adjust the correction factor based on the error in the previous time period. β is the correction weight for the prediction accuracy index of the fruit yield prediction model, with α>β>0. This correction process is continuous. In each time period, the new error E(T) and environmental variables are fed back into the prediction model, continuously optimizing the prediction results until the error converges to a smaller range, ensuring that the prediction is more accurate and reducing error accumulation.

[0074] In this embodiment, the orchard is first divided into M monitoring areas, and real-time data on the local temperature change rate and the water absorption time delay of citrus trees are obtained in each area. Next, based on the fluctuation amplitude and hysteresis of these data, the weights of the fruit yield prediction model output results in each time period are determined, and the model's prediction accuracy index is calculated through weighted average. Subsequently, the prediction accuracy index is compared with a preset gradient standard threshold to categorize it into three types: "accurate prediction," "incompletely accurate prediction," and "inaccurate prediction." For incompletely accurate predictions, the degree of abnormality is assessed. If the degree of abnormality is high, the prediction results are continuously revised to improve the accuracy of the prediction.

[0075] Example 2, a citrus fruit yield prediction system described in this example includes a data acquisition module, a prediction accuracy index calculation module, a result classification module, and a prediction result correction module;

[0076] Data acquisition module: The orchard is divided into M monitoring areas. When the internal environment of the orchard changes, the local temperature sudden change rate data and the citrus tree water absorption time delay data within several time periods in each monitoring area are obtained in real time;

[0077] Prediction Accuracy Index Calculation Module: This module determines the weighted values ​​for the accuracy of the fruit yield prediction model output results within each time period based on the fluctuation amplitude of the local temperature sudden change rate data and the lag degree of the citrus tree water absorption time delay data. The prediction accuracy index of the fruit yield prediction model is obtained by performing a weighted average calculation on the weighted values ​​for the accuracy of the fruit yield prediction model output results within each time period.

[0078] Result classification module: compares and analyzes the prediction accuracy index of the fruit yield prediction model with the gradient standard threshold, and classifies the accuracy of the fruit yield prediction results into accurate prediction, incompletely accurate prediction and inaccurate prediction according to the analysis results;

[0079] Forecast result correction module: For incompletely accurate predictions, the abnormal degree of prediction accuracy of the fruit yield prediction model within a fixed time period is evaluated. If the abnormal degree of accuracy is high, the fruit yield prediction results are continuously corrected.

[0080] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0081] It should be understood that the term "and / or" as used herein simply describes a relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A alone, A and B together, or B alone. A and B can be singular or plural. Furthermore, the character " / " as used herein generally indicates an "or" relationship between the associated objects, but it may also indicate an "and / or" relationship. For specific understanding, please refer to the context.

[0082] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0083] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. A method for predicting citrus fruit yield, characterized by: The following steps are included: S1: Divide the orchard into M monitoring areas. When the internal environment of the orchard changes, obtain the local temperature sudden change rate data and the citrus tree water absorption time delay data in real time in each monitoring area within several time periods; S2: Determine the weighted values ​​for the accuracy of the output results of the fruit yield prediction model in each time period based on the fluctuation amplitude of the local temperature sudden change rate data and the lag degree of the citrus tree water absorption time delay data, and calculate the prediction accuracy index of the fruit yield prediction model by performing a weighted average calculation on the weighted values ​​for the accuracy of the output results of the fruit yield prediction model in each time period; S3: Compare and analyze the prediction accuracy index of the fruit yield prediction model with the gradient standard threshold, and classify the accuracy of the fruit yield prediction results into accurate prediction, incompletely accurate prediction, and inaccurate prediction according to the analysis results; S4: For incompletely accurate predictions, the degree of abnormality in the prediction accuracy of the fruit yield prediction model within a fixed time period is evaluated. If the degree of abnormality in accuracy is high, the fruit yield prediction results are continuously revised.

