WO2020199453A1 - 光照精准预测的苗床调度方法、系统及介质 - Google Patents

光照精准预测的苗床调度方法、系统及介质 Download PDF

Info

Publication number
WO2020199453A1
WO2020199453A1 PCT/CN2019/100359 CN2019100359W WO2020199453A1 WO 2020199453 A1 WO2020199453 A1 WO 2020199453A1 CN 2019100359 W CN2019100359 W CN 2019100359W WO 2020199453 A1 WO2020199453 A1 WO 2020199453A1
Authority
WO
WIPO (PCT)
Prior art keywords
light
seedbed
time
data
rotation
Prior art date
Application number
PCT/CN2019/100359
Other languages
English (en)
French (fr)
Inventor
刘成良
贡亮
方锐
汪韬
吴伟
黄亦翔
Original Assignee
上海交通大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 上海交通大学 filed Critical 上海交通大学
Publication of WO2020199453A1 publication Critical patent/WO2020199453A1/zh

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A01—AGRICULTURE; FORESTRY; ANIMAL HUSBANDRY; HUNTING; TRAPPING; FISHING
    • A01G—HORTICULTURE; CULTIVATION OF VEGETABLES, FLOWERS, RICE, FRUIT, VINES, HOPS OR SEAWEED; FORESTRY; WATERING
    • A01G9/00—Cultivation in receptacles, forcing-frames or greenhouses; Edging for beds, lawn or the like
    • A01G9/28—Raised beds; Planting beds; Edging elements for beds, lawn or the like, e.g. tiles
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
    • G06Q10/06315—Needs-based resource requirements planning or analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/02—Agriculture; Fishing; Forestry; Mining

Definitions

  • the invention relates to the technical field of three-dimensional cultivation, in particular, to a seedbed scheduling method, system and medium for accurately predicting light.
  • the invention aims at the light receiving problem of plants in a three-dimensional cultivation plant factory using sunlight, and intelligently deploys seedbeds dynamically by predicting light intensity. To achieve the purpose of evenly receiving light on the crops in the three-dimensional seedbed.
  • the current optical radiation prediction methods are divided into physical models and statistical methods.
  • the physical model is based on the physical state and motion state of the atmosphere, also known as the numerical weather forecast model, which is considered the most suitable for the day-ahead and multi-day forecast range.
  • numerical weather prediction models are more rigid by weather factors, such as cloud cover, cloud evolution, and optical properties in the forecast area.
  • this model has a good prediction effect under clear sky conditions, but the prediction effect will be greatly reduced under the condition of more cloud cover.
  • the application of these physical models in long-term solar radiation prediction is also limited by computational complexity.
  • mathematical statistical models mathematical statistical models and machine learning algorithms. Mathematical statistics mainly include regression analysis, time series analysis, fuzzy theory, wavelet analysis and Kalman filtering.
  • Typical machine learning algorithms include: artificial neural networks (ANN), support vector machines (SVM) and heuristic intelligent optimization algorithms.
  • ANN artificial neural networks
  • SVM support vector machines
  • Hybrid artificial intelligence systems are quite effective for solar forecasting. Under unstable sky conditions, the improvement of machine learning technology in predicting the sun one hour in advance seems to be more obvious.
  • Patent document CN108076915A (application number: 201810026150.8) discloses an intelligent three-dimensional cultivation machine, including: a main support bracket, a track, a conveyor chain, a drive system, a planting tray, and a main controller; the track includes an upward slope Tracks, downward sloped tracks and horizontal tracks.
  • the purpose of the present invention is to provide a seedbed scheduling method, system and medium for accurately predicting illumination.
  • a seedbed scheduling method for accurate prediction of illumination includes:
  • Illumination prediction model establishment steps collect and process historical weather data and historical illumination data, and establish a light intensity prediction model
  • Light intensity prediction step collect real-time weather data, and predict light intensity according to the obtained light intensity prediction model, and output light forecast data;
  • Scheduling decision acquisition step according to the obtained light forecast data, the seedbed is scheduled.
  • the step of establishing the illumination prediction model includes:
  • Data collection step Collect historical weather data in the first area and historical weather data in the second area, and output the first and second area data sets; further, the first area refers to the area where the smallest weather forecast can be obtained For example, in the Minhang District of Shanghai, the second area refers to a certain greenhouse or a certain landmark that needs to predict the light;
  • Data processing steps remove continuous blank data segments longer than the preset duration in the first and second region data sets, fill the continuous blank data segments with the average value of two adjacent values on the time scale, and fill the filled first
  • the first and second regional data sets are averaged in hours, the data in the first and second regional data sets are combined with the same time as the standard to form sample data, and the model training features are constructed and output based on the obtained sample data;
  • Model building steps According to the obtained model training data, the light intensity is used as the prediction target, the integrated learning model is used for training, the error function is selected, and the parameters are adjusted by cross-validation to obtain the light intensity prediction model.
