A multi-model fusion method based on wheat yield estimation
Optimizing environmental characteristic data through time-varying importance factors and environmental sensitivity factors, combining biomass data to predict wheat yields, solving the problems of insufficient adaptability and static characteristic evaluation in the existing technology, and improving the accuracy of wheat yield prediction and Adaptability.
Patent Information
- Application Number
- CN202411149363.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-21
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2044-08-21
AI Technical Summary
The existing multi-model fusion method for wheat yield estimates is insufficient in the face of dynamically changing environmental factors, making it difficult to quickly adjust the weight, resulting in a decrease in prediction accuracy. The traditional static feature importance assessment cannot reflect the time changes of environmental factors. .
Time-varying importance factors and environmental sensitivity factors are used to optimize environmental characteristic data, optimize environmental characteristic sequences are obtained, and classified in combination with biomass data to establish a higher adaptable yield prediction model. Time-varying importance factor quantifies the time variation of characteristic importance by comparing the current correlation with historical data, and environmental sensitivity factor quantifies the immediate impact of environmental factors through local linear regression.
It improves the accuracy of plot classification and wheat yield prediction, especially in extreme weather or sudden pests and diseases, which can quickly respond to environmental changes, enhancing the adaptability and prediction accuracy of the model.
Smart Images

Figure CN118656657B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of data processing, and in particular to a multi-model fusion method based on wheat yield estimation. Background Art
[0002] As an important food crop in the world, wheat yield estimation is of great significance for agricultural planning, market regulation and food safety. With the development of information technology, multi-model fusion methods have shown great potential in improving the accuracy of wheat yield estimation. These methods usually combine multiple data sources and prediction models to capture the complex factors affecting wheat yield from different perspectives.
[0003] Among the existing multi-model fusion methods for wheat yield estimation, k-means clustering is a common technique used to group samples according to farmland characteristics in order to establish unique model weights for each cluster. This method can improve the accuracy of prediction to a certain extent by using the distance between the sample and the cluster center as the basis for weight adjustment. However, this method is not adaptable enough when faced with dynamically changing environmental factors, such as extreme weather events or sudden pests and diseases, and it is difficult to quickly adjust the weights to reflect environmental changes, resulting in a decrease in prediction accuracy. In addition, the traditional k-means clustering method usually uses a static feature importance evaluation, which cannot reflect the temporal changes in the importance of environmental factors, further limiting the adaptability of the model. Summary of the invention
[0004] In order to solve the above technical problems, the purpose of the present invention is to provide a multi-model fusion method based on wheat yield estimation, and the technical scheme adopted is as follows:
[0005] An embodiment of the present invention provides a multi-model fusion method based on wheat yield estimation, which specifically comprises:
[0006] Obtain the environmental characteristic data and biomass data of wheat in the current growth period and historical growth period of each plot of land, and form the current environmental characteristic sequence, current biomass sequence, historical environmental characteristic sequence, and historical biomass sequence respectively;
[0007] According to the Pearson correlation coefficient between each current environmental characteristic sequence and the current biomass sequence, and the Pearson correlation coefficient between each historical environmental characteristic sequence and the historical biomass sequence, the time-varying importance factor corresponding to each current environmental characteristic sequence is obtained; the current environmental characteristic sequence is optimized using the time-varying importance factor to obtain an optimized environmental characteristic sequence;
[0008] Evenly segment the optimized environmental feature sequence; obtain the slope of each segment of the segmented optimized environmental feature sequence; obtain the environmental sensitivity factor corresponding to each segment of the optimized environmental feature sequence according to the slope; optimize the optimized environmental feature sequence using the environmental sensitivity factor to obtain a first optimized environmental feature sequence;
[0009] The various first optimized environmental feature sequences and current biomass sequences of each plot of land are combined into a classification basis set; each plot of land is classified using the classification basis set; the yield prediction model of each type of land is obtained to predict wheat yield.
[0010] Preferably, the environmental characteristic data and biomass data include:
[0011] The environmental characteristic data are three types of data, namely temperature, precipitation, and soil moisture; the biomass data are the weight of all wheat plants in a field at a moment.