2. The method for predicting citrus fruit yield according to claim 1, wherein: In S2, a local temperature sudden change rate fluctuation index is generated according to the fluctuation amplitude of the local temperature sudden change rate data. The local temperature sudden change rate fluctuation index is obtained as follows: Obtain the temperature abrupt change rate data ΔT(t) within the Q time period from the monitoring area, and its calculation formula is: In the formula, T(t) and T(t + 1) are the temperatures at the current moment and the next moment respectively, Δt is the time interval, and the temperature abrupt change rate data is sorted into a time series R = {ΔT(1), ΔT(2), …, ΔT(N)}; where N is the length of the time series; determine the embedding dimension m, that is, the length of the subsequence used to construct the delay vector. According to the embedding dimension m, the time series R of the temperature abrupt change rate data is constructed into an m-dimensional embedding vector R m , m , m , m , m , , m+1 , , m+1 , m , m+1 (i), and its form is: R m (i) = {ΔT(i), ΔT(i + 1), …, ΔT(i + m - 1)}, i = 1, 2, …, N - m + 1; for each embedding vector R m (i), calculate the distance between it and the embedding vector R m (j), and the expression is: d[R m (i), R m (j)] = max(|ΔT(i + k) - ΔT(j + k)|), 0 ≤ k ≤ m - 1; the condition for judging whether two vectors are similar is: d[RX m (i), R m (j)] < r; in the formula, r is the tolerance threshold used to define the similarity of two trajectories, and calculate the ratio of all vector pairs that meet the conditions, denoted as C m (i), and its formula is: Calculate the average value B of the similar vector pairs for all i m , and the expression is: Similarly, raise the embedding dimension to m + 1, construct an m + 1-dimensional embedding vector X m+1 (i), and calculate the ratio C m+1 (i) and B m+1 ; Calculate the local temperature sudden change rate fluctuation index, the expression is: Where SQ is the local temperature sudden change rate fluctuation index.

3. A citrus fruit yield prediction method according to claim 2, characterized in that: In S2, a water absorption hysteresis index is generated according to the hysteresis degree of the water absorption time delay data of the citrus tree. The method for obtaining the water absorption hysteresis index is: Construct a soil moisture time series X = {x(1), x(2), ..., x(A)} with a length of A, reflecting the soil moisture content at different time points; construct a citrus tree water absorption time series Y = {y(1), y(2), ..., y(B)} with a length of B, and construct an A×B cumulative distance matrix D(i, j) to store the optimal matching distance between each point in the time series X and Y. Each element D(i, j) represents the alignment distance between the soil moisture series X(i) and the water absorption series Y(j). The initial condition is: D(0, 0) = 0, indicating the initial position of the two sequences when matching begins. Initialize D(i,0) and D(0,j) to represent the boundary conditions; for each pair of points x in the two time series i and y j , calculate the local Euclidean distance d(x i ,y j ), the expression is: d(x i ,y j )=(x i -y j ) 2 ; The local distance measures the difference between soil moisture and water absorption at each time point; the dynamic programming algorithm is used to fill the cumulative distance matrix D(i,j), and the cumulative distance is calculated as: The formula shows that the cumulative distance of the current position D(i,j) is the local distance d(x i ,y j ) and the sum of the minimum cumulative distances among its three neighboring points; after the cumulative distance matrix D is filled, backtrack to find the optimal path P from D(A,B) back to D(0,0), which represents the optimal alignment of soil moisture data and water absorption data on the time axis. Starting from D(A,B), choose the direction of the minimum value to backtrack until reaching D(0,0), and record each matching point (i,j) on the path. The path represents the best match between soil moisture X(i) and water absorption Y(j); through the optimal alignment path, determine the lag relationship between soil moisture data and water absorption data. The time difference Δt(i) = ji of each pair of (i,j) matching points represents the lag time between soil moisture change and water absorption response. The water absorption lag index EK is defined as the average value of the time difference of all matching points, and the expression is: Where EK is the moisture absorption hysteresis index, and K is the total number of matching points on the optimal alignment path.

4. A citrus fruit yield prediction method according to claim 3, characterized in that: In S2, the local temperature sudden change rate fluctuation index and the water absorption hysteresis index are converted into the first eigenvector, and the first eigenvector is used as the input of the machine learning model. The machine learning model uses the weighted assignment label of the accuracy of the fruit yield prediction model output results in each time period predicted by each group of first eigenvectors as the prediction target, and takes minimizing the sum of the prediction errors of the weighted assignment labels of the accuracy of the fruit yield prediction model output results in all time periods as the training target. The machine learning model is trained until the sum of the prediction errors reaches convergence, and the model training is stopped. The weighted assignment of the accuracy of the fruit yield prediction model output results in each time period is determined according to the model output results, wherein the machine learning model is a polynomial regression model, and the prediction accuracy index of the fruit yield prediction model is obtained by weighted average calculation of the weighted assignment of the accuracy of the fruit yield prediction model output results in each time period.

