Greenhouse temperature data monitoring method and system based on data analysis

Abstract

The invention discloses a greenhouse temperature data monitoring method and system based on data analysis. The method comprises the following steps: S1, collecting and constructing an environment data set by using a multi-point sensor array arranged in a greenhouse; s2, obtaining a preprocessed environment data set; s3, forming an initial partition structure; s4, when it is monitored that the environmental data change exceeds a threshold value, dynamically adjusting partition boundaries and the number of regions to obtain a self-adaptive partition result; s5, performing feature extraction on the environment data of each partition in the self-adaptive partition result to form a feature set for model training; s6, obtaining future temperature prediction results of different partitions in the greenhouse; and S7, generating greenhouse environment regulation and control suggestions in combination with future temperature prediction results of different partitions, and carrying out auxiliary regulation on a ventilation system, a heating system and a shading device. The problems of insufficient precision and poor adaptability in greenhouse temperature monitoring and regulation in the prior art are solved, and efficient technical support is provided for intelligent agriculture development.

Description

Technical Field

[0001] The present invention relates to the technical field of greenhouse temperature data monitoring, and in particular to a greenhouse temperature data monitoring method and system based on data analysis. Background Art

[0002] With the development of modern agricultural technology, greenhouse cultivation has gradually become one of the important means to improve crop yield and quality. In the greenhouse environment, temperature, as a key environmental factor affecting crop growth, needs to be accurately monitored and regulated. However, the temperature of the greenhouse microenvironment has significant dynamics and spatial differences. Traditional temperature monitoring and control technologies are difficult to meet the current needs of smart agricultural development.

[0003] At present, greenhouse temperature monitoring mainly relies on multi-point sensor arrays to collect environmental data and process them uniformly through a central control system. Some traditional methods use simple average value calculations to analyze the overall temperature in the greenhouse. Although this can reflect the overall trend of the greenhouse, it ignores the spatial temperature difference and local microenvironment changes inside the greenhouse, making it difficult to meet the needs of refined regulation. In addition, some improved methods have introduced the idea of ​​regional division, dividing the greenhouse space into several fixed areas, and realizing regional-level monitoring through regional mean calculation. However, the static nature of regional division cannot adapt to the dynamic changes of the greenhouse environment, resulting in insufficient monitoring accuracy.

[0004] In recent years, with the rapid development of artificial intelligence technology, machine learning algorithms have begun to be applied to greenhouse temperature prediction, including models based on simple linear regression or support vector machines. However, existing algorithms usually assume that temperature changes have a linear or simple nonlinear relationship, while greenhouse temperature changes are actually affected by a variety of complex factors such as light, humidity, and wind speed. Traditional algorithms are often inefficient in dealing with these high-dimensional nonlinear relationships and it is difficult to achieve high-precision predictions. At the same time, existing technologies also have deficiencies in the data processing link. For example, the processing of missing data and abnormal data is mostly simple interpolation or elimination, and the correlation between these data and the dynamic environmental characteristics of the greenhouse is not fully considered, resulting in insufficient data quality, which affects the training effect of the model.

[0005] In addition, the existing technology usually adopts preset rules or fixed thresholds in the generation of greenhouse environment control suggestions, lacks in-depth analysis and refined prediction of future temperature change trends, resulting in lag and energy waste in practical application of control measures. Overall, the existing technology has significant deficiencies in monitoring accuracy, data processing efficiency, prediction accuracy and intelligent control suggestions. An innovative method that can dynamically adapt to the characteristics of the greenhouse microenvironment and make full use of high-dimensional data characteristics is urgently needed to solve the above problems. Summary of the invention

[0006] One purpose of the present invention is to propose a greenhouse temperature data monitoring method and system based on data analysis. The present invention solves the problems of insufficient accuracy and poor adaptability of the prior art in greenhouse temperature monitoring and control, and provides efficient technical support for the development of smart agriculture.

[0007] A greenhouse temperature data monitoring method based on data analysis according to an embodiment of the present invention comprises the following steps: S1. Use a multi-point sensor array arranged in the greenhouse to collect and construct an environmental data set; S2. Perform missing value filling, outlier detection and normalization processing on the environmental data set to obtain a preprocessed environmental data set; S3. Perform multimodal regional division of the greenhouse space according to the spatial location characteristics and environmental parameter characteristics contained in the preprocessed environmental data set, and form an initial partition structure; S4. Evaluate the initial partition structure based on the real-time changing environmental data in the greenhouse. When the monitored environmental data changes exceed the threshold, dynamically adjust the partition boundaries and the number of regions to obtain an adaptive partition result. S5. extracting features from the environmental data of each partition in the adaptive partitioning result, combining the time series features, spatial distribution features, and correlation features between environmental parameters to form a feature set for model training; S6. Divide the feature set into samples, use the gradient boosting tree algorithm to build a temperature prediction model on each adaptive partition, use the temperature prediction model to input and infer the environmental data of each adaptive partition, and obtain the future temperature prediction results of different partitions in the greenhouse; S7. Generate greenhouse environment control suggestions based on the future temperature prediction results of different partitions, and perform auxiliary adjustments on the ventilation system, heating system and shading device. After the adjustment is completed, continue to collect new environmental data and return to step S1 to step S6 for loop execution, forming a dynamic closed-loop optimization process.

