Grape growth environment monitoring method and system based on artificial intelligence

Abstract

The invention relates to the technical field of environment monitoring, in particular to a grape growth environment monitoring method and system based on artificial intelligence, and the method comprises the steps: constructing a multi-dimensional environment data matrix; performing standardization processing on the multi-dimensional environment data matrix, and extracting dynamic trend features to obtain a standardized environment feature matrix; constructing a nonlinear relation model; if it is detected that the environmental data change exceeds a preset threshold value, sub-module parameters of the nonlinear relation model are updated; adjusting weight distribution according to the updated sub-module parameters; verifying the convergence of weight distribution by calibrating the nonlinear relation model; predicting an environment parameter change sequence according to the calibrated nonlinear relation model; and comparing the deviation between the predicted environmental parameter change sequence and the actual data, and if the deviation exceeds a preset tolerance threshold, re-optimizing the sub-module parameters. According to the invention, the problem that the monitoring precision is not high when the environmental parameters change is solved, and the environmental monitoring precision is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of environmental monitoring, and in particular to a grape growing environment monitoring method and system based on artificial intelligence. BACKGROUND

[0002] In the process of grape cultivation, environmental parameters such as temperature, humidity, and light have a significant impact on grape growth and yield. However, these parameters in the agricultural environment exhibit high dynamicity and nonlinearity due to complex and variable natural conditions. Traditional environmental monitoring and prediction models often struggle to accurately capture these trends and their interactions, resulting in insufficient prediction accuracy and affecting the reliability of grape cultivation management decisions.

[0003] Specifically, while environmental parameters can be obtained in real-time through various sensors, the high dimensionality and significant noise interference make it difficult to directly extract dynamic trend features, leading to uncertainty in model input. Furthermore, the non-linear relationships between environmental parameters are complex, and traditional linear models struggle to effectively fit this complexity. Existing prediction models lack sufficient dynamic adjustment mechanisms when faced with sudden environmental changes, often leading to the accumulation of prediction bias. In particular, under extreme weather or other abnormal conditions, the output of the prediction model may deviate, and in severe cases, it may affect decisions such as irrigation and fertilization in grape growth management, thereby affecting the healthy growth and yield of crops.

[0004] In addition, there is a contradiction between real-time performance and computational complexity in the current model calibration and optimization process. High-precision model calibration often requires a large amount of computational resources, but agricultural management scenarios have high requirements for response speed. Balancing the precision of the model while balancing real-time performance and computational resource consumption to meet the dynamic changing needs of agricultural environmental monitoring is a major challenge for current technology. SUMMARY

[0005] The present application provides a grape growing environment monitoring method and system based on artificial intelligence, which combines a nonlinear relationship model, a dynamic calibration mechanism, and a sub-module optimization to efficiently and accurately monitor the grape growing environment. In a first aspect, the method comprises: Step S1: Obtain environmental data from sensors and construct a multi-dimensional environmental data matrix; standardize the multi-dimensional environmental data matrix, extract dynamic trend features, and obtain a standardized environmental feature matrix; Step S2: Construct a nonlinear relationship model based on the standardized environmental feature matrix; collect incremental data of the environmental data, and if any environmental data in the incremental data changes beyond a preset threshold, update the sub-module parameters of the nonlinear relationship model; Step S3: adjusting the weight distribution of the sub-modules in the nonlinear relationship model according to the updated sub-module parameters, and verifying the convergence of the weight distribution by calibrating the nonlinear relationship model; Step S4: predicting an environmental parameter change sequence according to the calibrated nonlinear relationship model; comparing the deviation of the predicted environmental parameter change sequence from the actual data, and if the deviation exceeds a preset tolerance threshold, re-optimizing the sub-module parameters; Step S5: smoothing the output of the re-optimized nonlinear relationship model to generate an environmental parameter prediction sequence.

[0006] As a preferred technical solution of the present application, in step S1, the environmental data is obtained from the sensor, and a multi-dimensional environmental data matrix is constructed, including: Real-time monitoring data is obtained from temperature sensors, humidity sensors, and light sensors; the real-time monitoring data is segmented by incremental data stream analysis; and the segmented data is format-verified. A multi-dimensional environmental data matrix containing temperature, humidity, and light is constructed according to the verified data, wherein the multi-dimensional environmental data matrix is dimensioned by time series, and integrates the real-time values of temperature, humidity, and light.

[0007] As a preferred technical solution of the present application, in step S1, the multi-dimensional environmental data matrix is standardized to extract dynamic trend features and obtain a standardized environmental feature matrix, including: The multi-dimensional environmental data matrix is normalized using a data standardization algorithm; the dynamic trend features of temperature, humidity, and light are separated by incremental data feature extraction; and a standardized environmental feature matrix is generated based on the dynamic trend features, wherein the standardized environmental feature matrix contains normalized values and change trends of temperature, humidity, and light.

[0008] As a preferred technical solution of the present application, in step S2, a nonlinear relationship model is constructed based on the standardized environmental feature matrix, including: The standardized environmental feature matrix is analyzed for interaction by a nonlinear regression function; a nonlinear relationship function of temperature and humidity is determined based on the multi-dimensional feature coupling of temperature change and humidity fluctuation; a light adjustment effect model is constructed based on the adjustment effect of light intensity on temperature and humidity; and the nonlinear relationship function and the light adjustment effect model are integrated to obtain a nonlinear relationship model reflecting the corresponding relationship between light and temperature and humidity.