[0012] Preferably, the time-varying importance factor is:
[0013] ,
[0014] in, represents the time-varying importance factor corresponding to the i-th current environmental feature sequence; represents the Pearson correlation coefficient between the i-th current environmental characteristic sequence and the current biomass sequence; represents the Pearson correlation coefficient between the i-th historical environmental characteristic sequence and the historical biomass sequence t years away from the current growing period; T represents the maximum number of years from the historical growing period to the current growing period; represents the time decay factor; Indicates the number of types of environmental characteristic data.
[0015] Preferably, the current environmental feature sequence is optimized using the time-varying importance factor to obtain an optimized environmental feature sequence, including:
[0016] Each current environmental feature sequence is multiplied by the corresponding time-varying importance factor to obtain the optimized environmental feature sequence.
[0017] Preferably, the optimized environment feature sequence is evenly segmented, including:
[0018] A time window is set, and the time window is used to slide in the optimized environment feature sequence to evenly segment the optimized environment feature sequence, wherein the sliding step length of the time window is the length of the time window.
[0019] Preferably, obtaining the slope of each segment of data in the segmented optimized environment feature sequence includes:
[0020] The current biomass sequence is segmented using the time window; local linear regression analysis is performed using each segment of the segmented optimized environmental feature sequence and each segment of the corresponding current biomass sequence to obtain the local expression:
[0021] ,
[0022] in, Represents the data in the pth time window in the current biomass sequence; Represents the data in the pth time window in the optimized environment feature sequence corresponding to the i-th current environment feature sequence; and are the first and second regression coefficients, respectively; is the error term;
[0023] The second regression coefficient is the slope of the data in the pth time window in the optimized environmental feature sequence, that is, the slope of the pth segment of data.
[0024] Preferably, obtaining the environmental sensitivity factor corresponding to each segment of data in the optimized environmental feature sequence according to the slope includes:
[0025] Calculate the ratio of the standard deviation and the average value of each data segment in the optimized environment feature sequence as the corresponding change rate of each data segment;
[0026] According to the absolute value of the change rate and slope of each data segment in the optimized environmental characteristic sequence, the environmental sensitivity factor corresponding to each data segment is obtained. The calculation formula is: ,
[0027] in, represents the environmental sensitivity factor corresponding to the p-th segment of data in the optimized environmental feature sequence corresponding to the i-th current environmental feature sequence; Represents the normalized value of the absolute value of the slope of the p-th segment of data in the optimized environmental feature sequence corresponding to the i-th current environmental feature sequence; Represents the normalized value of the change rate of the p-th segment of data in the optimized environment feature sequence corresponding to the i-th current environment feature sequence.
[0028] Preferably, the calculation formula of the first optimized environmental feature sequence is:
[0029] ,
[0030] in, represents the first optimized environment feature sequence corresponding to the i-th current environment feature sequence; represents the weight coefficient; represents the optimized environment feature sequence corresponding to the i-th current environment feature sequence; Represents the environmental sensitivity factor corresponding to the p-th segment of data in the optimized environmental feature sequence corresponding to the i-th current environmental feature sequence.
[0031] Preferably, each plot of land is classified using a classification basis set, including:
[0032] The distance between the classification basis sets of every two plots of land is calculated, and each plot of land is classified according to the distance.
[0033] The embodiments of the present invention have at least the following beneficial effects: the present invention collects environmental characteristic data and biomass data of the current growth period and the historical growth period as the original data for subsequent classification of plots, not only taking into account the data of the current period and the historical period, but also considering multi-source data, so that the subsequent classification accuracy is higher; further, the time-varying importance factor corresponding to each current environmental characteristic sequence is obtained, and the current environmental characteristic sequence is optimized using the time-varying importance factor to obtain an optimized environmental characteristic sequence, which solves the problem that the traditional static characteristic importance evaluation method cannot fully capture the dynamic effect of environmental characteristics on wheat biomass, and further improves the accuracy of plot classification; then, on the basis of the optimized environmental characteristic sequence, the environmental sensitivity factor is obtained, and the optimized environmental characteristic sequence is optimized using the environmental sensitivity factor to obtain the first optimized environmental characteristic sequence, quantifying the immediate impact of different environmental factors on wheat biomass, and further improving the accuracy of subsequent classification of plots; after the classification of the plots is completed, the yield prediction model of each type of land is obtained, and the wheat yield is predicted. Different types of plots are adapted to different yield prediction models, which improves the adaptability of the plots and the yield prediction model, thereby improving the accuracy of the prediction of wheat yield. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0035] Figure 1 A method flow chart of a multi-model fusion method based on wheat yield estimation provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0036] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the multi-model fusion method based on wheat yield estimation proposed by the present invention, its specific implementation method, structure, characteristics and effects, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures, or characteristics in one or more embodiments may be combined in any suitable form.