[0008] Optionally, the S1 includes the following steps: S11. Arrange a multi-point sensor array in the greenhouse. The environmental parameters collected by each sensor node include temperature, humidity, light intensity and wind speed. The layout of the sensor array is arranged according to the greenhouse space structure and environmental gradient changes. S12. Collect data from the multi-point sensor nodes and construct an environmental data set D with time t and spatial position as indexes: }; in, represents the temperature data collected by the i-th sensor node, represents the humidity data collected by the i-th sensor node, represents the light intensity data collected by the i-th sensor node, represents the wind speed data collected by the i-th sensor node, ( ) represents the spatial position coordinates of the i-th sensor node, represents the timestamp of the collected data, and N represents the total number of sensor nodes in the greenhouse; S13. Perform a preliminary integrity check on the collected environmental data to ensure that all environmental parameters are included in each node data set. , and mark missing or abnormal data as unavailable points, recorded as set E: ; in,( ) is the spatial position coordinate of the unavailable point, The timestamp of the unavailable point; S14. Eliminate the unusable point data marked in step S13 to form a complete environmental data set : .

[0009] Optionally, S2 includes the following steps: S21. Complete environment dataset Perform analysis and fill in missing environmental parameter values ​​based on interpolation if any; S22. Perform outlier detection on the filled environmental data set, use the triple standard deviation method to detect the environmental parameter values, and use the weighted average method of adjacent nodes to correct the detected outlier set O; S23. Normalize the corrected environmental data set, map the environmental parameter values ​​to the interval [0,1], and record the normalized environmental data set .

[0010] Optionally, S3 includes the following steps: S31. From the preprocessed environmental dataset The spatial location features and environmental parameter features of each sensor node are extracted to form a multimodal feature vector : ; S32. Based on feature vector The spatial location features in the sensor nodes are used to calculate the spatial distance between the sensor nodes. : ; in, is the Euclidean distance between the i-th node and the j-th node; S33. Combined spatial distance and the Euclidean distance between the environmental parameter features , calculate the comprehensive similarity between sensor nodes : ; in, and are the weight coefficients of spatial distance and environmental parameter distance respectively, Represents the environmental parameter distance between the i-th and j-th nodes: ; S34. Based on comprehensive similarity The K-means algorithm is used to cluster the sensor nodes and generate K initial regions. The region division results are recorded as: ; in, represents the set of sensor nodes in the kth region, satisfying ; S35. Verify the generated initial area, calculate the mean value of the environmental parameters and the spatial compactness of the sensor nodes in each area, make the division result conform to the actual environmental characteristics in the greenhouse, and record the verified initial partition structure as the input area division result.

[0011] Optionally, S4 includes the following steps: S41. Continuously collect real-time environmental data sets using a sensor array deployed in the greenhouse And associate it with the initial partition structure R to form a real-time regional dataset: ; S42. Compute real-time regional datasets Changes in environmental parameters : ; in, Indicates the environmental parameter values ​​collected in real time. represents the historical mean of the sensor node in the initial partition structure, is the number of sensor nodes in the kth region; S43. Compare the environmental parameter change with the preset threshold Compare and mark the kth region as the region to be adjusted when the following conditions are met: ; in, The permissible range of greenhouse environmental parameters is set dynamically based on historical data; S44. Recalculate the comprehensive similarity for the marked areas that need adjustment , and re-divide the regional affiliation of sensor nodes based on the K-means algorithm to generate updated regional division results ; S45. Calculate the number of sensor nodes in each area after update and the mean of regional environmental parameters, if the number of nodes in a certain area is lower than the set threshold , then merge the area into the adjacent area. If the environmental parameter change of a certain area is greater than the preset value, refine the partition, add new partition boundaries, and adjust the optimized regional division results. Recorded as adaptive partitioning results.

[0012] Optionally, S5 includes the following steps: S51. Results for adaptive partitioning Each area , extract the environmental parameter change trend of each sensor node within the time window, and calculate the time series feature vector : ; in, Respectively represent the i-th sensor node in the time window The rate of change of interior temperature, humidity, light intensity, and wind speed; S52. Based on adaptive partition results Calculate the spatial distribution characteristics of sensor nodes in each partition, including the average distance between nodes and the standard deviation of the node distribution : ; ; in, represents the average distance from the i-th sensor node to other nodes; S53. For each partition Calculate the correlation coefficient matrix between various environmental parameters in the area , the elements in the correlation coefficient matrix Defined as: ; in, For parameters and The covariance of and The parameters are and The standard deviation of S54. Combine the time series features, spatial distribution features and environmental parameter association features extracted in steps S51, S52 and S53 to form a complete feature set : ; in, is the comprehensive feature vector of the kth partition.

