Greenhouse environment prediction method, device, equipment, storage medium and program product
By constructing an enhanced feature set and a multi-stage prediction model, the problems of simple model structure and data redundancy in greenhouse environment prediction are solved, achieving high-precision, real-time temperature and humidity prediction, adapting to complex environmental changes, and supporting intelligent greenhouse control.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-04
- Publication Date
- 2026-04-10
AI Technical Summary
Existing greenhouse environment prediction models suffer from a simple structure, insufficient multivariate interaction modeling, poor response to external disturbances, and redundant data inputs, resulting in limited prediction accuracy and poor real-time performance, making them difficult to adapt to complex scenarios.
An enhanced feature set is constructed, including day-night features, lag features, time-cycle features, and weighted meteorological features. Combined with a multi-stage prediction model, including a day-night classifier, a temperature prediction sub-network, and a humidity prediction sub-network, the prediction accuracy and adaptability are improved through multi-source data fusion and differentiated model design.
It improves the accuracy and stability of greenhouse environment forecasting, enhances the ability to respond to external disturbances, reduces data input redundancy, lowers the computational burden, and adapts to the real-time control needs of complex scenarios.
Smart Images
Figure CN121829644A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of precision agriculture, and in particular to a greenhouse environment prediction method, device, equipment, storage medium and program product. BACKGROUND
[0002] In the field of precision agriculture, greenhouse environment prediction is a core technology for intelligent regulation of crop growth environment, which needs to predict key parameters such as temperature and humidity based on indoor and outdoor environment data of the greenhouse, to provide basis for regulation of ventilation, light supplement, irrigation and other devices, so as to ensure the optimal growth conditions for crops. At present, deep learning technology has become the mainstream technology direction for greenhouse environment prediction due to its strong nonlinear fitting capability.
[0003] In the prior art, one type of scheme adopts a combination model of Convolutional Neural Networks (CNN) and Gated Recurrent Unit (GRU), constructs a temperature and humidity prediction model by collecting indoor and outdoor data of the greenhouse, and implements regulation in combination with the difference value of the target parameter; another type of scheme designs a double memory module and an autoregressive structure based on Long Short-Term Memory Network (LSTM), extracts long and short term thermal environment features to realize prediction; and there is also a scheme that introduces a Lyapunov function to construct a nonlinear controller, or drives prediction and control of ventilation equipment to adjust temperature, humidity and CO2 concentration through Internet of Things data.
[0004] However, the existing schemes have obvious defects: firstly, the model structure is single or the combination method is simple, and it mainly depends on the single network characteristics of CNN, LSTM and the like, which cannot capture the spatial features, short-term dependencies and long-term trends of the environment data at the same time, such as CNN-GRU combination which can only extract short-term features and is difficult to adapt to long-period scenarios such as day-night alternation and seasonal change; secondly, the multi-variable interaction modeling is insufficient, and the coupling relationship between parameters such as temperature, humidity and light is not explicitly constructed, resulting in that the prediction ignores the influence of variable linkage and the precision is limited; thirdly, the response ability to external disturbance is weak, and key features such as day-night state and precipitation lag are not introduced, so the prediction deviation is large when facing light abnormality and weather mutation; fourthly, the data input is redundant, and it needs to rely on complete historical sequence data, which has heavy calculation burden and poor real-time performance, and the control strategy of some schemes is rigid and lacks adaptive adjustment ability, which is difficult to adapt to the complex scene of actual greenhouse. SUMMARY
[0005] The application provides a greenhouse environment prediction method, device, equipment, storage medium and program product, to solve the problems of single structure of the existing greenhouse environment prediction model, insufficient multi-variable interaction modeling, poor response to external disturbance and redundant data input, can realize high-precision prediction of temperature and humidity, enhance the adaptability of complex environment, reduce the data volume and calculation burden, adapt to edge deployment and support greenhouse intelligent regulation and control.
[0006] The application provides a greenhouse environment prediction method, comprising: acquiring global environment perception data, wherein the global environment perception data comprises indoor sensing data and outdoor meteorological data of the greenhouse; constructing an enhanced feature set according to the global environment perception data, wherein the enhanced feature set comprises diurnal feature, lag feature, time period feature and weighted meteorological feature; inputting the global environment perception data and the enhanced feature set into a trained multi-stage prediction model to output temperature prediction results and humidity prediction results of the greenhouse at a future preset time step; wherein the multi-stage prediction model comprises a diurnal classifier, a temperature prediction sub-network and a humidity prediction sub-network, the diurnal feature is a binary classification feature for indicating the real diurnal state of the greenhouse, the lag feature is a highest precipitation feature of a preset historical time period extracted from the rainfall data, the time period feature is a numerical feature reflecting the periodicity of different time periods, and the weighted meteorological feature is a feature obtained by weight distribution of the outdoor meteorological data at the current time.
