Deep survival analysis-based peronophythora litchii risk prediction method
Through deep survival analysis model based on deep learning, key features are automatically learned and the risk of Phytophthora lychee cream is solved, and the problem of difficulty in effectively predicting Phytophthora lychee cream is achieved in the existing technology, achieving higher prediction accuracy and interpretability.
Patent Information
- Application Number
- CN202510084175.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-20
- Publication Date
- 2025-05-13
AI Technical Summary
The prior art is difficult to effectively predict the risk of Phytophthorasis in lychee cream, especially when dealing with complex nonlinear relationships and multidimensional features, traditional methods have the possibility of over-learning risks and artificial errors.
Using a deep survival analysis model based on deep learning, we use the deep survival analysis model to construct a feed-forward deep neural network to automatically learn key features and predict the risk of Phytophthora lychee cream, and use the Cox proportional hazard model to output survival probability and incidence risk levels.
Accurate prediction of the risk of Phytophthora lychee cream is achieved, the possibility of artificial intervention and error is reduced, the prediction accuracy and interpretability of the model are improved, and the limitations of traditional methods in nonlinear modeling and feature extraction are overcome.
Smart Images
Figure CN119993491A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of crop disease prevention and control, and relates to a litchi downy mildew disease risk prediction method based on deep survival analysis. Background Art
[0002] At present, most of the research on litchi downy mildew is based on disease prevention and control, and most of them are biological and chemical control. There is still a large gap in the application of electronic information technology.
[0003] Although the agronomy field has found a certain relationship between the occurrence of downy mildew and environmental factors such as temperature and humidity through field surveys, there is a lack of specific computer models to prove the specific relationship between the risk of litchi downy mildew and environmental factors. It is also impossible to use digital form to give specific risk data for the occurrence of downy mildew in the future through environmental factors.
[0004] Survival Analysis is a statistical method for analyzing and modeling the distribution of life expectancy or time to event. Also known as life data analysis or time to event analysis, it is a highly specialized branch of statistics. It focuses on exploring and quantifying the probability distribution of the time to an event, which usually involves some form of "failure" or "termination".
[0005] The interaction between downy mildew and environmental factors is obviously nonlinear. Traditional survival analysis methods usually assume that the relationship between variables and survival time is linear, which is insufficient when dealing with complex nonlinear relationships. In addition, traditional survival analysis methods have high requirements for data integrity and quality, and are sensitive to missing values and noisy data, which may affect the reliability of the results.
[0006] Although machine learning-based survival analysis methods (such as decision trees, random survival forests, support vector machines, and artificial neural networks) have overcome some limitations of traditional survival analysis methods to a certain extent, these methods still require manual feature selection and extraction during the training process, which will increase the cost of preprocessing for multidimensional features and may introduce human errors. In addition, because the interaction terms between features in survival analysis are often very complex. Traditional machine learning models (such as decision trees, support vector machines, etc.) require experts to manually design feature interaction terms or manually add interaction terms to the model. This not only increases the workload, but also fails to fully capture all potential interaction relationships.
[0007] Although traditional complex machine learning models also use some regularization methods, they still have a high risk of over-learning.
[0008] Therefore, how to provide a litchi downy mildew risk prediction method based on deep survival analysis that can effectively predict the occurrence risk of litchi downy mildew according to complex nonlinear relationships is an urgent problem that technicians in this field need to solve. Summary of the invention
[0009] In view of this, the present invention proposes a risk prediction method for litchi downy mildew based on deep survival analysis. The deep survival analysis model based on deep learning re-explains the interaction mechanism between the occurrence of litchi downy mildew and environmental factors, explains the complex nonlinear relationship therein, and effectively predicts the occurrence risk of downy mildew according to environmental factors.
[0010] In order to achieve the above object, the present invention adopts the following technical solution:
[0011] The present invention discloses a risk prediction method for litchi downy mildew based on deep survival analysis, comprising the following steps:
[0012] Constructing a feedforward deep neural network, including an input layer for receiving a feature matrix, wherein the input layer transmits the feature matrix data to a plurality of hidden layers connected in sequence, wherein the hidden layers include a fully connected layer and a dropout layer connected in sequence; the last hidden layer is connected to the risk function output layer through a linear layer with a single node, thereby obtaining a deep survival analysis prediction model;
[0013] Collecting litchi tree living environment data and fruit tree downy mildew infection rate data in a corresponding time period to construct a training data set, and using the training data set to train the deep survival analysis prediction model based on a back propagation algorithm to obtain a trained deep survival analysis prediction model;
[0014] The real-time living environment data of litchi trees is used as the feature matrix to input into the trained deep survival analysis prediction model to predict the survival probability of individual litchi trees and obtain the disease risk level data of individual litchi trees.