[0013] Optionally, the S6 includes the following steps: S61. Feature Set Multidimensional data enhancement is performed before sample division, and enhanced samples are constructed by expanding the time dimension and adding spatial gradient information. : ; in, Respectively represent partitions The temperature gradient along the x, y, z spatial axes at time t is defined as: ; ; ; S62. Based on enhanced samples Divide into training set and test set according to the time window, and use time and space gradient information as new input features to form a comprehensive sample set and ; S63. Initialize the temperature prediction model for each partition and construct the loss function Defined as: ; in, is the total loss of the temperature prediction model, Represents the model prediction value With actual value The error between is the neighborhood weight, is a coordination factor used to balance local errors and global consistency, is the number of training samples in the partition; S64. In the process of temperature prediction model training, the enhanced samples of S61 are combined to dynamically evaluate the contribution of the newly added temporal and spatial gradient features, and calculate the feature importance score I(f): ; in, is the dth feature in the feature set, and D is the number of dimensions of the feature set; Optimize the modeling capability of the temperature prediction model for key features of the greenhouse microenvironment by dynamically eliminating low-contribution features; S65. Input real-time environmental data to the trained temperature prediction model , combining the temporal and spatial gradient characteristics to predict the future temperature change trend of the sub-region: ; At the same time, the uncertainty of the prediction results is quantified to generate the prediction interval ; S66. Prediction results for all partitions Integrate to form the future temperature forecast results of the greenhouse as a whole At the same time, combined with the uncertainty quantification results, the partition triggers enhanced sample retraining when the prediction error is greater than the preset value, forming a closed-loop optimization process.

[0014] A greenhouse temperature data monitoring system based on data analysis includes the following modules: The data acquisition module is a multi-point sensor array arranged in the greenhouse, which is used to collect environmental data in real time. The environmental data includes temperature, humidity, light intensity and wind speed. The layout of the sensor array is determined according to the spatial structure of the greenhouse and the characteristics of environmental gradient changes; The data processing module is connected to the data collection module and is used to fill missing values, detect outliers and normalize the collected environmental data, and output the preprocessed environmental data set; The regional division module is used to perform multimodal regional division of the greenhouse space according to the preprocessed data set output by the data processing module, generate the initial partition structure based on the spatial location characteristics and environmental parameter characteristics of the sensor, and dynamically adjust the partition boundaries and the number of regions to generate adaptive partition results; The feature extraction module is used to extract the time series features, spatial distribution features and correlation features between environmental parameters of each partition from the adaptive partition results, and generate a feature set by combining the time and space gradient information to provide input data for model training; The model training module is used to divide the training set and the test set according to the time window based on the feature set generated by the feature extraction module, and use the improved gradient boosting tree algorithm to build the temperature prediction model on each adaptive partition, and optimize the model performance according to the importance of dynamic features; The prediction analysis module is connected to the model training module and is used to input real-time environmental data, combine the temperature prediction model of greenhouse partitions, generate the prediction results of future temperature change trends of each partition, and generate the prediction interval through uncertainty quantification analysis; The control suggestion module integrates the temperature change trends of each partition based on the future temperature prediction results generated by the prediction and analysis module, and generates greenhouse environment control suggestions based on the uncertainty analysis results, including adjustment plans for the ventilation system, heating system and shading device; The feedback optimization module connects the data collection module, model training module and control suggestion module. It is used to continuously collect new environmental data and monitor the control effects. It triggers feature resampling and model retraining based on real-time data to form a dynamic closed-loop optimization mechanism.

[0015] The beneficial effects of the present invention are: (1) The present invention realizes accurate characterization of temperature distribution in complex greenhouse space by collecting environmental data from multiple sensors in real time in the greenhouse and adjusting the partition boundaries and the number of regions according to the dynamic changes of environmental parameters. It can dynamically optimize the regional division according to the real-time changing greenhouse environmental data, effectively avoiding the regional monitoring distortion problem caused by static partitioning, and enabling the monitoring system to better adapt to the dynamic changes of the greenhouse environment, thereby realizing high-precision temperature monitoring in greenhouses with significant local temperature differences and complex microenvironments.

[0016] (2) The present invention improves the traditional gradient boosting tree algorithm in greenhouse temperature prediction. By introducing time series features and spatial gradient features, the algorithm's modeling ability for multidimensional nonlinear relationships is significantly improved. Combined with a multi-level optimization framework, the improved gradient boosting tree algorithm can not only effectively capture the complex spatiotemporal characteristics of temperature changes in the greenhouse, but also dynamically adjust the importance of features, so that the model can focus on key environmental parameters and eliminate redundant information. By introducing a global coordination term, the prediction consistency inside and outside the greenhouse partition is optimized, effectively alleviating the problem of local overfitting in greenhouse temperature prediction.