[0037] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
[0038] The following is a detailed description of a multi-model fusion method based on wheat yield estimation provided by the present invention in conjunction with the accompanying drawings.
[0039] Example:
[0040] The main application scenarios of the present invention are:
[0041] The main purpose of the present invention is to design a multi-model fusion method for wheat yield estimation that can quickly respond to environmental changes, especially in the face of extreme weather events or sudden pests and diseases, and can flexibly adjust the environmental feature weights to improve the accuracy and adaptability of the prediction, that is, to collect environmental feature data and biomass data when wheat grows in different plots to optimize the environmental feature weights. Use the optimized data set to perform cluster analysis and assign the best matching prediction model to the cluster cluster.
[0042] See also Figure 1 , which shows a flow chart of a multi-model fusion method based on wheat yield estimation provided by an embodiment of the present invention, the method comprising the following steps:
[0043] Step S1, obtaining environmental characteristic data and biomass data of each plot of land in the current growing period and the historical growing period; forming a current environmental characteristic sequence, a current biomass sequence, a historical environmental characteristic sequence, and a historical biomass sequence respectively.
[0044] When predicting wheat yield for each plot of land, the situation of each plot is different. The same prediction model may not be universal when making predictions. Therefore, it is necessary to classify the plots and then select the appropriate model for each type of plot for prediction. This can greatly improve the accuracy of the prediction.
[0045] The plots in the present invention default to one season per year. For each plot of land, the environmental characteristic data of wheat growth every day in the current growing period and the historical growing period of each plot of land are collected, including temperature, precipitation, and soil moisture. At the same time, biomass data also needs to be collected. The biomass data represents the weight of all wheat plants in a plot of land at a moment. It should be noted that the current growing period is also the growing period of wheat in the current plot of land, which represents the most recent growing period of wheat in a plot of land. The growing period of wheat includes the entire period from planting wheat to harvesting wheat; the historical growing period is also each growing period of wheat in history; for example, the current growing period is 2023, and the wheat growing period of each year before 2023 is a historical growing period.
[0046] Obtain the environmental characteristic data and biomass data of each wheat field during the current and historical growth periods, and form the current environmental characteristic sequence respectively , Current biomass sequence , historical environmental characteristics sequence , historical biomass series , each type of environmental feature data constitutes an environmental feature sequence. , It is expressed as: , which represents a large set of three current environment feature sequences. , The timestamp of sampling data for each plot is set at daily intervals. and Same as above, where i represents two meanings, one is the environmental characteristic data, and the other represents the number of types of environmental characteristic sequences. There are three types in total, namely temperature, precipitation, and soil moisture. The values of i are 1, 2, and 3, for example express , express ; j only represents biomass data, t represents the number of years from the current growing season, and t is 1, which means that the historical growing season is 1 year away from the current growing season. Furthermore, wheat yield data for each plot of land in the current growing season is also required, recorded as , X represents plot X. At the same time, it is also necessary to fill in the missing data in temperature, precipitation, soil moisture and biomass data to facilitate subsequent calculations. There are many algorithms for data filling, and implementers can choose according to actual conditions.
[0047] Step S2, according to the Pearson correlation coefficient between each current environmental feature sequence and the current biomass sequence, and the Pearson correlation coefficient between each historical environmental feature sequence and the historical biomass sequence, obtain the time-varying importance factor corresponding to each current environmental feature sequence; use the time-varying importance factor to optimize the current environmental feature sequence to obtain the optimized environmental feature sequence.