[0007] According to the greenhouse environment prediction method provided by the application, the global environment perception data is analyzed and periodically coded to generate the diurnal feature.
[0008] According to the greenhouse environment prediction method provided by the application, the global environment perception data is analyzed and periodically coded to generate the diurnal feature.
[0009] According to the greenhouse environment prediction method provided by the application, the global environment perception data is analyzed and periodically coded to generate the diurnal feature.
[0010] According to the greenhouse environment prediction method provided by the application, the global environment perception data is analyzed and periodically coded to generate the diurnal feature.
[0011] According to the greenhouse environment prediction method provided by the application, the method further comprises: obtaining a training data set, wherein the training data set comprises historical global environment perception data and corresponding historical enhanced feature set; training the day and night classifier based on the training data set; combining the output of the trained day and night classifier with the global environment perception data and the historical enhanced feature set in the training data set as inputs to train the temperature prediction subnetwork and the humidity prediction subnetwork respectively.
[0012] The application further provides a greenhouse environment prediction device, comprising the following modules: an acquisition module and a processing module; the acquisition module is used to acquire global environment perception data, wherein the global environment perception data comprises greenhouse internal sensing data and greenhouse external meteorological data; the processing module is used to construct an enhanced feature set according to the global environment perception data, wherein the enhanced feature set comprises day and night features, lag features, time period features and weighted meteorological features; input the global environment perception data and the enhanced feature set into a trained multi-stage prediction model to output temperature prediction results and humidity prediction results of the greenhouse in a future preset time step; wherein the multi-stage prediction model comprises a day and night classifier, a temperature prediction subnetwork and a humidity prediction subnetwork, the day and night features are used to indicate the binary classification features of the real day and night state of the greenhouse, the lag features are the highest precipitation features of a preset historical time period extracted from the rainfall data, the time period features are used to reflect the numerical features of the periodicity of different time periods, and the weighted meteorological features are the features obtained by weighting and distributing the current time outdoor meteorological data.
[0013] According to the greenhouse environment prediction device provided by the application, the construction of the enhanced feature set according to the global environment perception data comprises: analyzing and periodically encoding the timestamp information in the global environment perception data to generate the day and night features.
[0014] According to the greenhouse environment prediction device provided by the application, the construction of the enhanced feature set according to the global environment perception data comprises: extracting rainfall data from the global environment perception data, calculating the highest precipitation in one or more preset historical time periods based on the rainfall data, and determining the lag features based on the relationship between the highest precipitation and humidity change.
[0015] According to the greenhouse environment prediction device provided by the application, the construction of the enhanced feature set according to the global environment perception data comprises: extracting and encoding the numerical values reflecting the periodicity of different time periods from the timestamp information of the global environment perception data to obtain the time period features.
[0016] According to the greenhouse environment prediction device provided by the application, the global environment perception data is used to construct an enhanced feature set, which comprises: performing weight distribution processing on the current time outdoor meteorological data in the global environment perception data to obtain the weighted meteorological feature.
[0017] According to the greenhouse environment prediction device provided by the application, the acquisition module is further used to acquire a training data set, the training data set comprising historical global environment perception data and corresponding historical enhanced feature set; the processing module is further used to train the day and night classifier based on the training data set; the output of the trained day and night classifier is combined with the global environment perception data and the historical enhanced feature set in the training data set to serve as input together, and the temperature prediction sub-network and the humidity prediction sub-network are trained respectively.
[0018] The application further provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the greenhouse environment prediction method according to any one of the above when executing the computer program.
[0019] The application further provides a non-transitory computer readable storage medium having a computer program stored thereon, and the computer program is executed by a processor to implement the greenhouse environment prediction method according to any one of the above.
[0020] The application further provides a computer program product comprising a computer program, and the computer program is executed by a processor to implement the greenhouse environment prediction method according to any one of the above.