[0017] (3) The present invention realizes the intelligent linkage of the entire chain from data collection, model training, temperature prediction to the generation of control suggestions by constructing a dynamic closed-loop optimization process. In the greenhouse temperature control link, targeted control suggestions are generated in combination with real-time prediction results, and the partition model and prediction model are dynamically optimized according to the new data collected after the control, thereby forming a feedback closed loop of the control effect. Compared with the traditional control method based on fixed rules, the present invention has significantly improved the response speed and control accuracy, can effectively avoid energy waste, and at the same time ensure that crops grow at a suitable temperature. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 The present invention provides a flow chart of a greenhouse temperature data monitoring method and system based on data analysis. DETAILED DESCRIPTION

[0019] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0020] refer to Figure 1 , a greenhouse temperature data monitoring method based on data analysis, comprising the following steps: S1. Use a multi-point sensor array arranged in the greenhouse to collect and construct an environmental data set; S2. Perform missing value filling, outlier detection and normalization processing on the environmental data set to obtain a preprocessed environmental data set; S3. Perform multimodal regional division of the greenhouse space according to the spatial location characteristics and environmental parameter characteristics contained in the preprocessed environmental data set, and form an initial partition structure; S4. Evaluate the initial partition structure based on the real-time changing environmental data in the greenhouse. When the monitored environmental data changes exceed the threshold, dynamically adjust the partition boundaries and the number of regions to obtain an adaptive partition result. S5. extracting features from the environmental data of each partition in the adaptive partitioning result, combining the time series features, spatial distribution features, and correlation features between environmental parameters to form a feature set for model training; S6. Divide the feature set into samples, use the gradient boosting tree algorithm to build a temperature prediction model on each adaptive partition, use the temperature prediction model to input and infer the environmental data of each adaptive partition, and obtain the future temperature prediction results of different partitions in the greenhouse; S7. Generate greenhouse environment control suggestions based on the future temperature prediction results of different partitions, and perform auxiliary adjustments on the ventilation system, heating system and shading device. After the adjustment is completed, continue to collect new environmental data and return to step S1 to step S6 for loop execution, forming a dynamic closed-loop optimization process.

[0021] In this implementation, S1 includes the following steps: S11. Arrange a multi-point sensor array in the greenhouse. The environmental parameters collected by each sensor node include temperature, humidity, light intensity and wind speed. The layout of the sensor array is arranged according to the greenhouse space structure and environmental gradient changes. S12. Collect data from the multi-point sensor nodes and construct an environmental data set D with time t and spatial position as indexes: }; in, represents the temperature data collected by the i-th sensor node, represents the humidity data collected by the i-th sensor node, represents the light intensity data collected by the i-th sensor node, represents the wind speed data collected by the i-th sensor node, ( ) represents the spatial position coordinates of the i-th sensor node, represents the timestamp of the collected data, and N represents the total number of sensor nodes in the greenhouse; S13. Perform a preliminary integrity check on the collected environmental data to ensure that all environmental parameters are included in each node data set. , and mark missing or abnormal data as unavailable points, recorded as set E: ; in,( ) is the spatial position coordinate of the unavailable point, The timestamp of the unavailable point; S14. Eliminate the unusable point data marked in step S13 to form a complete environmental data set : .

[0022] In this implementation, S2 includes the following steps: S21. Complete environment dataset Perform analysis and fill in missing environmental parameter values ​​based on interpolation if any; S22. Perform outlier detection on the filled environmental data set, use the triple standard deviation method to detect the environmental parameter values, and use the weighted average method of adjacent nodes to correct the detected outlier set O; S23. Normalize the corrected environmental data set, map the environmental parameter values ​​to the interval [0,1], and record the normalized environmental data set .

[0023] In this implementation, S3 includes the following steps: S31. From the preprocessed environmental dataset The spatial location features and environmental parameter features of each sensor node are extracted to form a multimodal feature vector : ; S32. Based on feature vector The spatial location features in the sensor nodes are used to calculate the spatial distance between the sensor nodes. : ; in, is the Euclidean distance between the i-th node and the j-th node; S33. Combined spatial distance and the Euclidean distance between the environmental parameter features , calculate the comprehensive similarity between sensor nodes : ; in, and are the weight coefficients of spatial distance and environmental parameter distance respectively, Represents the environmental parameter distance between the i-th and j-th nodes: ; S34. Based on comprehensive similarity The K-means algorithm is used to cluster the sensor nodes and generate K initial regions. The region division results are recorded as: ; in, represents the set of sensor nodes in the kth region, satisfying ; S35. Verify the generated initial area, calculate the mean value of the environmental parameters and the spatial compactness of the sensor nodes in each area, make the division result conform to the actual environmental characteristics in the greenhouse, and record the verified initial partition structure as the input area division result.

[0024] In this implementation, S4 includes the following steps: S41. Continuously collect real-time environmental data sets using a sensor array deployed in the greenhouse And associate it with the initial partition structure R to form a real-time regional dataset: ; S42. Compute real-time regional datasets Changes in environmental parameters : ; in, Indicates the environmental parameter values ​​collected in real time. represents the historical mean of the sensor node in the initial partition structure, is the number of sensor nodes in the kth region; S43. Compare the environmental parameter change with the preset threshold Compare and mark the kth region as the region to be adjusted when the following conditions are met: ; in, The permissible range of greenhouse environmental parameters is set dynamically based on historical data; S44. Recalculate the comprehensive similarity for the marked areas that need adjustment , and re-divide the regional affiliation of sensor nodes based on the K-means algorithm to generate updated regional division results ; S45. Calculate the number of sensor nodes in each area after update and the mean of regional environmental parameters, if the number of nodes in a certain area is lower than the set threshold , then merge the area into the adjacent area. If the environmental parameter change of a certain area is greater than the preset value, refine the partition, add new partition boundaries, and adjust the optimized regional division results. Recorded as adaptive partitioning results.