  • the historical weather data of the first area includes any one or more of the following:
  • the historical weather data of the second area includes: the light intensity obtained by the sensor;
  • the model training features include: all sample data; the time corresponding to the sample data; the maximum and minimum values of the forecast temperature and humidity of the day corresponding to the time of the sample data; the time corresponding to the sample data at the same time the day before All weather elements of, including: temperature, relative humidity, rainfall, weather, wind speed, wind direction, local light intensity;
  • the integrated learning model is a progressive gradient regression tree algorithm, including:
  • the error function is the root mean square error, and the formula is as follows:
  • RMSE stands for Root Mean Squared Error, that is, Root Mean Squared Error
  • n the total number of samples
  • the cross-validation method is a K-fold validation method: the training set is randomly divided into K different subsets, each subset is called a fold, and then the decision tree model is trained and evaluated K times, that is, one is selected each time Fold for evaluation, use another K-1 fold for training, the output result is an array containing K evaluation scores;
  • the parameter adjustment includes: setting the model learning rate to a first preset value, setting the ratio of the random subset of input features to the second preset value during each training, and setting the number of leaf nodes of each regression tree to the third The preset value, the number of training iterations is set to the fourth preset value.
  • the step of predicting light intensity includes:
  • Collect real-time weather data use the obtained light intensity prediction model to predict the light intensity value l pred (t) with a granularity of one hour, and output the light prediction data.
  • the step of establishing the scheduling decision model is a predefined scheduling decision model
  • l represents the light intensity value received by the crop
  • the total amount of light that each layer of crop has received at time t is:
  • i the number of layers of the seedbed
  • t represents time
  • Li (t) represents the total amount of light that the i-th crop has received as of time t;
  • l i (t) represents the real-time light intensity of the i-th seedbed.
  • l i (t) is equal to the measured real-time light.
  • l i (t) is equal to 0;
  • T start represents the time when there is light on the day
  • L pred represents the total amount of light in a whole day calculated based on the measured light and predicted light intensity
  • l pred (t) represents the predicted light intensity value with one hour as the time granularity
  • T end represents the time when the day's illumination ends
  • N the total number of layers of the seedbed
  • i the number of layers of the seedbed
  • L min represents the minimum amount of light required for the growth of seedbed crops
  • T 1 represents the end of the first rotation of the seedbed
  • T 2 represents the end of the second rotation of the seedbed
  • c represents the multiple of the total amount of light in the second rotation to the total amount of light received in the third rotation; further, in order to avoid the total amount of light received by each seedbed due to the rapid change of light intensity during the third rotation Uneven, c is set between 1 and 5.
  • K represents the ratio of the light quantity in the second rotation calculated by c to the total light quantity in the second and third rotations
  • L jpred (t) represents the total amount of light that can be obtained from time t to the end of the next rotation
  • T j represents the time when the j-th rotation ends, and j represents the number of rotations
  • L j (t) represents the total amount of light received by each seedbed after the j-th rotation predicted at time t;
  • Li (t) represents the amount of light that the i-th seedbed has obtained at time t;
  • n the total number of layers that did not receive light in this rotation, m ⁇ N;
  • N the total number of layers of the seedbed
  • L ij (t) represents the amount of light that should be made up for the i-th seedbed before the j-th rotation
  • T ij represents the time when the i-th seedbed is scheduled from the top in the j-th rotation
  • T ij is calculated from the above formula
  • ⁇ t represents the preset interval time
  • the seedbed is scheduled according to the time T ij obtained from the top level of the i-th seedbed in the jth rotation at time t.
  • a seedbed scheduling system for accurate prediction of illumination includes:
  • Illumination prediction model establishment module collect and process historical weather data and historical illumination data, and establish a light intensity prediction model
  • Light intensity prediction module collect real-time weather data, and predict light intensity according to the obtained light intensity prediction model, and output light forecast data;
  • Scheduling decision acquisition module According to the obtained light forecast data, the seedbed is scheduled.