[0009] As a preferred technical solution of the present application, in step S2, if it is detected that the environmental data change exceeds a preset threshold, the sub-module parameters of the nonlinear relationship model are updated, including: detecting an incremental data input; determining whether the change amplitude of the temperature, humidity or illumination exceeds a preset threshold value through a dynamic threshold value; if the preset threshold value is exceeded, decomposing the nonlinear relationship model into a temperature submodule, a humidity submodule and an illumination submodule through submodule segmentation technology; performing local weight updating on the temperature submodule, the humidity submodule and the illumination submodule to obtain updated submodule parameter configurations.

[0010] As a preferred technical solution of the present application, in step S3, adjusting the weight distribution of the submodules in the nonlinear relationship model according to the updated submodule parameters comprises: performing local feature sensitivity analysis according to the updated submodule parameter configurations; if the temperature change amplitude exceeds the preset threshold value, increasing the weight of the temperature submodule; adjusting the weight of the humidity submodule through dynamic trend modeling of the humidity change; adjusting the weight of the illumination submodule according to the illumination change trend; and generating an adjusted weight distribution according to the adjusted weights.

[0011] As a preferred technical solution of the present application, in step S3, verifying the convergence of the weight distribution by calibrating the nonlinear relationship model comprises: verifying whether the adjusted weight distribution satisfies a local convergence condition through submodule convergence detection; performing real-time calibration on the parameter isolation optimization of the temperature submodule, the humidity submodule and the illumination submodule through dynamic calibration step size; updating the nonlinear relationship model according to the calibration result to obtain a calibrated nonlinear relationship model, wherein the calibrated nonlinear relationship model satisfies a preset convergence condition.

[0012] As a preferred technical solution of the present application, performing smoothing processing on the output of the re-optimized nonlinear relationship model to generate an environment parameter prediction sequence comprises: performing smoothing processing on the output of the re-optimized nonlinear relationship model through data distribution fitting technology; generating a smoothed environment parameter prediction sequence, wherein the environment parameter prediction sequence contains predicted values of the temperature, humidity and illumination; transmitting the environment parameter prediction sequence to a management system for generating planting management suggestions.

[0013] In a second aspect, the present application further provides a grape growing environment monitoring system based on artificial intelligence, which is used to implement the above method, and comprises: a data processing module, configured to obtain environment data from a sensor, construct a multi-dimensional environment data matrix, perform standardization processing on the multi-dimensional environment data matrix, extract dynamic trend features, and obtain a standardized environment feature matrix; A model construction module is configured to construct a nonlinear relationship model based on the standardized environment feature matrix; and an incremental data of the environment data is collected, and if it is detected that any environment data in the incremental data changes by more than a preset threshold, a submodule parameter of the nonlinear relationship model is updated; A weight updating module is configured to adjust a weight distribution of a submodule in the nonlinear relationship model according to the updated submodule parameter, and verify the convergence of the weight distribution by calibrating the nonlinear relationship model; A model optimization module is configured to predict an environment parameter change sequence according to the calibrated nonlinear relationship model; compare a deviation between the predicted environment parameter change sequence and actual data, and if the deviation exceeds a preset tolerance threshold, re-optimize the submodule parameter; A smoothing processing module is configured to perform smoothing processing on an output of the re-optimized nonlinear relationship model to generate an environment parameter prediction sequence.

[0014] In a third aspect, the present application further provides a computer readable storage medium, which stores instructions, and the instructions are executed by a processor to implement the above method.

[0015] The technical scheme provided by the embodiments of the present application can have the following beneficial effects: The present application provides a grape growing environment monitoring system based on artificial intelligence, which can efficiently and accurately monitor and predict the changes of environmental parameters such as temperature, humidity and light. By real-time acquisition of environmental data through sensors and construction of a multi-dimensional environment data matrix, the key features of environmental parameters are effectively extracted by using standardization processing and dynamic trend feature extraction technology. By constructing a nonlinear relationship model and automatically updating the submodule parameters of the model when detecting that the environmental change exceeds the preset threshold, the system ensures real-time adaptation to the dynamic changes of the vineyard environment. In the model optimization process, the system adjusts the weight and calibrates the mechanism to ensure the convergence and accuracy of the model. When the deviation between the predicted value and the actual data exceeds the preset tolerance threshold, the system will automatically re-optimize the submodule parameters to improve the accuracy of the model. In order to eliminate the noise and abnormal fluctuations in the predicted data, the system uses smoothing processing technology to generate a more stable and continuous environment parameter prediction sequence, thereby providing reliable data support for the management system of the vineyard. Through this series of optimization and calibration process, the present application significantly improves the prediction accuracy and real-time response capability of the vineyard environment monitoring system, and can provide precise environmental management suggestions for grape planting, such as irrigation, shading, ventilation and other operations, to promote the healthy growth and high yield of grapes and improve the overall planting efficiency and benefit. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A flowchart of the grape growing environment monitoring method based on artificial intelligence of the present application; Figure 2 A structure diagram of a grape growth environment monitoring system based on artificial intelligence according to the present application. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be described in detail below in combination with the drawings and specific embodiments.

[0018] As Figure 1 In the present embodiment, the grape growth environment monitoring method based on artificial intelligence can specifically include: Step S1: obtaining environment data from a sensor to construct a multi-dimensional environment data matrix; performing standardization processing on the multi-dimensional environment data matrix to extract dynamic trend features to obtain a standardized environment feature matrix.