[0021] The greenhouse environment prediction method, apparatus, equipment, storage medium, and program products provided by this invention, on the one hand, acquire comprehensive environmental perception data inside and outside the greenhouse and construct an enhanced feature set including day-night features, lag features, time-period features, and weighted meteorological features. This supplements key environmental correlation information and periodic patterns, solving the problems of insufficient multivariate interaction modeling and weak response to external disturbances in existing technologies, thus improving prediction stability in complex scenarios. On the other hand, because the multi-stage prediction model includes a day-night classifier and a dedicated temperature and humidity sub-network, it can be specifically adapted to different physical characteristics of temperature and humidity, avoiding the shortcomings of a single model in taking into account both long-term and short-term dependencies, significantly improving prediction accuracy. Furthermore, because the day-night features... The system accurately indicates the true day and night conditions, thus avoiding environmental misjudgments caused by light sensor distortion and improving the prediction stability for day-night transitions and abnormal lighting scenarios. Because the hysteresis feature captures the delayed impact of historical precipitation on current humidity, it enhances the model's response to sudden weather events and reduces systematic biases in post-rain humidity predictions. Since the time-cycle feature reflects the periodic patterns of different time periods, it helps the model fit long-term environmental trends and adapt to seasonal changes. Because the weighted meteorological feature can be obtained from current-time data through weight allocation without relying on complete historical meteorological sequences, it significantly reduces data input redundancy, lowers computational burden, and improves model real-time performance and deployability. Thus, through the synergistic effect of multi-source data fusion, enhanced feature supplementation, and differentiated model design, accurate and efficient prediction of greenhouse temperature and humidity can be achieved, providing reliable decision support for intelligent greenhouse control. Attached Figure Description
[0022] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0023] Figure 1 This is a flowchart illustrating the greenhouse environment prediction method provided by the present invention; Figure 2 This is a schematic diagram of the greenhouse environment prediction device provided by the present invention; Figure 3 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation
[0024] In order to make the objectives, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be described clearly and completely below with reference to the drawings in the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of the present application.
[0025] It should be noted that in the embodiments of the present application, the words such as "exemplary" or "for example" are used to mean serving as an example, instance, or illustration. Any embodiment or design solution described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or having more advantages than other embodiments or design solutions. In fact, the use of the words such as "exemplary" or "for example" is intended to present the relevant concept in a specific manner.
[0026] It should be noted that in this document, the terms "comprising", "containing", or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device that includes a list of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent in such a process, method, article, or device. Without more limitations, the element defined by the phrase "comprising a" does not exclude the presence of additional identical elements in the process, method, article, or device that includes the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, but can also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method can be performed in an order different from that described, and various steps can also be added, omitted, or combined. In addition, the features described with reference to certain examples can be combined in other examples.
[0027] In order to clearly describe the technical solutions of the embodiments of the present application, in the embodiments of the present application, the words "first", "second", etc. are used to distinguish the same or similar items with basically the same function and role, and those skilled in the art can understand that the words "first", "second", etc. are not limited in number and execution order.
[0028] The embodiments of the present application describe some exemplary embodiments for the purpose of illustration, and it should be understood that the present application can be implemented in other ways not specifically shown in the drawings.
[0029] As shown in Figure 1 The embodiments of the present application provide a greenhouse environment prediction method, which can be applied to a greenhouse environment prediction device. The greenhouse environment prediction method can include S101-S103: S101, the greenhouse environment prediction device acquires global environment perception data.
[0030] The global environment perception data includes greenhouse indoor sensing data and greenhouse outdoor meteorological data.
[0031] Specifically, the global environment perception data covers two types of data sources, indoor and outdoor. The greenhouse indoor sensing data is collected by deploying temperature sensors, humidity sensors and light sensors to accurately capture the dynamic changes of the core environmental parameters in the greenhouse. The greenhouse outdoor meteorological data includes temperature, humidity, rainfall and wind speed, which can fully cover the external meteorological factors affecting the greenhouse environment.
[0032] Optionally, the sampling frequency of the greenhouse indoor sensing data can be greater than that of the greenhouse outdoor meteorological data. For example, the sampling interval of the greenhouse indoor sensing data can be set to 10 minutes, and the sampling interval of the greenhouse outdoor meteorological data can be set to 1 hour.
[0033] Optionally, after collecting the global environment perception data, the greenhouse environment prediction device can perform data preprocessing on the global environment perception data. The data preprocessing includes anomaly detection, missing value processing and time synchronization operation.
[0034] Specifically, the anomaly detection can use Hampel filter based on Median Absolute Deviation (MAD) principle. The absolute deviation of each data point in the data sequence from the median of the sequence is calculated, and the median of the absolute deviation is taken as the MAD value, that is: ; When a data point satisfies , it is determined that the data point is an anomaly value and is processed; wherein is the data point to be detected, is the median of the data sequence.