5. A citrus fruit yield prediction method according to claim 4, characterized in that: In S3, the prediction accuracy index of the fruit yield prediction model is compared with the gradient standard threshold. According to the analysis results, the accuracy of the fruit yield prediction results is divided into accurate prediction, incompletely accurate prediction and inaccurate prediction, specifically: The obtained prediction accuracy index of the fruit yield prediction model is compared with the gradient standard threshold, the gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold, and the prediction accuracy index of the fruit yield prediction model is compared with the first standard threshold and the second standard threshold respectively; If the prediction accuracy index of the fruit yield prediction model is greater than the second standard threshold, it indicates that the prediction accuracy of the fruit yield prediction model is high, and a high-accuracy prediction signal is generated. The accuracy of the fruit yield prediction result is classified as an accurate prediction, and the current management strategy is continued without additional adjustment. If the prediction accuracy index of the fruit yield prediction model is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it means that the prediction accuracy of the fruit yield prediction model is medium. At this time, a medium accuracy prediction signal is generated, and the accuracy of the fruit yield prediction result is classified as an incompletely accurate prediction, and the current management strategy is dynamically adjusted; If the prediction accuracy index of the fruit yield prediction model is less than the first standard threshold, it means that the prediction accuracy of the fruit yield prediction model is low. At this time, a low-accuracy prediction signal is generated, and the accuracy of the fruit yield prediction result is classified as an inaccurate prediction. Since the prediction result is inaccurate, the grower should immediately take emergency measures.

6. The method for predicting citrus fruit yield according to claim 1, wherein: In S4, for incomplete prediction, within a fixed time period T, the model's predicted output is P pred (T), the actual output is P actual (T), then the prediction error E(T) is defined as: E(T) = |P actual (T)-P pred (T)|; Calculate the accuracy abnormality value MH of the fruit yield prediction model within a fixed time period T, the expression is: If the calculated accuracy anomaly value is greater than or equal to the preset accuracy anomaly value reference threshold, it means that the prediction error is too large and the prediction accuracy anomaly is high; When it is detected that the prediction error in the T time period is abnormally high, the fruit yield prediction results need to be continuously revised; the initial prediction result is marked as P pred (T), using the prediction error and environmental variables to correct, the corrected predicted yield P 修正 (T) is expressed as: P 修正 (T)=P pred (T)-α·E(T-1)+β·FG(T); where FG(T) is the prediction accuracy index of the fruit yield prediction model in time period T; α is used to adjust the correction coefficient based on the error of the previous time period, and β is the correction weight of the prediction accuracy index of the fruit yield prediction model, and α>β>0.

7. A citrus fruit yield prediction system for implementing the citrus fruit yield prediction method according to any one of claims 1 to 6, characterized in that: It includes data acquisition module, prediction accuracy index calculation module, result classification module and prediction result correction module; Data acquisition module: The orchard is divided into M monitoring areas. When the internal environment of the orchard changes, the local temperature sudden change rate data and the citrus tree water absorption time delay data within several time periods in each monitoring area are obtained in real time; Prediction Accuracy Index Calculation Module: This module determines the weighted values ​​for the accuracy of the fruit yield prediction model output results within each time period based on the fluctuation amplitude of the local temperature sudden change rate data and the lag degree of the citrus tree water absorption time delay data. The prediction accuracy index of the fruit yield prediction model is obtained by performing a weighted average calculation on the weighted values ​​for the accuracy of the fruit yield prediction model output results within each time period. Result classification module: compares and analyzes the prediction accuracy index of the fruit yield prediction model with the gradient standard threshold, and classifies the accuracy of the fruit yield prediction results into accurate prediction, incompletely accurate prediction and inaccurate prediction according to the analysis results; Forecast result correction module: For incompletely accurate predictions, the abnormal degree of prediction accuracy of the fruit yield prediction model within a fixed time period is evaluated. If the abnormal degree of accuracy is high, the fruit yield prediction results are continuously corrected.

Images