[0019] Among them, obtaining environment data from a sensor to construct a multi-dimensional environment data matrix includes: Obtaining real-time monitoring data from a temperature sensor, a humidity sensor and a light sensor; segmenting the real-time monitoring data through incremental data stream analysis; performing format verification on the segmented data; constructing a multi-dimensional environment data matrix containing temperature, humidity and light according to the verified data, wherein the multi-dimensional environment data matrix is dimensioned by time series, and integrates real-time values of the temperature, humidity and light.

[0020] Specifically, the system receives current ambient temperature, relative humidity percentage, and light intensity values ​​from distributed sensor nodes within the vineyard, along with temperature, humidity, and light sensors. This sensor data is synchronously acquired from the corresponding sensor nodes at a preset sampling frequency and includes information such as timestamps, sensor identifiers, data types, and measured values, forming raw data packets. To ensure efficient data processing and analysis, incremental data stream parsing technology is used to segment the data. Specifically, a sliding window mechanism is used to divide the continuously received data stream into time segments at fixed time intervals, ensuring sufficient data within each time window for subsequent analysis. Based on the boundary identifiers and sensor type identifiers of the raw data packets, the mixed data stream is separated into three sub-streams: temperature, humidity, and light intensity. The boundary identifiers are specific identifiers used to mark the boundaries of the data packets. Each sub-stream maintains independent time-series characteristics, avoiding cross-interference between different data types. Furthermore, each of the aforementioned sensor data points is validated. After the data format validation is passed, the system constructs a multi-dimensional environmental data matrix based on the validated data. The time dimension of this matrix is ​​row-indexed, with each row representing the complete environmental state at a specific point in time, including three types of environmental parameters: temperature, humidity, and light intensity. Each column corresponds to the variation trajectory of these parameters over the entire time series. This matrix ensures that the three types of environmental data at the same point in time are located in the same row by filling the verified environmental data into their corresponding positions in chronological order, guaranteeing data synchronization. For missing environmental data, the system uses nearest neighbor interpolation or boundary value truncation to fill in the gaps. This method effectively handles data loss or redundancy caused by intermittent sensor operation or communication interruptions. This technical solution can efficiently process various real-time monitoring data from the vineyard environment, ensuring data accuracy, completeness, and timeliness, thereby providing precise environmental monitoring data for grape growth and promoting optimized management of grape cultivation.

[0021] Furthermore, the multidimensional environmental data matrix is ​​standardized to extract dynamic trend features, resulting in a standardized environmental feature matrix, including: The multidimensional environmental data matrix is ​​normalized using a data standardization algorithm. The dynamic trend features of temperature, humidity, and light intensity are separated by incremental data feature extraction; A standardized environmental feature matrix is ​​generated based on the dynamic trend characteristics, wherein the standardized environmental feature matrix includes the normalized values ​​of temperature, humidity, and light intensity and their changing trends.

[0022] Specifically, the multi-dimensional environmental data matrix is processed by a min-max normalization method, i.e., a data standardization algorithm, to map the data ranges of environmental parameters such as temperature, humidity, and illumination to a unified interval [0, 1], ensuring that data of different dimensions are compared and analyzed on the same scale. Through incremental data feature extraction technology, the system captures the change trend of temperature, humidity, and illumination parameters, adopts a sliding time window mechanism to update data periodically and calculate the change rate between adjacent data points. Through a first-order difference method, the change value of temperature, humidity, and illumination within the window is calculated, and a moving average filter is used to smooth the change rate sequence to eliminate short-term noise. According to the statistical characteristics of the change rate, the trend pattern of each parameter is further identified, such as an upward, downward, or stable trend, to form stable dynamic trend features. The above dynamic trend features are integrated into a standardized environmental feature matrix, which is based on time series and contains normalized values and change trends of temperature, humidity, and illumination, reflecting the time sequence characteristics and change laws of each environmental parameter. Through correlation analysis, the system also calculates the correlation coefficient between environmental parameters and embeds it into the matrix to show the mutual relationship between parameters, such as the mutual influence of temperature and humidity, temperature and illumination. Through the above technical solutions, the standardized environmental feature matrix not only accurately records the change trend of each environmental parameter, but also reflects the dynamic correlation between them, providing data support for precise monitoring and prediction of grape growing environments.

[0023] Step S2: constructing a nonlinear relationship model based on the standardized environmental feature matrix; collecting incremental data of the environmental data, and if it is detected that any environmental data in the incremental data changes beyond a preset threshold, updating the sub-module parameters of the nonlinear relationship model; In step S2, the nonlinear relationship model is constructed based on the standardized environmental feature matrix, including: Through nonlinear regression function interaction analysis on the standardized environmental feature matrix, a nonlinear relationship function of temperature and humidity is determined based on the multi-dimensional feature coupling of temperature change on humidity fluctuation, and a light regulation effect model is constructed based on the regulation effect of light intensity on temperature and humidity. The nonlinear relationship function and the light regulation effect model are integrated to obtain a nonlinear relationship model reflecting the corresponding relationship between light and temperature and humidity.