[0035] The missing value processing can adopt different processing strategies for different missing time lengths. Linear interpolation method is used to complete the short-time missing data to ensure data continuity, and related outdoor meteorological data is used to assist in completing the long-time missing data to ensure data integrity.
[0036] The time synchronization operation refers to the use of resampling technology to unify the indoor and outdoor data to the time interval of the greenhouse indoor sensing data. Linear interpolation method is used to realize the time resolution of different sampling frequency data, providing standardized data basis for subsequent feature construction and model training.
[0037] S102, the greenhouse environment prediction device constructs an enhanced feature set according to the global environment perception data.
[0038] The above enhanced feature set includes a day-night feature, a lag feature, a time period feature, and a weighted weather feature. The day-night feature is a binary classification feature indicating the true day-night state of the greenhouse. The lag feature refers to the highest precipitation feature of a preset historical time period extracted from the rainfall data. The time period feature reflects the numerical features of the periodicity of different time periods. The weighted weather feature refers to the feature obtained by weighting the outdoor weather data at the current time.
[0039] Optionally, the enhanced feature set is constructed according to the global environment perception data, including: analyzing and periodically encoding the timestamp information in the global environment perception data to generate the day-night feature.
[0040] Specifically, the existing model excessively relies on the sensor raw data of the light intensity to determine the environment state. In an actual greenhouse, leaf shading, artificial shading, or light supplementing operations can severely distort the readings of the light sensor, causing the model to misjudge "midday after shading" as "night" or "night after light supplementing" as "day". This one-sided perception based on noisy data makes the model unable to understand the key environmental context of the true day-night state, thereby making completely wrong predictions, which seriously restricts the prediction accuracy and the rationality of the regulation.
[0041] Therefore, in order to solve the above problems, the timestamp information in the global environment perception data can be analyzed and periodically encoded to generate the day-night feature.
[0042] For example, the input timestamp information is converted into a day-night classification identifier through the periodicity of time by deep learning.
[0043] By introducing the day-night feature, robust perception and logical judgment of the environment state are realized. Instead of simply relying on unreliable light readings, the model autonomously infers the true day-night state through periodic encoding of time series and deep learning. This provides a reliable basis for the prerequisite regulation decisions (such as cooling in the daytime and keeping warm at night).
[0044] Optionally, the enhanced feature set is constructed according to the global environment perception data, including: extracting rainfall data from the global environment perception data, calculating the highest precipitation in one or more preset historical time periods based on the rainfall data, and determining the lag feature based on the relationship between the highest precipitation and the humidity change.
[0045] Specifically, the greenhouse environment is a high inertia system, and the influence of external disturbances often has a delay. The influence of precipitation on the humidity inside the greenhouse has significant hysteresis. Most existing models only consider the current time meteorological input, and cannot capture this hysteresis dynamic process, resulting in a persistent and systematic deviation in humidity prediction during the key time period after rainfall occurs, which cannot meet the demand for accurate regulation.
[0046] Therefore, in order to solve the above problems, the application can introduce historical precipitation as a feature, and the model can "remember" and "associate" the relationship between historical precipitation events and current humidity changes. This enables the model to perceive the trend of humidity rise in advance, and can reduce the prediction deviation in humidity prediction within several hours after rainfall occurs, and improve the predictability and initiative of regulation of the model.
[0047] Optionally, the constructing an enhanced feature set according to the global environment perception data comprises: extracting and encoding a numerical value reflecting periodicity of different time periods from time stamp information of the global environment perception data to obtain the time period feature.
[0048] Specifically, the construction of the time period feature is realized by converting absolute time information into periodic coded features, which can help the model accurately capture the environmental change law at different time scales such as day and night, week, season, etc.
[0049] For example, it can include hour period feature, week period feature and month period feature: Hour period feature: ; Week period feature: ; Month period feature: .
[0050] Optionally, the constructing an enhanced feature set according to the global environment perception data comprises: performing weight allocation processing on current time outdoor meteorological data in the global environment perception data to obtain the weighted meteorological feature.
[0051] Specifically, in order to improve accuracy, many models tend to introduce complete, long time span outdoor meteorological data sequence as input. This provides more information, but greatly increases the complexity and computational burden of the model. When facing edge computing devices or real-time control systems that require high frequency updates, this redundancy becomes a bottleneck for deployment, making it difficult to achieve low latency prediction under limited hardware resources.
[0052] Therefore, in order to solve the above problems, the present application performs weight distribution processing on the outdoor meteorological data at the current moment, and only uses the weighted meteorological features as the model input, which can provide sufficient external context information for the model, and the mathematical expression of the weight distribution is: ; wherein, is the outdoor meteorological data at the current moment.