  • the illumination prediction model establishment module includes:
  • Data collection module Collect historical weather data in the first area and historical weather data in the second area, and output the first and second area data sets;
  • Data processing module Eliminate continuous blank data segments longer than the preset duration in the first and second region data sets, and fill the continuous blank data segments with the average value of two adjacent values on the time scale, and fill the filled first
  • the first and second regional data sets are averaged in hours, the data in the first and second regional data sets are combined with the same time as the standard to form sample data, and the model training features are constructed and output based on the obtained sample data;
  • Model building module According to the obtained model training data, the light intensity is used as the prediction target, the integrated learning model is used for training, the error function is selected, and the parameters are adjusted by cross-validation to obtain the light intensity prediction model.
  • the historical weather data of the first area includes any one or more of the following:
  • the historical weather data of the second area includes: the light intensity obtained by the sensor;
  • the model training features include: all sample data; the time corresponding to the sample data; the maximum and minimum values of the forecast temperature and humidity of the day corresponding to the time of the sample data; the time corresponding to the sample data at the same time the day before All weather elements of, including: temperature, relative humidity, rainfall, weather, wind speed, wind direction, local light intensity;
  • the model building module :
  • the integrated learning model is a progressive gradient regression tree algorithm, including:
  • the error function is the root mean square error, and the formula is as follows:
  • RMSE stands for Root Mean Squared Error, that is, Root Mean Squared Error
  • n the total number of samples
  • the cross-validation method is a K-fold validation method: the training set is randomly divided into K different subsets, each subset is called a fold, and then the decision tree model is trained and evaluated K times, that is, one is selected each time Fold for evaluation, use another K-1 fold for training, the output result is an array containing K evaluation scores;
  • the parameter adjustment includes: setting the model learning rate to a first preset value, setting the ratio of the random subset of input features to the second preset value during each training, and setting the number of leaf nodes of each regression tree to the third The preset value, the number of training iterations is set to the fourth preset value.
  • the light intensity prediction module includes:
  • Collect real-time weather data use the obtained light intensity prediction model to predict the light intensity value l pred (t) with a granularity of one hour, and output the light prediction data;
  • the scheduling decision model establishment module :
  • l represents the light intensity value received by the crop
  • the total amount of light that each layer of crop has received at time t is:
  • i the number of layers of the seedbed
  • t represents time
  • Li (t) represents the total amount of light that the i-th crop has received as of time t;
  • l i (t) represents the real-time light intensity of the i-th seedbed.
  • l i (t) is equal to the measured real-time light.
  • l i (t) is equal to 0;
  • T start represents the time when there is light on the day
  • L pred represents the total amount of light in a whole day calculated based on the measured light and predicted light intensity
  • l pred (t) represents the predicted light intensity value with one hour as the time granularity
  • T end represents the time when the day's illumination ends
  • N the total number of layers of the seedbed
  • i the number of layers of the seedbed
  • L min represents the minimum amount of light required for the growth of seedbed crops
  • T 1 represents the end of the first rotation of the seedbed
  • T 2 represents the end of the second rotation of the seedbed
  • c represents the multiple of the total amount of light in the second rotation to the total amount of light received in the third rotation
  • K represents the ratio of the light quantity in the second rotation calculated by c to the total light quantity in the second and third rotations
  • L jpred (t) represents the total amount of light that can be obtained from time t to the end of the next rotation
  • T j represents the time when the j-th rotation ends, and j represents the number of rotations
  • L j (t) represents the total amount of light received by each seedbed after the j-th rotation predicted at time t;
  • Li (t) represents the amount of light that the i-th seedbed has obtained at time t;
  • n the total number of layers that did not receive light in this rotation, m ⁇ N;
  • N the total number of layers of the seedbed
  • L ij (t) represents the amount of light that should be made up for the i-th seedbed before the j-th rotation
  • T ij represents the time when the i-th seedbed is scheduled from the top in the j-th rotation
  • T ij is calculated from the above formula
  • ⁇ t represents the preset interval time
  • the seedbed is scheduled according to the time T ij obtained from the top level of the i-th seedbed in the jth rotation at time t.
  • a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of any one of the above-mentioned seedbed scheduling methods for accurate prediction of illumination.
  • the present invention has the following beneficial effects:
  • the present invention addresses the problem of inaccurate weather forecasting in small areas by weather forecast data and difficulty in forecasting the intensity of surface light.