[0024] Specifically, since the grape growing environment has an important influence on its growth and development, and environmental factors such as temperature, humidity, light, etc. have high time-varying and nonlinear characteristics, the traditional grape environment monitoring method often relies on a simplified model, often ignoring the complex interaction between these environmental factors. The above simplified method fails to fully consider the coupling effect between different environmental parameters, especially during the growth of grapes, the mutual adjustment between temperature, humidity and light has a significant influence on the growth and development of grapes, therefore, by analyzing the interaction of temperature, humidity and light and other environmental parameters through a nonlinear regression function, the complex relationship between them is revealed, the nonlinear regression method uses polynomial regression and radial basis function (RBF) to transform the data in the standardized environmental feature matrix, identifies the interaction pattern between environmental parameters, and can capture the nonlinear and complex interaction between environmental parameters in high-dimensional space, thereby providing more accurate data support for subsequent environmental prediction; wherein, in the process of constructing the nonlinear relationship between temperature and humidity, based on the above method of analyzing the temperature change rate and identifying the temperature rising, falling and stable change pattern using a sliding window method, the temperature change pattern identification matrix is obtained. Based on the temperature change pattern identification matrix, the amplitude, frequency and delay characteristics of humidity fluctuation are further analyzed to form the time sequence correlation between temperature and humidity. When the temperature rises, the humidity usually appears to lag behind. When the temperature drops, the humidity presents different response patterns. By using the multi-dimensional feature coupling method, the amplitude, rate and duration of temperature change are taken as input dimensions, and the fluctuation amplitude, frequency, phase delay of humidity are taken as output dimensions to construct a six-dimensional feature coupling space. The kernel function mapping maps the nonlinear relationship between temperature and humidity to a high-dimensional space, identifies the nonlinear coupling pattern between them, and fits the nonlinear relationship function between temperature and humidity by support vector regression method. The above nonlinear relationship function can describe the expected fluctuation range and trend of humidity under different temperature change conditions.

[0025] In the analysis of the nonlinear relationship between light and temperature and humidity, first, a grading system of light intensity is established, and the influence of different light intensities on temperature and humidity is analyzed; light intensity adjusts the changes of temperature and humidity by affecting plant transpiration and thermal radiation effect; by modeling the response relationship between temperature and humidity under different light conditions, a ternary nonlinear relationship model of the synergistic regulation effect of light on temperature and humidity is constructed; light intensity is used as the adjusting variable, and temperature and humidity are used as the adjusted variables to analyze the synergistic response characteristics of temperature and humidity to light changes; through the above ternary nonlinear relationship, the rapid response of light changes will affect the synergistic changes of temperature and humidity, and the above ternary nonlinear relationship can effectively describe the regulation effect of light on temperature and humidity; when the light intensity changes rapidly, the delay time and response amplitude of the synergistic response of temperature and humidity are calculated, and a light regulation effect model is established based on the above ternary nonlinear relationship; for example, the light regulation effect model adopts a 3-layer hidden layer structure, each layer containing 64 neurons, when the multilayer perceptron neural network is used to process the ternary nonlinear relationship between light and temperature and humidity; the current light intensity value, the temperature change sequence and the humidity change sequence in the past 1 hour are received by the input layer and are nonlinearly transformed by the ReLU activation function; during the network training process, it is found that when the light intensity is lower than the first preset threshold, the regulation effect is weak, and when it exceeds the second preset threshold, the regulation effect tends to be saturated, and this characteristic is effectively modeled by the nonlinear mapping capability of the neural network.

[0026] The nonlinear relationship function of temperature and humidity and the light regulation effect model are integrated to construct a comprehensive nonlinear relationship model containing the interaction of temperature, humidity and light, which can describe the complex correlation between multiple parameters in the vineyard environment, that is, by using the light intensity, temperature and humidity data in a preset time period before the current time to predict the light intensity, temperature and humidity prediction values in a future time period, the light intensity, temperature and humidity data all include corresponding time stamps, i.e. season, date and current time point.

[0027] Due to the complexity of the light regulation effect, it also reflects the asymmetry of its influence on temperature and humidity. When the light intensity increases, the temperature responds rapidly and the amplitude is large, while the humidity response is relatively lagging but the duration is longer. When the light intensity decreases, the temperature decreases slowly, and the humidity also shows a gradual recovery characteristic. The above technical solution can accurately predict the evolution trend of temperature and humidity under different light change conditions by establishing a light regulation effect model, and provides a scientific basis for precise environmental management of vineyards.

[0028] Further, in step S2, if it is detected that the environmental data changes exceed the preset threshold, the sub-module parameters of the nonlinear relationship model are updated, including: Collecting incremental data of the environmental data, judging whether the change range of temperature, humidity or illumination data in the incremental data exceeds a preset threshold value through a dynamic threshold value; if the preset threshold value is exceeded, the nonlinear relationship model is decomposed into a temperature submodule, a humidity submodule and an illumination submodule through submodule segmentation technology; and the temperature submodule, the humidity submodule and the illumination submodule are subjected to local weight updating to obtain updated submodule parameter configurations.

[0029] Specifically, in the grape environment monitoring, the changes of temperature, humidity and illumination and other factors have high time sequence and dynamics. The traditional environment monitoring system often ignores the instantaneousness and complexity of environmental changes, resulting in slow response of the model and reduced prediction accuracy when the environment mutates. This problem is particularly prominent in different stages of grape growth, such as spring, summer and autumn. The temperature and humidity change characteristics are different in each stage, and the sudden changes of sensor data may not be accurately reflected in the model in these change processes, for example, the dramatic fluctuations of temperature and the sudden changes of illumination intensity.