[0053] This design greatly reduces the model parameters and the amount of calculation while maintaining the prediction accuracy, and provides support for the real-time deployment and closed-loop control of the model on resource-limited embedded devices or edge servers.
[0054] S103, the greenhouse environment prediction device inputs the global environment perception data and the enhanced feature set into the trained multi-stage prediction model, and outputs the temperature prediction result and the humidity prediction result of the greenhouse at the future preset time step.
[0055] wherein, the multi-stage prediction model includes a day-night classifier, a temperature prediction subnetwork and a humidity prediction subnetwork.
[0056] Specifically, the day-night classifier adopts a lightweight Transformer architecture and focuses on day-night pattern recognition, and its input is historical sequence data and time features. The architecture includes a one-dimensional convolution layer, a batch normalization layer, a multi-head self-attention mechanism, a global average pooling layer and a fully connected layer in sequence, and the output is a day-night binary classification probability. The temperature prediction subnetwork is designed for the medium and long term dependence characteristics of temperature, and includes a multi-scale feature extraction layer, a time series dependence modeling layer and a global attention layer in sequence. The multi-scale feature extraction layer extracts multi-dimensional features by using one-dimensional convolution with multiple different convolution kernel sizes in parallel and then fuses them. The time series dependence modeling layer uses a double-layer LSTM network to capture long time series dependence. The global attention layer uses a multi-head self-attention mechanism and is configured with a residual connection. The humidity prediction subnetwork is designed for the rapid response characteristics of humidity, and includes a single-layer one-dimensional convolution, a single-layer LSTM and a simplified fully connected layer in sequence. The multi-stage prediction model is also configured with a feature fusion mechanism, which integrates sequence features, weighted meteorological features, time features, day-night features and lag features through splicing operation, to provide comprehensive input information for the prediction subnetwork.
[0057] Optionally, before inputting the global environment perception data and the enhanced feature set into the trained multi-stage prediction model, the greenhouse environment prediction device can obtain a training data set, the training data set including historical global environment perception data and corresponding historical enhanced feature set; train the day-night classifier based on the training data set; combine the output of the trained day-night classifier with the global environment perception data and the historical enhanced feature set in the training data set as inputs to train the temperature prediction sub-network and the humidity prediction sub-network, respectively.
[0058] Specifically, an early stopping mechanism and an adaptive learning rate strategy are configured in the model training process, the early stopping mechanism is realized by monitoring the validation set loss, and the adaptive learning rate strategy adopts the ReduceLROnPlateau strategy; the model training takes the mean square error (MSE) as the loss function, and takes the coefficient of determination (R 2 ), the mean absolute error (MAE), and the root mean square error (RMSE) as evaluation indexes; the multi-stage prediction model supports multi-step prediction output, and the output expression is: ; wherein, represents the prediction result of the future preset time step, can output the prediction result of multiple time steps in the future synchronously, t is the current time, and k is the prediction step, that is, the model outputs the greenhouse temperature or humidity prediction value sequence from the t+1 time to the t+k time; represents the historical sequence data, and n is the historical sequence length, that is, the model inputs the historical data sequence of the greenhouse indoor and outdoor environment from the t-n time to the t time; represents the enhanced feature set at the current time, that is, the feature set including the weighted weather features constructed at the current time; represents the time period feature at the current time, that is, the time regularity feature such as hour, week, month, etc. after periodic encoding at the current time; represents the day-night feature at the current time, represents the lag feature at the current time.
[0059] In the embodiments of the present application, on the one hand, the global environment perception data in the greenhouse is acquired, and an enhanced feature set containing diurnal characteristics, lag characteristics, time period characteristics and weighted weather characteristics is constructed, so as to supplement the key correlation information and periodic law of the environment, solve the problems of insufficient multivariate interaction modeling and weak response to external disturbance in the prior art, and improve the prediction stability in a complex scene. On the other hand, the multi-stage prediction model contains a day-night classifier and a humidity-specific sub-network, so that the different physical characteristics of temperature and humidity can be adapted, the defect that a single model cannot take into account long-term and short-term dependencies can be avoided, and the prediction accuracy can be significantly improved. On the other hand, the diurnal characteristics can accurately indicate the real day-night state, so that the environmental misjudgment caused by the distortion of the light sensor can be avoided, and the prediction stability in the day-night alternating and light abnormal scene can be improved. The lag characteristics can capture the delayed influence of historical precipitation on the current humidity, so that the response capability of the model to sudden weather can be enhanced, and the systematic deviation of humidity prediction after rainfall can be reduced. The time period characteristics can reflect the periodic law in different time periods, so that the model can fit the long-period environmental trend and adapt to scenes such as seasonal changes. The weighted weather characteristics can be obtained based on the current time data through weight distribution, without relying on complete historical weather sequences, so that the data input redundancy can be greatly reduced, the computing burden can be reduced, and the real-time performance and deployability of the model can be improved. In this way, through the synergistic effect of multi-source data fusion, enhanced feature supplement and differentiated model design, accurate and efficient prediction of the temperature and humidity in the greenhouse can be realized, and reliable decision support for intelligent regulation and control of the greenhouse can be provided.