[0048] During the growth of wheat, the impact of environmental factors on biomass will change dynamically over time, and the traditional static feature importance assessment method cannot capture this dynamicity. As a result, it is difficult for the model to adjust the importance weights of each factor when facing extreme weather events or sudden pests and diseases. To solve this problem, the present invention introduces time-varying importance factors. The calculation of time-varying importance factors requires dividing the collected wheat growth environment data and biomass data into two comparative data sets: the recent growth period and the historical growth period. The current growth period data reflects the current growth environment. The same selection is made for the historical growth period data corresponding to the time.
[0049] For each environmental feature (temperature, precipitation, soil moisture) of the data series, the correlation coefficients between it and wheat biomass in the current growth period and the historical growth period are calculated respectively. Here, the Pearson correlation coefficient is used to reflect the degree of linear correlation. By comparing the correlation coefficients of the current growth period and the historical growth period, the temporal changes in feature importance can be found. In order to quantify this change, the absolute value of the difference between the correlation coefficients of each feature in the current growth period and the historical growth period is calculated. The absolute value of the difference directly reflects the temporal change in feature importance. The larger the difference, the more significant the change in the importance of the feature. In order to make the time-varying importance factors of different features comparable, the calculated absolute values of the difference are normalized. The normalization process ensures that the time-varying importance factors of different features can be compared on the same scale, which is convenient for subsequent weight adjustment. At the same time, considering that data with a larger time interval have less impact on the current prediction, the time decay factor is introduced. The more recent data have a greater impact on the time-varying importance factor. The introduction of the time decay factor makes the time-varying importance factor pay more attention to recent changes.
[0050] Through the above calculation process, the time-varying importance factor value of each environmental feature of each plot in the current growing period can be obtained. A higher time-varying importance factor value indicates that the importance of the feature is changing significantly.
[0051] Specifically, it is necessary to obtain the Pearson correlation coefficient between each current environmental feature sequence and the current biomass sequence, as well as the Pearson correlation coefficient between each historical environmental feature sequence and the historical biomass sequence. The calculation formula is:
[0052] , ,in, represents the Pearson correlation coefficient between the i-th current environmental characteristic sequence and the current biomass sequence; represents the Pearson correlation coefficient between the i-th historical environmental characteristic sequence and the historical biomass sequence t years away from the current growing season; represents the i-th current environment feature sequence, Represents the current biomass sequence; represents the i-th historical environmental characteristic sequence t years away from the current growing period, Represents the historical biomass sequence t years from the current growing season.
[0053] Considering that data from the current growth period may have little impact on the current forecast, a time decay factor is introduced: .in, is the time attenuation coefficient, which is set to 0.5 in the embodiment of the present invention and can be optimized according to the actual clustering effect. is the time interval. Next, calculate the time-varying importance factor In order to make the time-varying importance factors of different features comparable, normalization processing is required during calculation. Specifically, the calculation formula of the time-varying importance factor is:
[0054] ,
[0055] in, represents the time-varying importance factor corresponding to the i-th current environmental feature sequence; represents the Pearson correlation coefficient between the i-th current environmental characteristic sequence and the current biomass sequence; represents the Pearson correlation coefficient between the i-th historical environmental characteristic sequence and the historical biomass sequence t years away from the current growing period; T represents the maximum number of years from the historical growing period to the current growing period; represents the time decay factor; Indicates the number of types of environmental characteristic data. It is the absolute difference of Pearson's coefficient, reflecting the degree of temporal variation of the time-varying importance factor.
[0056] Thus, the time-varying importance factor corresponding to each current environmental feature sequence can be obtained, and the current environmental feature sequence can be optimized using the time-varying importance factor. Specifically, each current environmental feature sequence is multiplied by the corresponding time-varying importance factor to obtain the optimized environmental feature sequence. The calculation formula is: = * , Represents the optimized environment feature sequence corresponding to the i-th current environment feature sequence.