[0030] By detecting incremental data from temperature, humidity and illumination sensors, continuously receiving and comparing real-time data streams, identifying new data through timestamp comparison, and determining whether the change range of temperature, humidity or illumination exceeds the preset threshold value according to the dynamic threshold value judgment mechanism; for temperature, the dynamic threshold value is set according to the statistical distribution of historical data and adjusted according to different seasons to ensure more accurate response to temperature changes in different growth stages; the threshold value of humidity is determined by double standards, based on the relative humidity change rate and absolute humidity difference respectively, to ensure the monitoring sensitivity of humidity fluctuations in different growth stages; the threshold value of illumination intensity is set with different standards according to the illumination changes in different time periods of a day to further improve the response to illumination fluctuations.

[0031] When the change amount of temperature, humidity or illumination exceeds the corresponding threshold value, the above nonlinear relationship model is analyzed through submodule segmentation technology, specifically decomposing the original model into temperature submodule, humidity submodule and illumination submodule. Each submodule independently processes the corresponding environmental data, i.e. the above illumination, temperature and humidity data, avoiding interference between environmental factors. For each submodule, the neuron connection weight related to the environmental parameters, i.e. the above illumination, temperature and humidity, is extracted, and the contribution of each parameter to the model output is determined through gradient analysis; the temperature submodule receives temperature current value, historical value and change rate and other feature data, the humidity submodule processes humidity value, historical data and humidity and temperature correlation coefficient and other information, and the illumination submodule specifically processes the time sequence change characteristics of illumination intensity.

[0032] Through local weight updating, the network weights can be dynamically adjusted within each sub-module to improve prediction accuracy. The temperature sub-module updates weights through the Adam optimization algorithm, and the learning rate is dynamically adjusted according to the severity of temperature change. The humidity sub-module uses the RMSprop algorithm, focusing on the coupling relationship between humidity and temperature. The illumination sub-module is periodically optimized through the stochastic gradient descent algorithm with momentum according to the characteristics of the daily and seasonal cycles. After independent updating of each sub-module, the model parameters can be accurately adjusted to ensure higher prediction accuracy and real-time response capability when facing different environmental changes.

[0033] Through the above technical solution, the nonlinear relationship model can be updated in a timely manner according to the changes in the vineyard environment, thereby providing more accurate environmental prediction results and providing reliable data support for precise management of the vineyard. It helps to optimize the grape growing conditions and ensure the healthy growth and high yield of grapes.

[0034] Step S3: Adjust the weight distribution of the sub-modules in the nonlinear relationship model according to the updated sub-module parameters, and verify the convergence of the weight distribution by calibrating the nonlinear relationship model. In step S3, adjusting the weight distribution of the sub-modules in the nonlinear relationship model according to the updated sub-module parameters comprises: Performing local feature sensitivity analysis according to the updated sub-module parameter configuration. If the temperature change amplitude exceeds a preset threshold, increase the weight of the temperature sub-module. Model the dynamic trend of humidity change to adjust the weight of the humidity sub-module. Adjust the weight of the illumination sub-module according to the illumination change trend. Generate an adjusted weight distribution according to the adjusted weight.

[0035] Specifically, by sub-module segmentation technology, the nonlinear relationship model is decomposed into independent temperature, humidity and illumination sub-modules, each of which is independently optimized to avoid interference between parameters, thereby improving the calculation efficiency and prediction accuracy of the model. In cooperation with the above technical solution, and in order to further enable the above nonlinear relationship model to quickly respond and optimize the prediction result according to real-time environmental changes (such as a sharp rise in temperature, humidity change or light mutation), when the change amplitude of these environmental parameters is detected to exceed the preset threshold, the weight distribution of the sub-model in the nonlinear relationship model is adjusted through the parameters of the sub-modules, local feature sensitivity analysis is performed based on the sub-module parameter configuration, the partial derivative of the temperature, humidity and illumination sub-modules with respect to the change of the environmental feature matrix is calculated by extracting the parameter vector of the temperature, humidity and illumination sub-modules, and then the parameter sensitivity matrix is obtained. Further, the feature sensitivity index of each sub-module is calculated by weighted summation. The weight adjustment of the temperature sub-module is proportional to the temperature change amplitude and its feature sensitivity index, the weight of the humidity sub-module is adjusted by dynamic trend modeling, and the weight of the light sub-module is adjusted according to the light change gradient.

[0036] In the temperature sub-module, when the temperature change amplitude is detected to exceed the preset threshold, the weight of the sub-module is proportionally increased according to the feature sensitivity index; to ensure that the influence of temperature on the model output is reasonably strengthened. Similarly, the adjustment of the humidity sub-module weight depends on the dynamic change trend of humidity, the humidity change sequence is extracted by the sliding window method, and the trend fitting is performed by the exponential smoothing method. When the humidity change trend shows an upward or downward trend, the system adjusts the weight of the humidity sub-module according to the fitting accuracy; the adjustment of the light sub-module is based on the gradient of the change of the light intensity, and when the change gradient exceeds the preset threshold, the weight of the light sub-module is adjusted according to its feature sensitivity index. Finally, the weights of the adjusted temperature, humidity and light sub-modules are normalized to ensure that their sum is 1, and the final weight distribution is obtained.

[0037] The above technical solution solves the problem in the prior art that the model parameters cannot be adjusted in time when facing complex and dynamically changing vineyard environments, so as to automatically adjust the importance of environmental parameters such as temperature, humidity and light according to real-time changes, thereby optimizing the response speed and prediction accuracy of the environmental monitoring system and improving the environmental management efficiency and accuracy in the grape planting process.