[0025] In this implementation, S5 includes the following steps: S51. Results for adaptive partitioning Each area , extract the environmental parameter change trend of each sensor node within the time window, and calculate the time series feature vector : ; in, Respectively represent the i-th sensor node in the time window The rate of change of interior temperature, humidity, light intensity, and wind speed; S52. Based on adaptive partition results Calculate the spatial distribution characteristics of sensor nodes in each partition, including the average distance between nodes and the standard deviation of the node distribution : ; ; in, represents the average distance from the i-th sensor node to other nodes; S53. For each partition Calculate the correlation coefficient matrix between various environmental parameters in the area , the elements in the correlation coefficient matrix Defined as: ; in, For parameters and The covariance of and The parameters are and The standard deviation of S54. Combine the time series features, spatial distribution features and environmental parameter association features extracted in steps S51, S52 and S53 to form a complete feature set : ; in, is the comprehensive feature vector of the kth partition.

[0026] In this implementation, S6 includes the following steps: S61. Feature Set Multidimensional data enhancement is performed before sample division, and enhanced samples are constructed by expanding the time dimension and adding spatial gradient information. : ; in, Respectively represent partitions The temperature gradient along the x, y, z spatial axes at time t is defined as: ; ; ; S62. Based on enhanced samples Divide into training set and test set according to the time window, and use time and space gradient information as new input features to form a comprehensive sample set and ; S63. Initialize the temperature prediction model for each partition and construct the loss function Defined as: ; in, is the total loss of the temperature prediction model, Represents the model prediction value With actual value The error between is the neighborhood weight, is a coordination factor used to balance local errors and global consistency, is the number of training samples in the partition; S64. In the process of temperature prediction model training, the enhanced samples of S61 are combined to dynamically evaluate the contribution of the newly added temporal and spatial gradient features, and calculate the feature importance score I(f): ; in, is the dth feature in the feature set, and D is the number of dimensions of the feature set; Optimize the modeling capability of the temperature prediction model for key features of the greenhouse microenvironment by dynamically eliminating low-contribution features; S65. Input real-time environmental data to the trained temperature prediction model , combining the temporal and spatial gradient characteristics to predict the future temperature change trend of the sub-region: ; At the same time, the uncertainty of the prediction results is quantified to generate the prediction interval ; S66. Prediction results for all partitions Integrate to form the future temperature forecast results of the greenhouse as a whole At the same time, combined with the uncertainty quantification results, the partition triggers enhanced sample retraining when the prediction error is greater than the preset value, forming a closed-loop optimization process.

[0027] A greenhouse temperature data monitoring system based on data analysis includes the following modules: The data acquisition module is a multi-point sensor array arranged in the greenhouse, which is used to collect environmental data in real time. The environmental data includes temperature, humidity, light intensity and wind speed. The layout of the sensor array is determined according to the spatial structure of the greenhouse and the characteristics of environmental gradient changes; The data processing module is connected to the data collection module and is used to fill missing values, detect outliers and normalize the collected environmental data, and output the preprocessed environmental data set; The regional division module is used to perform multimodal regional division of the greenhouse space according to the preprocessed data set output by the data processing module, generate the initial partition structure based on the spatial location characteristics and environmental parameter characteristics of the sensor, and dynamically adjust the partition boundaries and the number of regions to generate adaptive partition results; The feature extraction module is used to extract the time series features, spatial distribution features and correlation features between environmental parameters of each partition from the adaptive partition results, and generate a feature set by combining the time and space gradient information to provide input data for model training; The model training module is used to divide the training set and the test set according to the time window based on the feature set generated by the feature extraction module, and use the improved gradient boosting tree algorithm to build the temperature prediction model on each adaptive partition, and optimize the model performance according to the importance of dynamic features; The prediction analysis module is connected to the model training module and is used to input real-time environmental data, combine the temperature prediction model of greenhouse partitions, generate the prediction results of future temperature change trends of each partition, and generate the prediction interval through uncertainty quantification analysis; The control suggestion module integrates the temperature change trends of each partition based on the future temperature prediction results generated by the prediction and analysis module, and generates greenhouse environment control suggestions based on the uncertainty analysis results, including adjustment plans for the ventilation system, heating system and shading device; The feedback optimization module connects the data collection module, model training module and control suggestion module. It is used to continuously collect new environmental data and monitor the control effects. It triggers feature resampling and model retraining based on real-time data to form a dynamic closed-loop optimization mechanism.

[0028] Example 1: During the period from May 1, 2024 to May 31, 2024, the greenhouse microenvironment temperature monitoring system of the present invention was deployed in a modern greenhouse in the province. The greenhouse area is 2,000 square meters, and the crop planted is tomatoes. The ideal growth temperature range is 22°C to 28°C. The goal of this experiment is to use the method of the present invention to monitor and intelligently control the microenvironment temperature in the greenhouse in real time, and verify its feasibility and effectiveness in practical applications.