[0060] The above describes the scheme provided by the embodiments of the present application mainly from the perspective of the method. To implement the above functions, the hardware structure and / or software modules corresponding to each function are included. Those skilled in the art should easily realize that, in combination with the units and algorithm steps of the examples described in the embodiments disclosed herein, the embodiments of the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a certain function is implemented in hardware or computer software driven hardware depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.
[0061] The execution subject of the greenhouse environment prediction method provided by the embodiments of the present application can be a greenhouse environment prediction device, or a control module for greenhouse environment prediction in the greenhouse environment prediction device. In the embodiments of the present application, the greenhouse environment prediction device is taken as an example to execute the greenhouse environment prediction method, and the greenhouse environment prediction device provided by the embodiments of the present application is described.
[0062] It should be noted that the embodiments of the present application can divide the function modules of the greenhouse environment prediction device according to the above method examples. For example, each function module can be divided according to each function, or two or more functions can be integrated into one processing module. The integrated module can be realized in the form of hardware or in the form of a software function module. Optionally, the division of the modules in the embodiments of the present application is illustrative, and is only a logical function division. In actual implementation, another division mode can be used.
[0063] As shown in Figure 2 The greenhouse environment prediction device 200 includes an acquisition module 201 and a processing module 202. The acquisition module 201 is configured to acquire global environment perception data, wherein the global environment perception data includes indoor sensing data and outdoor meteorological data of the greenhouse. The processing module 202 is configured to construct an enhanced feature set according to the global environment perception data, wherein the enhanced feature set includes a day-night feature, a lag feature, a time period feature, and a weighted meteorological feature. The global environment perception data and the enhanced feature set are input into a trained multi-stage prediction model to output temperature prediction results and humidity prediction results of the greenhouse at a future preset time step. The multi-stage prediction model includes a day-night classifier, a temperature prediction subnetwork, and a humidity prediction subnetwork. The day-night feature is a binary classification feature indicating a real day-night state of the greenhouse. The lag feature is a highest precipitation feature of a preset historical time period extracted from rainfall data. The time period feature is a numerical feature reflecting periodicity of different time periods. The weighted meteorological feature is a feature obtained by performing weight distribution on current outdoor meteorological data.
[0064] Optionally, constructing the enhanced feature set according to the global environment perception data includes analyzing and periodically encoding timestamp information in the global environment perception data to generate the day-night feature.
[0065] Optionally, constructing the enhanced feature set according to the global environment perception data includes extracting rainfall data from the global environment perception data, calculating a highest precipitation in one or more preset historical time periods based on the rainfall data, and determining the lag feature based on a relationship between the highest precipitation and humidity change.
[0066] Optionally, constructing the enhanced feature set according to the global environment perception data includes extracting and encoding a numerical value reflecting periodicity of different time periods from timestamp information in the global environment perception data to obtain the time period feature.
[0067] Optionally, the constructing the enhanced feature set according to the global environment perception data comprises: performing weight distribution processing on current time outdoor meteorological data in the global environment perception data to obtain the weighted meteorological feature.
[0068] Optionally, the obtaining module 201 is further configured to obtain a training data set, the training data set comprising historical global environment perception data and corresponding historical enhanced feature set; the processing module 202 is further configured to train the day and night classifier based on the training data set; and the output of the trained day and night classifier is combined with the global environment perception data and the historical enhanced feature set in the training data set as input to train the temperature prediction sub-network and the humidity prediction sub-network respectively.