  • the present invention establishes a small area illumination prediction model based on Web forecast information, and realizes the small area illumination intensity prediction;
  • the present invention solves the problem of the energy consumption of artificial light sources and the low land utilization rate of solar-utilizing greenhouses, and realizes the forecasting of light in small areas and the dynamic optimal deployment of three-dimensional seedbeds, thereby improving the scientific, scientific, Degree of precision and intelligence;
  • the present invention aims at the low utilization rate of natural light in the three-dimensional seedbed cultivation system and the high energy consumption of artificial light. According to the characteristics of the three-dimensional cultivation equipment using sunlight, the present invention establishes a three-dimensional cultivation seedbed that is dynamically driven in real time according to the light forecast data Scheduling decision model. Improve the natural light utilization rate of the three-dimensional seedbed and the intelligence of the plant factory cultivation system.
  • Fig. 1 is a schematic diagram of the flow of illumination prediction provided by the present invention.
  • Fig. 2 is a schematic diagram of a scheduling decision algorithm for a three-dimensional cultivation seedbed provided by the present invention.
  • Figure 3 is a schematic diagram of the three-dimensional cultivation seedbed structure provided by the present invention.
  • Fig. 4 is a schematic flow chart of a seedbed scheduling method for accurate prediction of illumination provided by the present invention.
  • a seedbed scheduling method for accurate prediction of illumination includes:
  • Illumination prediction model establishment steps collect and process historical weather data and historical illumination data, and establish a light intensity prediction model
  • Light intensity prediction step collect real-time weather data, and predict light intensity according to the obtained light intensity prediction model, and output light forecast data;
  • Scheduling decision acquisition step according to the obtained light forecast data, the seedbed is scheduled.
  • the step of establishing the illumination prediction model includes:
  • Data collection step Collect historical weather data in the first area and historical weather data in the second area, and output first and second area data sets; further, the first area refers to the place where the smallest weather forecast can be obtained in the area For example, in the Minhang District of Shanghai, the second area refers to a certain greenhouse or a certain landmark that needs to predict the light;
  • Data processing steps remove continuous blank data segments longer than the preset duration in the first and second region data sets, fill the continuous blank data segments with the average value of two adjacent values on the time scale, and fill the filled first
  • the first and second regional data sets are averaged in hours, the data in the first and second regional data sets are combined with the same time as the standard to form sample data, and the model training features are constructed and output based on the obtained sample data;
  • Model building steps According to the obtained model training data, the light intensity is used as the prediction target, the integrated learning model is used for training, the error function is selected, and the parameters are adjusted by cross-validation to obtain the light intensity prediction model.
  • the historical weather data of the first area includes any one or more of the following:
  • the historical weather data of the second area includes: the light intensity obtained by the sensor;
  • the model training features include: all sample data; the time corresponding to the sample data; the maximum and minimum values of the forecast temperature and humidity of the day corresponding to the time of the sample data; the time corresponding to the sample data at the same time the day before All weather elements of, including: temperature, relative humidity, rainfall, weather, wind speed, wind direction, local light intensity;
  • the integrated learning model is a progressive gradient regression tree algorithm, including:
  • the error function is the root mean square error, and the formula is as follows:
  • RMSE stands for Root Mean Squared Error, that is, Root Mean Squared Error
  • n the total number of samples
  • the cross-validation method is a K-fold validation method: the training set is randomly divided into K different subsets, each subset is called a fold, and then the decision tree model is trained and evaluated K times, that is, one is selected each time Fold for evaluation, use another K-1 fold for training, the output result is an array containing K evaluation scores;
  • the parameter adjustment includes: setting the model learning rate to a first preset value, setting the ratio of the random subset of input features to the second preset value during each training, and setting the number of leaf nodes of each regression tree to the third The preset value, the number of training iterations is set to the fourth preset value.
  • the step of predicting light intensity includes:
  • Collect real-time weather data use the obtained light intensity prediction model to predict the light intensity value l pred (t) with a granularity of one hour, and output the light prediction data.
  • the step of establishing the scheduling decision model is a predefined scheduling decision model
  • l represents the light intensity value received by the crop
  • the total amount of light that each layer of crop has received at time t is:
  • i the number of layers of the seedbed
  • t represents time
  • Li (t) represents the total amount of light that the i-th crop has received as of time t;
  • l i (t) represents the real-time light intensity of the i-th seedbed.
  • l i (t) is equal to the measured real-time light.