[0015] Preferably, the litchi tree living environment data includes temperature, humidity and light intensity data.
[0016] Preferably, the living environment data of each individual litchi tree in the training data set includes numerical data, categorical data and covariate data; the living environment data of the litchi tree is preprocessed, and the survival time of the individual litchi tree and the occurrence data of fruit tree disease events are marked; the real-time living environment data of the litchi tree includes numerical data and categorical data.
[0017] Preferably, the step of training the deep survival analysis prediction model based on the back propagation algorithm includes: optimizing through the back propagation algorithm to obtain individual risk scores and hyperparameters of the best deep survival analysis prediction model.
[0018] Preferably, the risk function output layer adopts the Cox proportional hazard model, in which the individual risk score r is used i The survival probability S(t i ):
[0019]
[0020] In the formula, h 0 (t i ) is the baseline risk function of individual i at time point t, r i is the individual risk score, S(t i is the survival probability of a fruit tree at a given time point in the future, and the risk of downy mildew at that time point is 1-S(t i ).
[0021] It can be seen from the above technical solution that, compared with the prior art, the beneficial effects of the present invention include:
[0022] (1) Adaptive learning. The present invention does not require the selection of covariates in advance, and the model can automatically learn key features from the data, which reduces the possibility of human intervention and errors and improves the prediction accuracy of the model.
[0023] (2) Nonlinear modeling capability. Thanks to the nonlinear activation function of deep neural networks, the present invention captures the complex nonlinear relationship between covariates and risks, which is unmatched by the traditional Cox model. It better simulates the complex relationship between covariates and risks, does not need to limit the distribution of survival time, finds the interaction mechanism between litchi downy mildew and environmental factors, and outputs the risk of future downy mildew more accurately. It overcomes the problem of the lack of scientific and effective tools for predicting the occurrence of downy mildew in litchi planting, and has higher interpretability, accuracy and versatility of survival prediction in overall prediction.
[0024] (3) Automatic feature extraction. Compared with the traditional method that requires manual selection and extraction of features, the present invention can automatically extract important features from the data, simplifying the model building process.
[0025] (4) The present invention uses a neural network to replace the linear part in the traditional Cox model, thereby achieving more flexible mapping and more accurate prediction.
[0026] (5) The present invention can make individualized survival probability predictions based on the multidimensional characteristic data of the litchi living environment, and provide personalized disease risk warnings accordingly, which is of great significance for improving the precise prevention and control level of litchi downy mildew and the economic benefits of orchards. BRIEF DESCRIPTION OF THE DRAWINGS
[0027] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art are briefly introduced below. Obviously, the drawings in the following description are only embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on the provided drawings without creative work.
[0028] Figure 1 A flow chart of a method for predicting litchi downy mildew risk based on deep survival analysis provided in an embodiment of the present invention;
[0029] Figure 2 A structural diagram of a deep survival analysis prediction model provided by an embodiment of the present invention;
[0030] Figure 3 A diagram of the execution process of the litchi downy mildew disease risk prediction method based on deep survival analysis provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0031] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in 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. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0032] Litchi downy mildew is a common disease in litchi cultivation. In China, it often breaks out from March to June each year, and when it breaks out, it can easily spread rapidly in the orchard, causing a great impact on the yield and quality of litchi. It is a serious plant disease.
[0033] Survival analysis refers to a method of analyzing and inferring the survival time of the observed subjects based on the data obtained from experiments or surveys, and studying the relationship between survival time and outcomes and many influencing factors, as well as the extent of the relationship.
[0034] Deep survival analysis is a survival analysis method based on deep learning. It combines deep neural networks and survival analysis models to predict the survival time or survival probability of patients. Deep survival analysis uses the powerful learning ability of deep neural networks to capture complex features and patterns in the data, thereby improving the accuracy of survival time prediction. Figure 1As shown, an embodiment of the present invention provides a litchi downy mildew disease risk prediction method based on deep survival analysis. The litchi downy mildew disease prediction method based on deep survival analysis uses environmental variables such as temperature, humidity, and light intensity in which fruit trees survive as features to predict the disease risk of fruit trees at future time points and provide prediction prompts for disease risk levels as needed.