[0029] At 2 a.m. on May 3, 2024, the system collected environmental data in real time through a multi-point sensor array arranged in the greenhouse, and found that the temperature in area R7 suddenly dropped from 26.3°C to 20.8°C, and the humidity rose sharply from 67.5% to 91.3%. The sensor array automatically transmitted these data to the system's data processing module. After missing value filling and outlier detection, the data was confirmed to be correct. The system dynamically adjusted the regional division based on the adaptive partitioning modeling method, and recognized that the change trend of the R7 area was significantly different from that of the surrounding areas, and judged that there might be abnormal temperature control equipment or local poor ventilation in this area.

[0030] The system immediately started the feature extraction module, combined the time series characteristics (the temperature drop rate was 0.92°C / minute), spatial distribution characteristics (the average spacing between sensor nodes in the R7 area was 4.3 meters, which was denser than other areas) and the correlation characteristics of environmental parameters (the negative correlation coefficient between humidity and temperature changed from -0.75 to -0.23) to generate a feature set and input it into the gradient boosting tree model of area R7. The model predicted that if the current trend continues, the temperature in R7 will drop to 18°C ​​within the next 15 minutes, and the humidity may exceed 95%, significantly exceeding the suitable range for crops.

[0031] At the same time, the system generated a control suggestion, recommending to enable the heating equipment in the R7 area and adjust the ventilation system parameters to increase the wind speed from the current 0.5m / s to 1.2m / s to stabilize the temperature and humidity. The control suggestion was immediately transmitted to the greenhouse control center, and the management staff manually adjusted the heating and ventilation equipment according to the system feedback. Real-time data monitoring shows that after the adjustment, the temperature in the R7 area recovered to 24.1°C within 10 minutes, and the humidity dropped to 78.6%. The system confirmed that the environment has returned to a suitable range.

[0032] At 14:30 on May 12, 2024, the greenhouse system detected that the temperature in the R3 area rose from 28.5°C to 32.7°C, and the light intensity increased sharply to 58,000lx, which was much higher than the average of 34,000lx in other areas on that day. Based on adaptive zoning modeling, the system determined that the R3 area was at risk of overheating due to concentrated light and failure to adjust the shading system in time. After model prediction, the system estimated that if no control measures were taken, the temperature in the R3 area would exceed 35°C within 20 minutes, which might cause leaf burns on crops. The system generated control suggestions, automatically reduced the light transmittance of the shading equipment to 30%, and increased the frequency of the regional spray system to increase the evaporation and cooling effect. After the above control, the temperature of R3 gradually dropped to 27.5°C, and the light intensity returned to normal levels, avoiding crop losses.

[0033] In addition, in terms of regulation effect, the method of the present invention can complete the regulation response within an average of 7 minutes, while the traditional method takes about 20 minutes, and there are cases of excessive or insufficient regulation in some areas. The present invention shows significant advantages in temperature control response speed and accuracy through a closed-loop optimization process, providing reliable guarantee for greenhouse environment regulation.

[0034] This embodiment verifies the effectiveness and applicability of the present invention through actual environmental monitoring and regulation of a greenhouse. It can quickly and accurately identify microenvironmental changes in the greenhouse, put forward efficient regulation suggestions and optimize greenhouse energy consumption, significantly improving the shortcomings of traditional technologies in dynamic adaptability and regulation efficiency.

[0035] The present invention realizes accurate characterization of temperature distribution in complex greenhouse space by real-time acquisition of environmental data from multi-point sensors in the greenhouse, and adjusting partition boundaries and the number of regions according to dynamic changes in environmental parameters. It can dynamically optimize regional division according to real-time changing greenhouse environmental data, effectively avoiding regional monitoring distortion problems caused by static partitioning, and enabling the monitoring system to better adapt to dynamic changes in the greenhouse environment, thereby realizing high-precision temperature monitoring in greenhouses with significant local temperature differences and complex microenvironments.

[0036] The present invention improves the traditional gradient boosting tree algorithm in greenhouse temperature prediction. By introducing time series features and spatial gradient features, the algorithm's modeling ability for multidimensional nonlinear relationships is significantly improved. Combined with a multi-level optimization framework, the improved gradient boosting tree algorithm can not only effectively capture the complex spatiotemporal characteristics of temperature changes in the greenhouse, but also dynamically adjust the importance of features, so that the model can focus on key environmental parameters and eliminate redundant information. By introducing a global coordination term, the prediction consistency inside and outside the greenhouse partition is optimized, effectively alleviating the problem of local overfitting in greenhouse temperature prediction.

[0037] The present invention realizes the intelligent linkage of the entire chain from data collection, model training, temperature prediction to the generation of control suggestions by constructing a dynamic closed-loop optimization process. In the greenhouse temperature control link, targeted control suggestions are generated in combination with real-time prediction results, and the partition model and the prediction model are dynamically optimized according to the new data collected after the control, thereby forming a feedback closed loop of the control effect. Compared with the traditional control method based on fixed rules, the present invention has significantly improved the response speed and control accuracy, can effectively avoid energy waste, and ensure that crops grow at a suitable temperature.