  • l i (t) is equal to 0;
  • T start represents the time when there is light on the day
  • L pred represents the total amount of light in a whole day calculated based on the measured light and predicted light intensity
  • l pred (t) represents the predicted light intensity value with one hour as the time granularity
  • T end represents the time when the day's illumination ends
  • N the total number of layers of the seedbed
  • i the number of layers of the seedbed
  • L min represents the minimum amount of light required for the growth of seedbed crops
  • T 1 represents the end of the first rotation of the seedbed
  • T 2 represents the end of the second rotation of the seedbed
  • c represents the multiple of the total amount of light in the second rotation to the total amount of light received in the third rotation; further, in order to avoid the total amount of light received by each seedbed due to the rapid change of light intensity during the third rotation Uneven, c is set between 1 and 5.
  • K represents the ratio of the light quantity in the second rotation calculated by c to the total light quantity in the second and third rotations
  • L jpred (t) represents the total amount of light that can be obtained from time t to the end of the next rotation
  • T j represents the time when the j-th rotation ends, and j represents the number of rotations
  • L j (t) represents the total amount of light received by each seedbed after the j-th rotation predicted at time t;
  • Li (t) represents the amount of light that the i-th seedbed has obtained at time t;
  • n the total number of layers that did not receive light in this rotation, m ⁇ N;
  • N the total number of layers of the seedbed
  • L ij (t) represents the amount of light that should be made up for the i-th seedbed before the j-th rotation
  • T ij represents the time when the i-th seedbed is scheduled from the top in the j-th rotation
  • T ij is calculated from the above formula
  • ⁇ t represents the preset interval time
  • the seedbed is scheduled according to the time T ij obtained from the top level of the i-th seedbed in the jth rotation at time t.
  • the seedbed scheduling system for accurate prediction of illumination provided by the present invention can be realized by the step flow of the seedbed scheduling method for accurate prediction of illumination provided by the present invention.
  • Those skilled in the art can understand the seedbed scheduling method with accurate light prediction as a preferred example of the seedbed scheduling system with accurate light prediction.
  • a seedbed scheduling system for accurate prediction of illumination includes:
  • Illumination prediction model establishment module collect and process historical weather data and historical illumination data, and establish a light intensity prediction model
  • Light intensity prediction module collect real-time weather data, and predict light intensity according to the obtained light intensity prediction model, and output light forecast data;
  • Scheduling decision acquisition module According to the obtained light forecast data, the seedbed is scheduled.
  • the illumination prediction model establishment module includes:
  • Data collection module Collect historical weather data in the first area and historical weather data in the second area, and output the first and second area data sets;
  • Data processing module Eliminate continuous blank data segments longer than the preset duration in the first and second region data sets, and fill the continuous blank data segments with the average value of two adjacent values on the time scale, and fill the filled first
  • the first and second regional data sets are averaged in hours, the data in the first and second regional data sets are combined with the same time as the standard to form sample data, and the model training features are constructed and output based on the obtained sample data;
  • Model building module According to the obtained model training data, the light intensity is used as the prediction target, the integrated learning model is used for training, the error function is selected, and the parameters are adjusted by cross-validation to obtain the light intensity prediction model.
  • the historical weather data of the first area includes any one or more of the following:
  • the historical weather data of the second area includes: the light intensity obtained by the sensor;
  • the model training features include: all sample data; the time corresponding to the sample data; the maximum and minimum values of the forecast temperature and humidity of the day corresponding to the time of the sample data; the time corresponding to the sample data at the same time the day before All weather elements of, including: temperature, relative humidity, rainfall, weather, wind speed, wind direction, local light intensity;
  • the model building module :
  • the integrated learning model is a progressive gradient regression tree algorithm, including:
  • the error function is the root mean square error, and the formula is as follows:
  • RMSE stands for Root Mean Squared Error, that is, Root Mean Squared Error
  • n the total number of samples
  • the cross-validation method is a K-fold validation method: the training set is randomly divided into K different subsets, each subset is called a fold, and then the decision tree model is trained and evaluated K times, that is, one is selected each time Fold for evaluation, use another K-1 fold for training, the output result is an array containing K evaluation scores;
  • the parameter adjustment includes: setting the model learning rate to a first preset value, setting the ratio of the random subset of input features to the second preset value during each training, and setting the number of leaf nodes of each regression tree to the third The preset value, the number of training iterations is set to the fourth preset value.
  • the light intensity prediction module includes:
  • Collect real-time weather data use the obtained light intensity prediction model to predict the light intensity value l pred (t) with a granularity of one hour, and output the light prediction data;
  • the scheduling decision model establishment module :
  • l represents the light intensity value received by the crop
  • the total amount of light that each layer of crop has received at time t is:
  • i the number of layers of the seedbed
  • t represents time
  • Li (t) represents the total amount of light that the i-th crop has received as of time t;
  • l i (t) represents the real-time light intensity of the i-th seedbed.