[0057] When optimizing the multi-model fusion method for wheat yield estimation, this step introduces a time-varying importance factor. Traditional static feature importance assessment methods cannot fully capture the dynamic impact of environmental characteristics on wheat biomass. The time-varying importance factor quantifies the temporal changes in the importance of environmental characteristics by comparing the correlation between the current growth period and the historical growth period data. Obtaining dynamic assessments enables the model to adjust the weights of various environmental factors in a timely manner, especially in response to environmental changes when facing extreme weather events or sudden pests and diseases. The introduction of time-varying importance factors provides feature importance assessment capabilities for subsequent k-means clustering analysis.
[0058] Step S3, evenly segmenting the optimized environmental feature sequence; obtaining the slope of each segment of the segmented optimized environmental feature sequence; obtaining the environmental sensitivity factor corresponding to each segment of the optimized environmental feature sequence according to the slope; optimizing the optimized environmental feature sequence using the environmental sensitivity factor to obtain a first optimized environmental feature sequence.
[0059] Under the optimization of step S2, the model is able to dynamically adjust the weights of features according to the temporal changes in the importance of environmental features, which improves the adaptability of the model to trend changes. However, this optimization focuses mainly on the dynamic changes in feature importance without directly considering the sensitivity of environmental factors to the immediate effects of wheat biomass. In actual agricultural production, small changes in environmental factors can sometimes have an immediate and significant impact on biomass, especially in extreme weather events or sudden outbreaks of pests and diseases. Therefore, even if the importance of a feature does not change significantly, its immediate impact may be a key factor in determining the accuracy of biomass prediction.
[0060] Although the time-varying importance factor introduced in step S2 can capture the overall trend of changes in the importance of environmental features, it still has limitations in dealing with short-term, sudden environmental changes. For example, the overall importance of an environmental factor (such as precipitation) may not change significantly during a growing season, but a small change in precipitation in the short term during a critical growth stage (such as flowering) may have a huge impact on biomass. The time-varying importance factor may not be able to reflect this short-term sensitivity change in a timely manner.
[0061] On the basis of step S2, step S3 of the present invention introduces an environmental sensitivity factor, and the introduction of the environmental sensitivity factor is intended to quantify the immediate impact of different environmental factors on wheat biomass. By performing local linear regression analysis within a set time window, the environmental sensitivity factor can capture the short-term relationship changes between environmental factors and biomass. This method not only takes into account the impact of environmental factors on biomass, but also combines the change range of environmental factors themselves, thereby fully reflecting the dynamic importance of environmental factors.
[0062] In order to capture environmental changes in the short term, it is necessary to process the optimized environmental feature sequence obtained in step S2, specifically, to obtain a smaller time window. The window size is set according to the crop growth characteristics, and is set to 7 days in the present invention. The implementer can adjust it according to the actual situation. This sliding window mechanism can effectively capture the short-term fluctuations of environmental factors and enhance the ability to identify sudden changes. In each window, a local linear regression analysis is performed on each environmental feature and biomass. Local analysis can capture the local linear features in the nonlinear relationship and more accurately reflect the immediate impact of environmental factors on biomass. The slope value is extracted from the local linear regression as a sensitivity index of the impact of environmental factors on biomass. The larger the absolute value of the slope, the more significant the impact of the slight change of the environmental factor on biomass. This method can accurately quantify the impact of different environmental factors on biomass. At the same time, the relative rate of change of each environmental factor in the current time window is calculated, reflecting the range of change of the environmental factor itself. The sensitivity index (absolute value of the slope) is combined with the rate of change of the environmental factor to obtain a preliminary environmental sensitivity factor value. The environmental sensitivity factor takes into account the impact of the environmental factor and its own range of change, and fully reflects the importance of the environmental factor. It can further capture the short-term sensitivity changes of environmental factors based on the long-term importance changes provided by the time-varying importance factor. This dual optimization mechanism enables the model to more accurately adjust the distance metric when performing k-means clustering, which not only considers the long-term importance of the feature, but also reflects its short-term sensitivity.