[0035] like Figure 1 As shown, the specific steps include:
[0036] Construct a configurable feedforward deep neural network, including an input layer for receiving a feature matrix, which transmits the feature matrix data to multiple hidden layers connected in sequence, and the hidden layers include fully connected layers and dropout layers connected in sequence; the last hidden layer is connected to the risk function output layer through a linear layer with a single node to obtain a deep survival analysis prediction model;
[0037] Collect the living environment data of litchi trees and the infection rate data of downy mildew of fruit trees in the corresponding time period to construct a training data set, and use the training data set to train a deep survival analysis prediction model based on the back propagation algorithm to obtain a trained deep survival analysis prediction model;
[0038] The real-time living environment data of litchi trees is used as the feature matrix to input into the trained deep survival analysis prediction model to predict the survival probability of individual litchi trees and obtain the disease risk level data of individual litchi trees.
[0039] The deep survival analysis prediction model of this embodiment adopts a deep neural network, such as Figure 2 As shown in Figure 1, the network propagates inputs through many hidden layers. The hidden layers consist of fully connected nonlinear activation functions followed by dropout. The last layer is a single node that performs linear combinations of hidden features. The output layer of the network is constructed in the form of a Cox proportional hazard model, which aims to predict the survival risk of an individual, and its output is a survival risk value.
[0040] Through multi-layer nonlinear transformation, deep neural networks can gradually extract high-dimensional features of input data at different levels. The multi-layer nonlinear structure of deep neural networks, where the output of each layer of the network is a nonlinear transformation of the input of the previous layer, enables the network to automatically extract important features from the original input data, not only to find more accurate mappings in complex data patterns, but also to automatically learn and model complex feature interactions without manual intervention, thereby improving the efficiency of the model.
[0041] The deep survival analysis model uses the dropout regularization technique to randomly "discard" some neurons in the network during training, forcing the model to rely on different combinations of neurons each time it is trained, thereby reducing over-reliance on a single influencing feature. In addition, the deep survival analysis method applies L2 regularization to the weights of the network, penalizing parameters with excessive weights in the model, thereby preventing the model from overfitting the noise of the training data, helping the model maintain a small complexity, and significantly reducing the risk of over-learning.
[0042] The deep neural network is an end-to-end training process. This training method incorporates tasks such as feature learning, risk prediction, and survival time modeling into a framework. The training process itself is an optimization process that can automatically adjust network parameters to find the most important combination among multiple features, reducing the work of manual selection. End-to-end training can optimize data feature extraction and prediction results simultaneously through batch processing, avoiding the accumulation of pallets between various links in traditional machine learning methods. In addition, during the training process, the deep survival analysis method dynamically adjusts the learning rate through the Adam optimizer, which not only helps to accelerate training, but also avoids overfitting caused by learning too quickly.
[0043] In one embodiment, the litchi tree living environment data includes temperature, humidity and light intensity data.
[0044] In this embodiment, Figure 2 As shown in the figure, by setting up a weather station in the litchi orchard to predict the sensors, the temperature, humidity, light intensity and other environmental data of the litchi tree living environment are collected in real time. Field surveys are conducted regularly to obtain the infection rate data of fruit tree downy mildew.
[0045] In one embodiment, the living environment data of each individual litchi tree in the training data set includes numerical data, categorical data and covariate data; the litchi tree living environment data is preprocessed, and the survival time of the individual litchi tree and the occurrence data of the fruit tree disease events are marked; the real-time litchi tree living environment data includes numerical data and categorical data.
[0046] In this implementation, based on the principle of survival analysis, there is no need to rely on expert knowledge to select features from raw data, but to directly preprocess the collected multidimensional raw data. For each observed individual, numerical data can be directly used as input, and categorical data is first encoded and then used as input. All covariates form a feature matrix x i And mark the survival time t of the observed individuals according to the field survey results i and the occurrence of fruit tree diseases i (represents the individual at observation time t iWhether a termination event occurs, if so, it is recorded as 1, otherwise it is recorded as 0).
[0047] In one embodiment, the step of training the deep survival analysis prediction model based on the back propagation algorithm includes: optimizing through the back propagation algorithm to obtain the individual risk score and the hyperparameters of the best deep survival analysis prediction model.
[0048] In this embodiment, the pre-processed survival data set (X, T, E) of the fruit trees is input into the deep survival analysis model for training, and the individual risk score r is obtained by optimization through the back propagation algorithm. i Among them, the survival time provides the target that the model needs to predict, and the event marker reminds the model to consider the individual's missing situation. The model predicts the individual's risk score r by learning the relationship between the feature input and the survival time and event marker. i , which represents the survival risk of an individual relative to other individuals, with higher scores representing a higher risk of an event occurring.
[0049] In one embodiment, the risk function output layer adopts the Cox proportional risk model, in which the risk score and the survival function have the following relationship:
[0050] h(t i )=h 0 (t)·exp(r i )
[0051] In the formula, h(t i ) is the risk of individual i at time point t, h 0 (t) is the base risk.