[0038] Further, in step S3, the convergence of the weight distribution is verified by calibrating the nonlinear relationship model, including: whether the adjusted weight distribution meets the local convergence condition is verified by sub-module convergence detection; The parameters of the temperature submodule, humidity submodule, and illumination submodule are calibrated in real time through dynamic calibration step size optimization. The nonlinear relationship model is updated based on the calibration results to obtain a calibrated nonlinear relationship model, wherein the calibrated nonlinear relationship model satisfies a preset convergence condition.

[0039] Specifically, the adjusted weight distribution is verified to satisfy the local convergence condition through submodule convergence detection, the weight change gradient of the temperature, humidity, and illumination submodules in the continuous iteration process is calculated, and the convergence index of each submodule is obtained through gradient norm calculation. If the convergence detection results of all submodules are lower than the preset local convergence threshold, it is determined that the weight distribution satisfies the local convergence condition. Otherwise, it does not satisfy. To ensure the efficient optimization and precision of the above nonlinear relationship model, a dynamic calibration step size mechanism is introduced to adjust the step size in real time according to the convergence state of each submodule and the amplitude of environmental parameter changes. For submodules that have not converged, a larger calibration step size is used to speed up the convergence process, and for submodules that have converged, a smaller calibration step size is used to maintain their stability. By constructing independent optimization spaces for the temperature, humidity, and illumination submodules, each submodule is optimized independently. The temperature submodule extracts temperature-related neural network layer weights, the humidity submodule is based on its dynamic change trend, and the illumination submodule considers the periodic changes of sunlight intensity, thereby being updated independently to avoid mutual interference between submodules.

[0040] In addition, the calculation of the dynamic calibration step size reflects the precise adaptation to the change characteristics of different environmental parameters. For example, when the temperature submodule detects a temperature rising trend at three consecutive time points, the step size will increase to speed up the learning of temperature changes. The humidity submodule adjusts the step size in a timely manner to better capture the reverse change law of humidity according to the negative correlation between humidity and temperature. The illumination submodule adjusts the step size at different times according to the change of sunlight time to ensure that the model can accurately predict the dynamic changes of illumination.

[0041] The parameter synchronization mechanism ensures the consistency of the optimized parameters of each submodule in the overall model. When the parameters of a submodule are updated, the system checks its coupling relationship with other submodules and fine-tunes if necessary to ensure the stability of the overall model. Finally, the updated nonlinear relationship model integrates the optimization parameters of each submodule through a parameter fusion algorithm and verifies its prediction accuracy through a cross-validation mechanism. When the prediction error of the model on the test set is lower than the preset standard, it is confirmed that the calibration process is complete and the model meets the expected convergence condition.

[0042] The technical scheme solves the problem that the traditional model cannot respond and accurately adjust in time when the environment changes dramatically in vineyard environment monitoring, ensures that the model can adaptively optimize the weight distribution of each environment parameter, improves the prediction ability and real-time performance of the system to dynamic environment changes, and further supports accurate decision-making and effective management in grape planting management.

[0043] Step S4: predicting an environment parameter change sequence according to the calibrated nonlinear relationship model; comparing the deviation of the predicted environment parameter change sequence and actual data, and re-optimizing the submodule parameters if the deviation exceeds a preset tolerance threshold.

[0044] Specifically, the light, temperature and humidity data in the current preset time period are input into the calibrated nonlinear relationship model to predict the above-mentioned environment parameter change sequence in the future time period. The environment parameters are the light, temperature and humidity in the grape growing environment. Due to the complexity of environmental changes, there may be a certain deviation between the prediction result and the actual observation data. Therefore, by comparing the difference between the predicted value and the actual data, if it is found that the deviation exceeds the preset tolerance threshold, it indicates that the prediction accuracy of the model does not meet the expectation, and the above-mentioned re-optimization of the parameters of the submodule needs to be repeated to ensure that the model can more accurately predict the change of the future environment parameters.

[0045] Step S5: smoothing the output of the re-optimized nonlinear relationship model to generate an environment parameter prediction sequence, specifically including: The output of the re-optimized nonlinear relationship model is smoothed by a data distribution fitting technology to generate a smoothed environment parameter prediction sequence, wherein the environment parameter prediction sequence contains the predicted values of the temperature, humidity and light; and the environment parameter prediction sequence is transmitted to a management system for generating planting management suggestions.

[0046] Specifically, to improve the accuracy and reliability of temperature, humidity, and light prediction values, the output of the re-optimized nonlinear relationship model is smoothed to generate an environmental parameter prediction sequence, which in turn provides accurate decision-making basis for the vineyard management system. First, the output of the re-optimized model is smoothed by data distribution fitting technology to eliminate local mutations and noise interference that may occur during model calculation. This smoothing process includes three main steps: preliminary smoothing, secondary fitting, and continuous processing. Specifically, the moving average fitting method is used to preliminarily smooth the original prediction data. By setting a sliding window length (e.g., 5 time points), the arithmetic mean of the data points within the window is calculated to obtain the smoothed value at each time point. Then, the preliminarily smoothed data is fitted again using a Gaussian kernel function, which uses Gaussian weights for weighted averaging to further eliminate minor fluctuations and maintain the overall trend of the data. The standard deviation parameter of the Gaussian kernel function is differentiated according to the characteristics of each environmental parameter to ensure optimal smoothing effect for different environmental parameters such as temperature, humidity, and light. Finally, the smoothed discrete data is continuously processed using a cubic spline interpolation method to generate a smooth continuous curve, ensuring smooth transition between adjacent prediction points in the prediction sequence and avoiding abrupt changes.