[0063] For example, when an environmental factor shows high sensitivity in the short term (high environmental sensitivity factor value), even if its long-term time-varying importance factor is relatively low, it will be given a higher weight when calculating the distance between samples. This optimization ensures that the clustering results can better reflect the impact of current environmental conditions on wheat growth, thereby improving the accuracy of wheat yield prediction.
[0064] A time window is set. Preferably, the time window P of the present invention is 7. The time window is used to slide in the optimized environment feature sequence, and the optimized environment feature sequence is evenly segmented. The length of each data segment is 7, and the sliding step of the time window is the length of the time window.
[0065] At the same time, the time window is used to segment the current biomass sequence, and then each segment of the data in the segmented optimized environmental feature sequence and each segment of the corresponding current biomass sequence are used for local linear regression analysis to obtain the local expression:
[0066] ,
[0067] in, Represents the data in the pth time window in the current biomass sequence, a total of 7; Represents the data in the pth time window in the optimized environment feature sequence corresponding to the i-th current environment feature sequence, and there are also 7; and are the first and second regression coefficients, respectively; is the error term. The second regression coefficient is the slope of the data in the pth time window in the optimized environmental feature sequence, which is also the slope of the pth segment of data. It should be noted that the data in the pth time window is also the data in the pth segment of data after segmentation in the sequence.
[0068] From the regression results, the absolute value of the slope is extracted as a sensitivity indicator: . It reflects the impact of changes in environmental factors on biomass.
[0069] Furthermore, the rate of change of environmental factors is calculated, that is, the rate of change of the data in the pth time window in the optimized environmental feature sequence: ,in, Represents the standard deviation of the data in the pth time window in the optimized environment feature sequence corresponding to the i-th current environment feature sequence; Represents the average value of the data in the pth time window in the optimized environment feature sequence corresponding to the i-th current environment feature sequence.
[0070] According to the absolute value of the change rate and slope of each data segment in the optimized environmental characteristic sequence, the environmental sensitivity factor corresponding to each data segment is obtained. The calculation formula is: ,in, represents the environmental sensitivity factor corresponding to the p-th segment of data in the optimized environmental feature sequence corresponding to the i-th current environmental feature sequence; Represents the normalized value of the absolute value of the slope of the p-th segment of data in the optimized environmental feature sequence corresponding to the i-th current environmental feature sequence; It represents the normalized value of the change rate of the pth segment of data in the optimized environmental feature sequence corresponding to the i-th current environmental feature sequence. Before calculating the environmental sensitivity factor, the absolute value and change rate of the slope need to be normalized.
[0071] Furthermore, according to the environmental sensitivity factor of each data segment in the optimized environmental feature sequence, the environmental sensitivity factor is used for secondary optimization to obtain the data value of each plot for distance measurement, so as to achieve the purpose of secondary optimization of the optimized environmental feature sequence. The calculation formula is: ,
[0072] in, represents the first optimized environment feature sequence corresponding to the i-th current environment feature sequence; represents the weight coefficient; represents the optimized environment feature sequence corresponding to the i-th current environment feature sequence; It represents the environmental sensitivity factor corresponding to the pth segment of data in the optimized environmental feature sequence corresponding to the i-th current environmental feature sequence. The initial value of the weight α is set to 0.5, which can be determined by cross-validation based on actual results. During the calculation process, The data in and the corresponding environmental sensitivity factor want to be calculated, for example If a certain data in is the data in the fifth time window, that is, the data in the fifth segment, then The p in is 5.
[0073] By using the standard deviation and mean to calculate the rate of change, the changing trend of environmental characteristics is captured more effectively and the interference caused by a single extreme value is avoided. This method comprehensively considers the long-term trend and short-term fluctuations of environmental characteristics, making the model more accurate and robust in responding to sudden environmental changes.
[0074] Step S4, combining various first optimized environmental feature sequences and current biomass sequences of each plot of land into a classification basis set; classifying each plot of land using the classification basis set; obtaining a yield prediction model for each type of land to predict wheat yield.