  • l i (t) is equal to the measured real-time light.
  • l i (t) is equal to 0;
  • T start represents the time when there is light on the day
  • L pred represents the total amount of light in a whole day calculated based on the measured light and predicted light intensity
  • l pred (t) represents the predicted light intensity value with one hour as the time granularity
  • T end represents the time when the day's illumination ends
  • N the total number of layers of the seedbed
  • i the number of layers of the seedbed
  • L min represents the minimum amount of light required for the growth of seedbed crops
  • T 1 represents the end of the first rotation of the seedbed
  • T 2 represents the end of the second rotation of the seedbed
  • c represents the multiple of the total amount of light in the second rotation to the total amount of light received in the third rotation
  • K represents the ratio of the light quantity in the second rotation calculated by c to the total light quantity in the second and third rotations
  • L jpred (t) represents the total amount of light that can be obtained from time t to the end of the next rotation
  • T j represents the time when the j-th rotation ends, and j represents the number of rotations
  • L j (t) represents the total amount of light received by each seedbed after the j-th rotation predicted at time t;
  • Li (t) represents the amount of light that the i-th seedbed has obtained at time t;
  • n the total number of layers that did not receive light in this rotation, m ⁇ N;
  • N the total number of layers of the seedbed
  • L ij (t) represents the amount of light that should be made up for the i-th seedbed before the j-th rotation
  • T ij represents the time when the i-th seedbed is scheduled from the top in the j-th rotation
  • T ij is calculated from the above formula
  • ⁇ t represents the preset interval time
  • the seedbed is scheduled according to the time T ij obtained from the top level of the i-th seedbed in the jth rotation at time t.
  • a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of any one of the above-mentioned seedbed scheduling methods for accurate prediction of illumination.
  • the invention provides a new type of seedbed scheduling method based on data-driven precise prediction of small area illumination. As shown in Figure 1, it is a schematic diagram of the flow of illumination prediction provided by the present invention, including the following steps:
  • Step 1 Data collection and processing, feature construction, and preparation for model training.
  • Step 2 Establish an illumination prediction model. Taking the light intensity as the prediction target, the integrated learning model is used for training, the error function is selected, and the parameters are adjusted by cross-validation to obtain the optimal model.
  • Step 3 Invoke the model to predict the illumination, and establish a three-dimensional cultivation seedbed scheduling decision model
  • the historical large-area forecast weather data in step 1.1 includes five variables obtained from weather forecasts published on the Internet, including temperature, relative humidity, rainfall, weather, wind speed, and wind direction.
  • the small area weather data includes the light intensity obtained by the local sensor.
  • the preliminary screening and processing of all data in the step 1.2 includes: removing continuous blank data segments larger than three hours in the original size area data set. Fill in the blank values and outliers in the large and small area data set after the passage with the average of two adjacent values on the time scale. The large and small area data set after the passage is averaged in hours. The data of the large area and the small area are combined to form a data sample based on the same time.
  • the feature constructed in step 1.3 includes: all sample data processed in step 1.2. Forecast the maximum and minimum values of temperature and humidity for the day. All weather elements at the same time on the previous day.
  • the integrated learning model in the second step is a Gradient Boosting Regression Tree (GBRT).
  • GBRT Gradient Boosting Regression Tree
  • GBRT Gradient Boosting Regression Tree
  • the error function in the second step is the root mean square error
  • the formula is
  • the cross-validation method in the second step is a K-fold validation method.
  • the training set is randomly divided into K different subsets, each subset is called a fold, and then the decision tree model is trained and evaluated K times-each time a fold is selected for evaluation, and another K- 1 fold for training.
  • the output result is an array containing K evaluation scores;
  • the parameter adjustment in the second step includes: setting the model learning rate to 0.02, setting the ratio of selecting a random subset of input features to 0.7 during each training, setting the number of leaf nodes of each regression tree to 60, and training The number of iterations is set to 1500.
  • the scheduling decision algorithm of the three-dimensional cultivation seedbed in the step 3.2 is as follows:
  • the calculation method of the total amount of light L of the crop from t1 to t2 is as follows:
  • l i (t) represents the real-time light intensity of the i-th seedbed.
  • T start represents the time when the light starts on that day.
  • Li (t) represents the total amount of light that the i-th crop has received as of time t.