[0047] The smoothed temperature, humidity, and light prediction values are reorganized into a time sequence and subjected to consistency verification to ensure that the physical relationship between these environmental parameters is reasonable. For example, when the temperature rises sharply, the predicted value of humidity should decrease accordingly to maintain the coordination between environmental parameters. After ensuring that these parameters conform to the actual rules, the generated environmental parameter prediction sequence is transmitted to the management system for generating planting management recommendations, including irrigation, shading, ventilation, and other operations.

[0048] Through the above technical solution, the vineyard growth environment monitoring system can obtain smoothed and accurate environmental parameter prediction sequences to provide scientific decision support for grape growth. In addition, the adaptive smoothing mechanism can dynamically adjust the smoothing strategy according to the characteristics of the original prediction data. For example, when the temperature changes sharply, the length of the moving average window is increased to obtain a stronger smoothing effect. The smoothing of humidity and light is set with different Gaussian kernel standard deviations and interpolation strategies according to their respective change characteristics to ensure that each environmental parameter receives the most suitable smoothing effect. This efficient and accurate smoothing process can maximize the stability and reliability of the model prediction results, thereby providing efficient data support for vineyard environmental management.

[0049] The present application also provides a grape growth environment monitoring system based on artificial intelligence for implementing the above method, as shown in Figure 2 The system includes: The data processing module is configured to acquire environment data from the sensor, construct a multi-dimensional environment data matrix, perform standardization processing on the multi-dimensional environment data matrix, extract a dynamic trend feature, and obtain a standardized environment feature matrix. The model construction module is configured to construct a nonlinear relationship model based on the standardized environment feature matrix, collect incremental data of the environment data, and update a submodule parameter of the nonlinear relationship model if it is detected that any environment data in the incremental data changes by more than a preset threshold. The weight updating module is configured to adjust a weight distribution of a submodule in the nonlinear relationship model according to the updated submodule parameter, and verify the convergence of the weight distribution by calibrating the nonlinear relationship model. The model optimization module is configured to predict an environment parameter change sequence according to the calibrated nonlinear relationship model, compare a deviation between the predicted environment parameter change sequence and actual data, and re-optimize a submodule parameter if the deviation exceeds a preset tolerance threshold. The smoothing processing module is configured to perform smoothing processing on an output of the re-optimized nonlinear relationship model, and generate an environment parameter prediction sequence.

[0050] The application further provides a computer readable storage medium, which stores instructions, and the instructions are executed by a processor to implement the method.

[0051] In summary, the present application collects real-time data from temperature, humidity and light sensors, and processes the data through incremental data stream analysis techniques. Then, according to the real-time changes of the data, the model determines whether to exceed the preset threshold through a dynamic threshold judgment mechanism. If it exceeds, the model is updated. Specifically, the temperature, humidity and light data are processed separately using sub-module segmentation techniques to ensure independent optimization of each sub-module and avoid mutual interference. Within each sub-module, weight updates are adjusted according to real-time environmental data to ensure the model's rapid response to environmental changes and accurate prediction. Through local feature sensitivity analysis, the weights of each sub-module are dynamically adjusted according to changes in temperature, humidity and light to optimize the accuracy and stability of the model. For cases where environmental data fluctuates greatly, a dynamic calibration step adjustment strategy is used to speed up the convergence process, and an isolation optimization method is used to ensure that each sub-module can focus on optimizing its own parameters, avoiding interference between parameters. All optimized sub-module parameters are integrated into a nonlinear relationship model to obtain the calibrated model. Through smoothing techniques, the system can eliminate noise and mutations in the predicted values, making the predicted sequences of temperature, humidity and light smooth and reasonable, ensuring that the final generated environmental parameter prediction sequence conforms to the actual law and can provide reliable planting recommendations for vineyard management systems. Through the mutual cooperation between the above steps, the system can quickly and accurately adjust the weight distribution of each environmental parameter when facing different environmental changes, thereby achieving accurate prediction and intelligent management of vineyard environmental changes, providing effective environmental control strategies for grape growers, and ultimately promoting the optimization and high yield of grape growth.

[0052] The above is only a preferred embodiment of the present application, but the protection scope of the present application is not limited thereto. Any skilled person in the art can make equivalent replacements or changes to the technical solutions and inventive concepts of the present application within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A method of monitoring a grape growing environment based on artificial intelligence, characterized by, Comprise: Step S1: obtaining environment data from sensors, constructing a multi-dimensional environment data matrix; standardizing the multi-dimensional environment data matrix, extracting dynamic trend characteristics, obtaining a standardized environment feature matrix, the environment data including temperature, humidity and illumination data; Step S2: based on the standardized environment feature matrix, a nonlinear relationship model is constructed; incremental data of the environment data is collected, if it is detected that any environment data change in the incremental data exceeds a preset threshold, the sub-module parameters of the nonlinear relationship model are updated; Step S3: according to the updated sub-module parameters, the weight distribution of the sub-modules in the nonlinear relationship model is adjusted, and the nonlinear relationship model is calibrated to verify the convergence of the weight distribution; Step S4: according to the calibrated nonlinear relationship model, the environment parameter change sequence is predicted; the deviation between the predicted environment parameter change sequence and the actual data is compared, if the deviation exceeds the preset tolerance threshold, the sub-module parameters are re-optimized; Step S5: the output of the re-optimized nonlinear relationship model is smoothed to generate an environment parameter prediction sequence.