[0075] The wheat growth environment data was optimized by using the time-varying importance factor and environmental sensitivity factor to obtain the first optimized environmental feature sequence for each plot. The data space used for clustering calculation specifically combines the first environmental feature sequence and the current biomass sequence to form a classification basis set , in the classification basis set, any data point is represented as: ,in They respectively represent the g-th data in the first optimized environmental feature sequence, that is, the optimized value of the g-th data in the three current environmental feature sequences of a piece of land.
[0076] because is multivariate time series data, and in order to further process this data for use in k-means clustering, we use dynamic time warping ( ) Calculate the distance between the classification basis sets of each two plots. Used as the distance measure between data points in the subsequent k-means clustering calculation (each plot is regarded as a data point). In the calculation, the elbow rule is used to determine the K value and perform cluster division. The algorithm and k-means clustering process are both existing algorithms and will not be described in detail here.
[0077] After completing the k-means clustering, clusters of wheat samples with similar growth characteristics and environmental conditions were obtained. Each cluster represents a group of wheat samples (plots) with similar attributes. A dedicated yield prediction model needs to be assigned to each category of plots. These yield prediction models include machine learning models such as linear regression, random forest, and support vector machine, or physical models based on crop growth models. The same model can be used for plots of different categories.
[0078] Specifically, a model pool is established that contains all the prediction models that need to be used. The data of each category of plots is divided into 70% training set and 30% test set. Since each plot is regarded as a data point and there is no time series relationship between plots, it is not necessary to maintain the continuity of continuous data when dividing the data and it can be divided randomly. Then, for each category of plots, all models in the model pool are trained using its training data set (environmental characteristic data and corresponding wheat yield data of each plot in each category of plots). The yield prediction performance of each model is then evaluated using the test set. The evaluation indicator is: root mean square error (RMSE) (mean absolute error (MAE), R² coefficient of determination, etc. can also be used). Based on the evaluation results, the best yield prediction model is then selected for each category of land. The above are existing methods and will not be described in detail here.
[0079] After assigning a yield prediction model to each category of plots, use these models to predict the yield of new plots. For the data point of a new plot, it is first necessary to determine which cluster it is closest to. This is achieved by calculating the distance between the new data point and the cluster center of each category of plots. The original data is optimized using the time-varying importance factor and environmental sensitivity factor used in the above calculation, and then the new plot will be assigned to the closest cluster, that is, to determine which category of plot the new plot belongs to, and then the yield prediction model of the cluster is used to predict the wheat yield of the new plot.
[0080] Step S4 uses the above optimized data set to perform k-means clustering. The purpose is to group wheat samples with similar growth characteristics and environmental conditions. When calculating the distance between samples, the weight of each feature is dynamically adjusted according to its time-varying importance factor and environmental sensitivity factor. A high time-varying importance factor value indicates that the importance of the feature is changing significantly and should be given a higher weight; a high environmental sensitivity factor value indicates that the environmental factor currently has a greater impact on biomass and should also be given a higher weight. The time-varying importance factor provides the long-term trend of feature importance, and the environmental sensitivity factor captures the short-term fluctuations of environmental characteristics. This method can automatically adapt to changes in environmental conditions and adjust the prediction strategy under different growth stages and environmental scenarios. For example, during droughts, environmental characteristics related to soil moisture will obtain higher time-varying importance factor and environmental sensitivity factor values, thereby receiving more attention in the clustering process, and ultimately obtaining more accurate yield predictions under drought conditions.
[0081] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The above is a description of a specific embodiment of this specification. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0082] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.