[0038] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A greenhouse temperature data monitoring method based on data analysis, characterized in that: The steps include: S1. Use a multi-point sensor array arranged in the greenhouse to collect and construct an environmental data set; S2. Perform missing value filling, outlier detection and normalization processing on the environmental data set to obtain a preprocessed environmental data set; S3. Perform multimodal regional division of the greenhouse space according to the spatial location characteristics and environmental parameter characteristics contained in the preprocessed environmental data set, and form an initial partition structure; S4. Evaluate the initial partition structure based on the real-time changing environmental data in the greenhouse. When the monitored environmental data changes exceed the threshold, dynamically adjust the partition boundaries and the number of regions to obtain an adaptive partition result. S5. extracting features from the environmental data of each partition in the adaptive partitioning result, combining the time series features, spatial distribution features, and correlation features between environmental parameters to form a feature set for model training; S6. Divide the feature set into samples, use the gradient boosting tree algorithm to build a temperature prediction model on each adaptive partition, use the temperature prediction model to input and infer the environmental data of each adaptive partition, and obtain the future temperature prediction results of different partitions in the greenhouse; S7. Generate greenhouse environment control suggestions based on the future temperature prediction results of different partitions, and perform auxiliary adjustments on the ventilation system, heating system and shading device. After the adjustment is completed, continue to collect new environmental data and return to step S1 to step S6 for loop execution, forming a dynamic closed-loop optimization process.

2. A greenhouse temperature data monitoring method based on data analysis according to claim 1, characterized in that: The S1 comprises the following steps: S11. Arrange a multi-point sensor array in the greenhouse. The environmental parameters collected by each sensor node include temperature, humidity, light intensity and wind speed. The layout of the sensor array is arranged according to the greenhouse space structure and environmental gradient changes. S12. Collect data from the multi-point sensor nodes and construct an environmental data set D with time t and spatial position as indexes: }； in, represents the temperature data collected by the i-th sensor node, represents the humidity data collected by the i-th sensor node, represents the light intensity data collected by the i-th sensor node, represents the wind speed data collected by the i-th sensor node, ( ) represents the spatial position coordinates of the i-th sensor node, represents the timestamp of the collected data, and N represents the total number of sensor nodes in the greenhouse; S13. Perform a preliminary integrity check on the collected environmental data to ensure that all environmental parameters are included in each node data set. , and mark missing or abnormal data as unavailable points, recorded as set E: ； in,( ) is the spatial position coordinate of the unavailable point, The timestamp of the unavailable point; S14. Eliminate the unusable point data marked in step S13 to form a complete environmental data set : 。 3. A greenhouse temperature data monitoring method based on data analysis according to claim 1, characterized in that: The S2 comprises the following steps: S21. For the complete environment dataset Perform analysis and fill in missing environmental parameter values ​​based on interpolation if any; S22. Perform outlier detection on the filled environmental data set, use the triple standard deviation method to detect the environmental parameter values, and use the weighted average method of adjacent nodes to correct the detected outlier set O; S23. Normalize the corrected environmental data set, map the environmental parameter values ​​to the interval [0, 1], and record the normalized environmental data set. .

4. A greenhouse temperature data monitoring method based on data analysis according to claim 1, characterized in that: The S3 comprises the following steps: S31. From the preprocessed environmental dataset The spatial location features and environmental parameter features of each sensor node are extracted to form a multimodal feature vector : ； S32. Based on feature vector The spatial location features in the sensor nodes are used to calculate the spatial distance between the sensor nodes. : ； in, is the Euclidean distance between the i-th node and the j-th node; S33. Combined spatial distance and the Euclidean distance between the environmental parameter features , calculate the comprehensive similarity between sensor nodes : ； in, and are the weight coefficients of spatial distance and environmental parameter distance respectively, Represents the environmental parameter distance between the i-th and j-th nodes: ； S34. Based on comprehensive similarity The K-means algorithm is used to cluster the sensor nodes and generate K initial regions. The region division results are recorded as: ； in, represents the set of sensor nodes in the kth region, satisfying ; S35. Verify the generated initial area, calculate the mean value of the environmental parameters and the spatial compactness of the sensor nodes in each area, make the division result conform to the actual environmental characteristics in the greenhouse, and record the verified initial partition structure as the input area division result.

5. The greenhouse temperature data monitoring method based on data analysis according to claim 1 is characterized in that: The S4 comprises the following steps: S41. Continuously collect real-time environmental data sets using a sensor array deployed in the greenhouse And associate it with the initial partition structure R to form a real-time regional dataset: ； S42. Compute real-time regional datasets Changes in environmental parameters : ； in, Indicates the environmental parameter values ​​collected in real time. represents the historical mean of the sensor node in the initial partition structure, is the number of sensor nodes in the kth region; S43. Compare the environmental parameter change with the preset threshold Compare and mark the kth region as the region to be adjusted when the following conditions are met: ； in, The permissible range of greenhouse environmental parameters is set dynamically based on historical data; S44. Recalculate the comprehensive similarity for the marked areas that need adjustment , and re-divide the regional affiliation of sensor nodes based on the K-means algorithm to generate updated regional division results ; S45. Calculate the number of sensor nodes in each area after update and the mean of regional environmental parameters, if the number of nodes in a certain area is lower than the set threshold , then merge the area into the adjacent area. If the environmental parameter change of a certain area is greater than the preset value, refine the partition, add new partition boundaries, and adjust the optimized regional division results. Recorded as adaptive partitioning results.