2. The method of claim 1, wherein, In step S1, the environment data is obtained from the sensors, and a multi-dimensional environment data matrix is constructed, comprising: Real-time monitoring data is obtained from temperature sensors, humidity sensors and illumination sensors; the real-time monitoring data is segmented by incremental data stream analysis; the segmented data is format verified; A multi-dimensional environment data matrix containing temperature, humidity and illumination is constructed according to the verified data, wherein the multi-dimensional environment data matrix is dimensioned by time sequence, and the real-time values of temperature, humidity and illumination are integrated.

3. The method of claim 1, wherein, In step S1, the multi-dimensional environment data matrix is standardized to extract dynamic trend characteristics to obtain a standardized environment feature matrix, comprising: The multi-dimensional environment data matrix is normalized by a data standardization algorithm; the dynamic trend characteristics of temperature, humidity and illumination are separated by incremental data feature extraction; a standardized environment feature matrix is generated according to the dynamic trend characteristics, wherein the standardized environment feature matrix contains the normalized values and change trends of temperature, humidity and illumination.

4. The method of claim 1, wherein, In step S2, a nonlinear relationship model is constructed based on the standardized environment feature matrix, comprising: The standardized environment feature matrix is analyzed by a nonlinear regression function; based on the multi-dimensional feature coupling of temperature change to humidity fluctuation, the nonlinear relationship function of temperature and humidity is determined; based on the adjusting effect of illumination intensity on temperature and humidity, an illumination adjusting effect model is constructed; the nonlinear relationship function and the illumination adjusting effect model are integrated to obtain a nonlinear relationship model reflecting the corresponding relationship between illumination and temperature and humidity.

5. The method of claim 4, wherein, In step S2, if it is detected that the environment data change exceeds the preset threshold, the sub-module parameters of the nonlinear relationship model are updated, comprising: detecting an incremental data input; determining whether the change amplitude of the temperature, humidity or illumination exceeds a preset threshold value through a dynamic threshold value; if the change amplitude exceeds the preset threshold value, decomposing the nonlinear relationship model into a temperature submodule, a humidity submodule and an illumination submodule through submodule segmentation technology; performing local weight updating on the temperature submodule, the humidity submodule and the illumination submodule to obtain updated submodule parameter configurations.

6. The method of claim 1, wherein, In step S3, the weight distribution of the submodules in the nonlinear relationship model is adjusted according to the updated submodule parameter configurations, including: performing local feature sensitivity analysis according to the updated submodule parameter configurations; if the temperature change amplitude exceeds the preset threshold value, increasing the weight of the temperature submodule; adjusting the weight of the humidity submodule through dynamic trend modeling of the humidity change; adjusting the weight of the illumination submodule according to the illumination change trend; and generating an adjusted weight distribution according to the adjusted weights.

7. The method of claim 6, wherein, In step S3, the nonlinear relationship model is calibrated to verify the convergence of the weight distribution, including: verifying whether the adjusted weight distribution meets a local convergence condition through submodule convergence detection; and performing real-time calibration on the parameter isolated optimization of the temperature submodule, the humidity submodule and the illumination submodule through a dynamic calibration step; updating the nonlinear relationship model according to the calibration result to obtain a calibrated nonlinear relationship model, wherein the calibrated nonlinear relationship model meets a preset convergence condition.

8. The method of claim 1, wherein, performing smoothing processing on the output of the re-optimized nonlinear relationship model to generate an environment parameter prediction sequence, including: performing smoothing processing on the output of the re-optimized nonlinear relationship model through data distribution fitting technology; generating a smoothed environment parameter prediction sequence, wherein the environment parameter prediction sequence contains predicted values of the temperature, humidity and illumination; transmitting the environment parameter prediction sequence to a management system for generating planting management suggestions.

9. An artificial intelligence-based grape growing environment monitoring system for implementing the method of any one of claims 1-8, characterized by, The system includes: a data processing module configured to acquire environment data from sensors, construct a multi-dimensional environment data matrix, perform standardization processing on the multi-dimensional environment data matrix, extract dynamic trend features, and obtain a standardized environment feature matrix; a model construction module configured to construct a nonlinear relationship model based on the standardized environment feature matrix, and collect incremental data of the environment data, wherein if it is detected that any environment data in the incremental data changes by more than a preset threshold value, the submodule parameters of the nonlinear relationship model are updated; a weight updating module configured to adjust the weight distribution of the submodules in the nonlinear relationship model according to the updated submodule parameters, and verify the convergence of the weight distribution by calibrating the nonlinear relationship model; a model optimization module configured to predict an environment parameter change sequence according to the calibrated nonlinear relationship model, compare the deviation between the predicted environment parameter change sequence and actual data, and re-optimize the submodule parameters if the deviation exceeds a preset tolerance threshold; a smoothing processing module configured to perform smoothing processing on the output of the re-optimized nonlinear relationship model to generate an environment parameter prediction sequence.

10. A computer-readable storage medium having stored thereon instructions, the instructions comprising, The instructions, when executed by a processor, implement the method of any one of claims 1-8.

Images