[0052] The prediction process uses the overall information of the training data to calculate the baseline risk function h 0 (t), and then use the individual risk score r i The survival probability S(t i ):
[0053]
[0054] In the formula, S(t i 0 is the survival probability of the fruit tree at a given time point in the future, and the risk of downy mildew at that time point is 1-S(t i ).
[0055] In specific implementation, the risk of disease can be set to be less than or equal to 0.1 as a low risk level, the risk of disease greater than 0.1 and less than or equal to 0.5 as a medium risk level, and the risk of disease greater than 0.5 as a high risk level. Finally, the predicted individual risk level is output.
[0056] In one embodiment, the deep survival analysis model can also be customized and optimized according to the needs of specific problems, for example, by adjusting the network structure, optimizing the algorithm, etc. to improve the performance of the model.
[0057] The prediction method for litchi downy mildew based on deep survival analysis focuses on the survival time of litchi trees and the occurrence of disease events. Through the above technical scheme, it is possible to explain the interaction between litchi downy mildew and environmental variables from the perspective of time series data, and can predict the risk of fruit trees being infected with downy mildew under the influence of future environmental factors according to actual needs, thereby achieving the purpose of using electronic information technology to assist orchards in early prevention and control and early warning of possible outbreaks of litchi downy mildew in the future.
[0058] The deep survival analysis proposed in the present invention has stronger generalization ability, can process high-dimensional data and nonlinear relationships, and is applicable to various types of survival data. The litchi downy mildew prediction method based on deep survival analysis can effectively use the time series data of the litchi living environment for the training of the deep survival analysis model, thereby generating a prediction model and predicting the survival probability and disease risk of the litchi fruit trees according to the changes in the future environment of the litchi fruit trees over time, and giving the disease risk level of the fruit trees at a future time point according to actual needs, which is helpful to predict the occurrence and spread of litchi diseases. For traditional survival analysis methods, dealing with multivariate dependence and competing risk problems may be a challenge, and deep survival analysis can better solve such problems, provide decision-making basis for the precise prevention and control of litchi downy mildew, and assist orchard managers in carrying out precise prevention and control of litchi downy mildew in advance.
[0059] The above is a detailed introduction to the litchi downy mildew disease risk prediction method based on deep survival analysis provided by the present invention. In this embodiment, specific examples are used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for general technical personnel in the field, according to the idea of the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.
[0060] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined in the present embodiments may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown in the present embodiment, but will conform to the widest range consistent with the principles and novel features disclosed in the present embodiment.
Claims
1. A risk prediction method for litchi downy mildew based on deep survival analysis, characterized in that: The steps include: Constructing a feedforward deep neural network, including an input layer for receiving a feature matrix, wherein the input layer transmits the feature matrix data to a plurality of hidden layers connected in sequence, wherein the hidden layers include a fully connected layer and a dropout layer connected in sequence; the last hidden layer is connected to the risk function output layer through a linear layer with a single node, thereby obtaining a deep survival analysis prediction model; Collecting litchi tree living environment data and fruit tree downy mildew infection rate data in a corresponding time period to construct a training data set, and using the training data set to train the deep survival analysis prediction model based on a back propagation algorithm to obtain a trained deep survival analysis prediction model; The real-time living environment data of litchi trees is used as the feature matrix to input into the trained deep survival analysis prediction model to predict the survival probability of individual litchi trees and obtain the disease risk level data of individual litchi trees.
2. The method for predicting litchi downy mildew risk based on deep survival analysis according to claim 1, characterized in that: The litchi tree living environment data includes temperature, humidity and light intensity data.
3. The method for predicting litchi downy mildew risk based on deep survival analysis according to claim 1, characterized in that: The living environment data of each individual litchi fruit tree in the training data set includes numerical data, categorical data and covariate data; the living environment data of the litchi fruit trees are preprocessed, and the survival time of the individual litchi fruit trees and the occurrence data of the fruit tree disease events are marked; The real-time litchi tree living environment data includes numerical data and categorical data.
4. The method for predicting litchi downy mildew risk based on deep survival analysis according to claim 1, characterized in that: The step of training the deep survival analysis prediction model based on the back propagation algorithm includes: optimizing through the back propagation algorithm to obtain individual risk scores and hyperparameters of the best deep survival analysis prediction model.
5. The method for predicting litchi downy mildew risk based on deep survival analysis according to claim 4, characterized in that: The risk function output layer adopts the Cox proportional hazard model, in which the individual risk score r is used i The survival probability S(t i ): In the formula, h 0( u ) is the benchmark risk function, r i is the individual risk score, S ( t i) is the survival probability of a fruit tree at a given time point in the future, where the risk of downy mildew at that time point is 1-S ( t i) .