6. A greenhouse temperature data monitoring method based on data analysis according to claim 1, characterized in that: The S5 comprises the following steps: S51. Results for adaptive partitioning Each area , extract the environmental parameter change trend of each sensor node within the time window, and calculate the time series feature vector : ； in, Respectively represent the i-th sensor node in the time window The rate of change of interior temperature, humidity, light intensity, and wind speed; S52. Based on adaptive partition results Calculate the spatial distribution characteristics of sensor nodes in each partition, including the average distance between nodes and the standard deviation of the node distribution : ； ； in, represents the average distance from the i-th sensor node to other nodes; S53. For each partition Calculate the correlation coefficient matrix between various environmental parameters in the area , the elements in the correlation coefficient matrix Defined as: ； in, For parameters and The covariance of and The parameters are and The standard deviation of S54. Combine the time series features, spatial distribution features and environmental parameter association features extracted in steps S51, S52 and S53 to form a complete feature set : ； in, is the comprehensive feature vector of the kth partition.

7. A greenhouse temperature data monitoring method based on data analysis according to claim 1, characterized in that: The S6 comprises the following steps: S61. Feature Set Multidimensional data enhancement is performed before sample division, and enhanced samples are constructed by expanding the time dimension and adding spatial gradient information. : ； in, Respectively represent partitions The temperature gradient along the x, y, z spatial axes at time t is defined as: ； ； ； S62. Based on enhanced samples Divide into training set and test set according to the time window, and use time and space gradient information as new input features to form a comprehensive sample set and ; S63. Initialize the temperature prediction model for each partition and construct the loss function Defined as: ； in, is the total loss of the temperature prediction model, Represents the model prediction value With actual value The error between is the neighborhood weight, is a coordination factor used to balance local errors and global consistency, is the number of training samples in the partition; S64. In the process of temperature prediction model training, the enhanced samples of S61 are combined to dynamically evaluate the contribution of the newly added temporal and spatial gradient features, and calculate the feature importance score I(f): ； in, is the dth feature in the feature set, and D is the number of dimensions of the feature set; Optimize the modeling capability of the temperature prediction model for key features of the greenhouse microenvironment by dynamically eliminating low-contribution features; S65. Input real-time environmental data to the trained temperature prediction model , combining the temporal and spatial gradient characteristics to predict the future temperature change trend of the sub-region: ； At the same time, the uncertainty of the prediction results is quantified to generate the prediction interval ; S66. Prediction results for all partitions Integrate to form the future temperature forecast results of the greenhouse as a whole At the same time, combined with the uncertainty quantification results, the partition triggers enhanced sample retraining when the prediction error is greater than the preset value, forming a closed-loop optimization process.

8. A greenhouse temperature data monitoring system based on data analysis, used to execute a greenhouse temperature data monitoring method based on data analysis according to any one of claims 1 to 7, characterized in that: Includes the following modules: The data acquisition module is a multi-point sensor array arranged in the greenhouse, which is used to collect environmental data in real time. The environmental data includes temperature, humidity, light intensity and wind speed. The layout of the sensor array is determined according to the spatial structure of the greenhouse and the characteristics of environmental gradient changes; The data processing module is connected to the data collection module and is used to fill missing values, detect outliers and normalize the collected environmental data, and output the preprocessed environmental data set; The regional division module is used to perform multimodal regional division of the greenhouse space according to the preprocessed data set output by the data processing module, generate the initial partition structure based on the spatial location characteristics and environmental parameter characteristics of the sensor, and dynamically adjust the partition boundaries and the number of regions to generate adaptive partition results; The feature extraction module is used to extract the time series features, spatial distribution features and correlation features between environmental parameters of each partition from the adaptive partition results, and generate a feature set by combining the time and space gradient information to provide input data for model training; The model training module is used to divide the training set and the test set according to the time window based on the feature set generated by the feature extraction module, and use the improved gradient boosting tree algorithm to build the temperature prediction model on each adaptive partition, and optimize the model performance according to the importance of dynamic features; The prediction analysis module is connected to the model training module and is used to input real-time environmental data, combine the temperature prediction model of greenhouse partitions, generate the prediction results of future temperature change trends of each partition, and generate the prediction interval through uncertainty quantification analysis; The control suggestion module integrates the temperature change trends of each partition based on the future temperature prediction results generated by the prediction and analysis module, and generates greenhouse environment control suggestions based on the uncertainty analysis results, including adjustment plans for the ventilation system, heating system and shading device; The feedback optimization module connects the data collection module, model training module and control suggestion module. It is used to continuously collect new environmental data and monitor the control effects. It triggers feature resampling and model retraining based on real-time data to form a dynamic closed-loop optimization mechanism.

Images