[0083] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims
1. A multi-model fusion method based on wheat yield estimation, characterized in that: The method includes: Obtain the environmental characteristic data and biomass data of wheat in the current growth period and historical growth period of each plot of land, and form the current environmental characteristic sequence, current biomass sequence, historical environmental characteristic sequence, and historical biomass sequence respectively; According to the Pearson correlation coefficient between each current environmental characteristic sequence and the current biomass sequence, and the Pearson correlation coefficient between each historical environmental characteristic sequence and the historical biomass sequence, the time-varying importance factor corresponding to each current environmental characteristic sequence is obtained; the current environmental characteristic sequence is optimized using the time-varying importance factor to obtain an optimized environmental characteristic sequence; Evenly segment the optimized environmental feature sequence; obtain the slope of each segment of the segmented optimized environmental feature sequence; obtain the environmental sensitivity factor corresponding to each segment of the optimized environmental feature sequence according to the slope; optimize the optimized environmental feature sequence using the environmental sensitivity factor to obtain a first optimized environmental feature sequence; Combining various first optimized environmental feature sequences and current biomass sequences of each plot of land into a classification basis set; classifying each plot of land using the classification basis set; obtaining a yield prediction model for each type of land to predict wheat yield; The time-varying importance factor is: , in, represents the time-varying importance factor corresponding to the i-th current environmental feature sequence; represents the Pearson correlation coefficient between the i-th current environmental characteristic sequence and the current biomass sequence; represents the Pearson correlation coefficient between the i-th historical environmental characteristic sequence and the historical biomass sequence t years away from the current growing period; T represents the maximum number of years from the historical growing period to the current growing period; represents the time decay factor; Indicates the number of types of environmental characteristic data; The step of obtaining the environmental sensitivity factor corresponding to each segment of data in the optimized environmental feature sequence according to the slope includes: Calculate the ratio of the standard deviation and the average value of each data segment in the optimized environment feature sequence as the corresponding change rate of each data segment; According to the absolute value of the change rate and slope of each data segment in the optimized environmental characteristic sequence, the environmental sensitivity factor corresponding to each data segment is obtained. The calculation formula is: , in, represents the environmental sensitivity factor corresponding to the p-th segment of data in the optimized environmental feature sequence corresponding to the i-th current environmental feature sequence; Represents the normalized value of the absolute value of the slope of the p-th segment of data in the optimized environmental feature sequence corresponding to the i-th current environmental feature sequence; Represents the normalized value of the change rate of the p-th segment of data in the optimized environmental feature sequence corresponding to the i-th current environmental feature sequence; The calculation formula of the first optimized environmental feature sequence is: , in, represents the first optimized environment feature sequence corresponding to the i-th current environment feature sequence; represents the weight coefficient; represents the optimized environment feature sequence corresponding to the i-th current environment feature sequence; Represents the environmental sensitivity factor corresponding to the p-th segment of data in the optimized environmental feature sequence corresponding to the i-th current environmental feature sequence.
2. A multi-model fusion method based on wheat yield estimation according to claim 1, characterized in that: The environmental characteristic data and biomass data include: The environmental characteristic data are three types of data, namely temperature, precipitation, and soil moisture; the biomass data are the weight of all wheat plants in a field at a moment.
3. The multi-model fusion method based on wheat yield estimation according to claim 1, characterized in that: The method of optimizing the current environmental feature sequence by using the time-varying importance factor to obtain the optimized environmental feature sequence includes: Each current environmental feature sequence is multiplied by the corresponding time-varying importance factor to obtain the optimized environmental feature sequence.
4. The multi-model fusion method based on wheat yield estimation according to claim 1, characterized in that: The step of uniformly segmenting the optimized environment feature sequence comprises: A time window is set, and the time window is used to slide in the optimized environment feature sequence to evenly segment the optimized environment feature sequence, wherein the sliding step length of the time window is the length of the time window.
5. The multi-model fusion method based on wheat yield estimation according to claim 1, characterized in that: The step of obtaining the slope of each segment of data in the segmented optimized environment feature sequence includes: The current biomass sequence is segmented using the time window; local linear regression analysis is performed using each segment of the segmented optimized environmental feature sequence and each segment of the corresponding current biomass sequence to obtain the local expression: , in, Represents the data in the pth time window in the current biomass sequence; Represents the data in the pth time window in the optimized environment feature sequence corresponding to the i-th current environment feature sequence; and are the first and second regression coefficients, respectively; is the error term; The second regression coefficient is the slope of the data in the pth time window in the optimized environmental feature sequence, that is, the slope of the pth segment of data.
6. The multi-model fusion method based on wheat yield estimation according to claim 1, characterized in that: The method of classifying each piece of land by using a classification basis set includes: The distance between the classification basis sets of every two plots of land is calculated, and each plot of land is classified according to the distance.