[0069] In the embodiments of the present application, on the one hand, since the global environment perception data inside and outside the greenhouse is obtained and the enhanced feature set containing day and night features, lag features, time period features and weighted meteorological features is constructed, the key associated information and periodic rules of the environment can be supplemented, the problems of insufficient multi-variable interaction modeling and weak response to external disturbances in the prior art can be solved, and the prediction stability in complex scenarios can be improved; on the other hand, since the multi-stage prediction model contains a day and night classifier and humidity and temperature special sub-networks, the different physical characteristics of humidity and temperature can be adaptively adapted, the defect that a single model is difficult to balance long-term and short-term dependencies can be avoided, and the prediction accuracy can be significantly improved; on the other hand, since the day and night features can accurately indicate the real day and night state, the environmental misjudgment caused by the distortion of the light sensor can be avoided, and the prediction stability in day and night alternating and light abnormal scenarios can be improved; since the lag features can capture the delayed influence of historical precipitation on the current humidity, the response capability of the model to sudden weather can be enhanced, and the systematic deviation of humidity prediction after rainfall can be reduced; since the time period features can reflect the periodicity of different time periods, the model can help to fit the long-period environmental trend and adapt to scenarios such as seasonal changes; since the weighted meteorological features can be obtained based on the current time data through weight distribution, the model does not need to rely on complete historical meteorological sequences, so the data input redundancy can be greatly reduced, the computing burden can be reduced, and the real-time performance and deployability of the model can be improved. In this way, through the synergistic effect of multi-source data fusion, enhanced feature supplement and differentiated model design, accurate and efficient prediction of greenhouse temperature and humidity can be realized, and reliable decision support for greenhouse intelligent regulation and control can be provided.
[0070] Figure 3 An example of an entity structure diagram of an electronic device is shown in FIG. 1. Figure 3As shown, the electronic device can include a processor 310, a communications interface 320, a memory 330, and a communications bus 340, wherein the processor 310, the communications interface 320, and the memory 330 communicate with each other through the communications bus 340. The processor 310 can invoke the logic instructions in the memory 330 to execute the greenhouse environment prediction method, which includes: acquiring global environment perception data, the global environment perception data including indoor sensing data and outdoor weather data of the greenhouse; constructing an enhanced feature set according to the global environment perception data, the enhanced feature set including day-night features, lag features, time period features, and weighted weather features; inputting the global environment perception data and the enhanced feature set into a trained multi-stage prediction model to output temperature prediction results and humidity prediction results of the greenhouse at a future preset time step; wherein the multi-stage prediction model includes a day-night classifier, a temperature prediction subnetwork, and a humidity prediction subnetwork, the day-night features are used to indicate the binary classification features of the real day-night state of the greenhouse, the lag features are the highest precipitation features of a preset historical time period extracted from the rainfall data, the time period features are used to reflect the numerical features of the periodicity of different time periods, and the weighted weather features are the features obtained by weighting the outdoor weather data at the current time.
[0071] In addition, the logic instructions in the memory 330 described above can be implemented in the form of a software functional unit and sold or used as an independent product, which can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the part of the technical solutions that essentially contribute to the prior art or the part of the technical solutions can be embodied in the form of a software product, which is stored in a storage medium and includes instructions for making a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in various embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various program code storage media.
[0072] In another aspect, the present application also provides a computer program product, which comprises a computer program, the computer program being stored in a non-transitory computer-readable storage medium, and the computer program being executable by a processor to enable a computer to perform the greenhouse environment prediction method provided by any of the above methods, the method comprising: obtaining global environment perception data, the global environment perception data comprising greenhouse internal sensing data and greenhouse external weather data; constructing an enhanced feature set according to the global environment perception data, the enhanced feature set comprising diurnal feature, lag feature, time period feature, and weighted weather feature; inputting the global environment perception data and the enhanced feature set into a trained multi-stage prediction model to output temperature prediction results and humidity prediction results of the greenhouse at a future preset time step; wherein the multi-stage prediction model comprises a diurnal classifier, a temperature prediction sub-network, and a humidity prediction sub-network, the diurnal feature is a binary classification feature for indicating a real diurnal state of the greenhouse, the lag feature is a highest precipitation feature of a preset historical time period extracted from rainfall data, the time period feature is a numerical feature for reflecting periodicity of different time periods, and the weighted weather feature is a feature obtained by weight distribution on current time outdoor weather data.