  • Figure 3 it is a schematic diagram of the three-dimensional cultivation seedbed structure of the present invention.
  • K 2/3, that is: the minimum amount of light that meets the growth requirements of each seedbed at the end of the first rotation, and the total amount of light that meets the needs of the second rotation at the end of the second rotation is the total amount of light received in the third rotation 2 times.
  • L j (t) represents the total amount of light received by each layer of crops after the j-th rotation predicted at time t.
  • m represents the total number of layers that have not received light in this rotation (m ⁇ N).
  • L ij (t) represents the amount of light that should be made up for the i-th seedbed before the j-th rotation. T ij can be calculated from the above two formulas.
  • a new type of seedbed scheduling method based on data-driven precise prediction of small area illumination is characterized in that the specific steps are as follows:
  • Step 1 Data collection and processing, feature construction, and preparation for model training.
  • Step 2 Establish an illumination prediction model. Taking the light intensity as the prediction target, the integrated learning model is used for training, the error function is selected, and the parameters are adjusted by cross-validation to obtain the optimal model.
  • Step 3 Invoke the model to predict the illumination, and establish a three-dimensional cultivation seedbed scheduling decision model
  • the historical large-region forecast weather data in 1.1 includes five variables obtained from the weather forecast published on the Internet, including temperature, relative humidity, rainfall, weather, wind speed, and wind direction.
  • the small area weather data includes the light intensity obtained by the local sensor.
  • the preliminary screening and processing of the data in 1.2 includes: removing continuous blank data segments larger than three hours in the original size area data set. Fill in the blank values and outliers in the large and small area data set after the passage with the average of two adjacent values on the time scale. The large and small area data set after the passage is averaged in hours. The data of the large area and the small area are combined to form a data sample based on the same time.
  • the features constructed in 1.3 include: all sample data processed in step 1.2. Corresponds to the time of each processed sample data. The maximum and minimum values of forecast temperature and humidity corresponding to the time of each processed sample. Corresponds to all weather elements at the same time on the day before the time of each processed sample, including temperature, relative humidity, rainfall, weather, wind speed, wind direction, and local light intensity.
  • the integrated learning model is a Gradient Boosting Regression Tree (GBRT) algorithm.
  • GBRT Gradient Boosting Regression Tree
  • the error function is the root mean square error, and the formula is
  • RMSE Root Mean Squared Error
  • n the total number of samples
  • the cross-validation method is a K-fold verification method.
  • the training set is randomly divided into K different subsets, each subset is called a fold, and then the decision tree model is trained and evaluated K times-each time a fold is selected for evaluation, and another K- 1 fold for training.
  • the output result is an array containing K evaluation scores;
  • the parameter adjustment includes: setting the model learning rate to 0.02, setting the ratio of selecting a random subset of input features to 0.7 during each training, setting the number of leaf nodes of each regression tree to 60, and setting the number of training iterations to 1500.
  • the step 3.1 is characterized in that the three-dimensional cultivation seedbed scheduling decision-making algorithm driven by dynamic real-time light prediction data is as follows:
  • the seedbeds are rotated three times, and the first rotation meets the minimum amount of light required for the growth of crops in each seedbed, L min .
  • the calculation method of the total amount of light L of the crop from t1 to t2 is as follows:
  • l represents the light intensity value received by the crop
  • i the number of layers of the seedbed
  • t represents time
  • Li (t) represents the total amount of light that the i-th crop has received as of time t;
  • l i (t) represents the real-time light intensity of the i-th seedbed.
  • l i (t) is equal to the measured real-time light.
  • l i (t) is equal to 0;
  • T start represents the time when there is light on the day
  • L pred represents the total amount of light in a whole day calculated based on the measured light and predicted light intensity
  • l pred represents the predicted light intensity
  • T end represents the time when the day's light ends
  • i the number of layers of the seedbed
  • L min represents the minimum amount of light required for seedbed crop growth
  • T 1 represents the end of the first rotation of the seedbed
  • T 2 represents the end of the second rotation of the seedbed
  • K 2/3, that is: the minimum light required for the growth of each layer of seedbed is met at the end of the first rotation, and the total light received in the second rotation at the end of the second rotation is the total light received in the third rotation 2 times.