[0073] In yet another aspect, the present application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, the computer program being executable by a processor to implement a greenhouse environment prediction method provided by any of the above methods, the method comprising: obtaining global environment perception data, the global environment perception data comprising greenhouse internal sensing data and greenhouse external weather data; constructing an enhanced feature set according to the global environment perception data, the enhanced feature set comprising diurnal feature, lag feature, time period feature, and weighted weather feature; inputting the global environment perception data and the enhanced feature set into a trained multi-stage prediction model to output temperature prediction results and humidity prediction results of the greenhouse at a future preset time step; wherein the multi-stage prediction model comprises a diurnal classifier, a temperature prediction sub-network, and a humidity prediction sub-network, the diurnal feature is a binary classification feature for indicating a real diurnal state of the greenhouse, the lag feature is a highest precipitation feature of a preset historical time period extracted from rainfall data, the time period feature is a numerical feature for reflecting periodicity of different time periods, and the weighted weather feature is a feature obtained by weight distribution on current time outdoor weather data.
[0074] The device embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected to achieve the purposes of the embodiments according to actual needs. Those skilled in the art can understand and implement without creative labor.
[0075] Through the description of the above embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and the necessary general hardware platform, and of course can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.
[0076] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.
Claims
1. A method for predicting greenhouse environment conditions, characterized in that, include: Acquire comprehensive environmental sensing data, which includes sensor data inside the greenhouse and meteorological data outside the greenhouse; Based on the comprehensive environmental perception data, an enhanced feature set is constructed, which includes day and night features, lag features, time period features, and weighted meteorological features. The multi-stage prediction model trained by inputting the full-domain environmental perception data and the enhanced feature set outputs the temperature prediction result and humidity prediction result of the greenhouse at a preset time step in the future. The multi-stage prediction model includes a day-night classifier, a temperature prediction subnetwork, and a humidity prediction subnetwork. The day-night features are binary classification features used to indicate the true day-night state of the greenhouse. The lag features refer to the highest precipitation features extracted from the rainfall data for a preset historical time period. The time period features are numerical features used to reflect the periodic patterns of different time periods. The weighted meteorological features refer to the features obtained by weighting the outdoor meteorological data at the current moment.
2. The greenhouse environment prediction method according to claim 1, characterized in that, The step of constructing an enhanced feature set based on the global environmental perception data includes: The timestamp information in the global environmental perception data is analyzed and periodically encoded to generate the day-night features.
3. The greenhouse environment prediction method according to claim 1, characterized in that, The step of constructing an enhanced feature set based on the global environmental perception data includes: Rainfall data is extracted from the global environmental perception data, and the highest rainfall in one or more recent preset historical time periods is calculated based on the rainfall data. The lag feature is determined based on the relationship between the highest rainfall and humidity changes.
4. The greenhouse environment prediction method according to claim 1, characterized in that, The step of constructing an enhanced feature set based on the global environmental perception data includes: The time period features are obtained by extracting and encoding numerical values that reflect the periodic patterns of different time periods from the timestamp information of the global environmental perception data.
5. The greenhouse environment prediction method according to claim 1, characterized in that, The step of constructing an enhanced feature set based on the global environmental perception data includes: The weighted meteorological features are obtained by weighting the outdoor meteorological data at the current time in the global environmental perception data.
6. The greenhouse environment prediction method according to any one of claims 1-5, characterized in that, The method further includes: Obtain a training dataset, which includes historical global environment perception data and corresponding historical enhanced feature sets; The day / night classifier is trained based on the training dataset. The output of the trained day / night classifier is combined with the global environmental perception data and historical enhanced feature set in the training dataset, and used as input to train the temperature prediction subnetwork and the humidity prediction subnetwork, respectively.
7. A greenhouse environment prediction device, characterized in that, include: Acquisition module and processing module; The acquisition module is used to acquire global environmental perception data, which includes sensor data inside the greenhouse and meteorological data outside the greenhouse. The processing module is used to construct an enhanced feature set based on the global environmental perception data. The enhanced feature set includes day and night features, lag features, time period features, and weighted meteorological features. The multi-stage prediction model trained by inputting the full-domain environmental perception data and the enhanced feature set outputs the temperature prediction result and humidity prediction result of the greenhouse at a preset time step in the future. The multi-stage prediction model includes a day-night classifier, a temperature prediction subnetwork, and a humidity prediction subnetwork. The day-night features are binary classification features used to indicate the true day-night state of the greenhouse. The lag features refer to the highest precipitation features extracted from the rainfall data for a preset historical time period. The time period features are numerical features used to reflect the periodic patterns of different time periods. The weighted meteorological features refer to the features obtained by weighting the outdoor meteorological data at the current moment.
8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the greenhouse environment prediction method as described in any one of claims 1 to 6.
9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the greenhouse environment prediction method as described in any one of claims 1 to 6.
10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the greenhouse environment prediction method as described in any one of claims 1 to 6.