  • L jpred (t) represents the total amount of light that can be obtained from time t to the end of the next rotation
  • T j represents the end of the jth rotation
  • j represents the number of rotations
  • L j (t) represents the total amount of light received by each seedbed after the j-th rotation predicted at time t;
  • Li (t) represents the amount of light that the i-th seedbed has obtained at time t;
  • n the total number of layers that did not receive light in this rotation, m ⁇ N;
  • N the total number of layers of the seedbed
  • L ij (t) represents the amount of light that should be made up for the i-th seedbed before the j-th rotation
  • T ij represents the time when the i-th seedbed is scheduled from the top in the j-th rotation
  • T ij can be calculated from the above two formulas.

Landscapes

  • Business, Economics & Management (AREA)
  • Engineering & Computer Science (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • General Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Tourism & Hospitality (AREA)
  • Marketing (AREA)
  • Development Economics (AREA)
  • Quality & Reliability (AREA)
  • Operations Research (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Game Theory and Decision Science (AREA)
  • Mining & Mineral Resources (AREA)
  • Health & Medical Sciences (AREA)
  • Marine Sciences & Fisheries (AREA)
  • Primary Health Care (AREA)
  • General Health & Medical Sciences (AREA)
  • Animal Husbandry (AREA)
  • Agronomy & Crop Science (AREA)
  • Environmental Sciences (AREA)
  • Educational Administration (AREA)
  • Cultivation Of Plants (AREA)
  • Circuit Arrangement For Electric Light Sources In General (AREA)

Abstract

一种光照精准预测的苗床调度方法、系统及介质，包括：光照预测模型建立步骤：采集并处理历史天气数据及历史光照数据，建立光照强度预测模型；光照强度预测步骤：采集实时的天气数据，并根据获得的光照强度预测模型，预测光照强度，输出光照预测数据；调度决策获取步骤：根据获得的光照预测数据，对苗床进行调度。

Description

光照精准预测的苗床调度方法、系统及介质 技术领域
本发明涉及立体栽培技术领域，具体地，涉及光照精准预测的苗床调度方法、系统及介质。
背景技术
提高农业生产效率，实现农业装备精准控制一直是农业自动化的研究重点。随着温室种植在世界范围的推广应用，针对温室种植的相关设备得到了迅速发展，出现了以自动化多层栽培装备为代表的高度专业化、自动化农业设备。温室以其不受外界环境干扰、土地利用率高、自动化程度高的优点，在欧洲与日本等土地资源匮乏、环境恶劣的地区得到了广泛的应用。近年来，国际上植物工厂技术研发极为活跃，一方面不断引入和应用高新科技成果，朝着更加智能精准控制的方向发招；另一方面朝着更加节能和低运行成本的实用化方向发展，以实现技术的普遍化。
就人工光植物工厂而言，多采用多层式立体栽培以提高土地资源的利用效率，但人工光源的电能消耗约占植物工厂总体的80％。因此，高能耗一直是人工光植物工厂的难题。而太阳光利用型植物工厂因为考虑到所有的作物都要能够收到光照，要求作物生长方向上不能有遮挡物，一般使用单层式栽培方案，土地资源的利用效率低下。为了解决两种类型的温室问题，有必要将两者的优势进行结合，既能利用太阳光作为作物的生长资料，又能最大化土地资源利用率。本发明针对太阳光利用型立体栽培植物工厂的植物受光问题，通过对光照强度进行预测，智能化动态部署苗床。达到使立体苗床中的作物均匀受光的目的。
目前的光辐射预测方法分为物理模型和统计方法。物理模型是以大气的物理状态和运动状态为基础的，又称数值天气预报模型，被认为最适合与日前和多日预报范围。然而，数值天气预报模型受天气因素硬性较大，如预报区域的云量、云演化和光学性质等。一般这种模型在晴空条件下预测效良好，但在云量较多的情况下预测效果会大大降低。此外，这些物理模型在长期太阳辐射预测中的应用也收到计算复杂度的限制。统计模型分为两种：数理统计模型和机器学习算法。数理统计主要包括回归分析、时间序列分析、 模糊理论、小波分析和卡尔曼滤波。在实际应用中，由于各种因素导致的参数随时间变化，统计方法的预测精度不如数值天气预报模型精度高。典型的机器学习算法包括：人工神经网络(ANN)、支持向量机(SVM)和启发式智能优化算法。混合人工智能系统对于太阳能预测是相当有效的，在不稳定的天空条件下，机器学习技术在提前一小时预测太阳的改善似乎更明显。
专利文献CN108076915A(申请号：201810026150.8)公开了一种智能化立体栽培机，包括：主支撑支架、轨道、输送链、驱动系统、种植穴盘和主控制器；所述轨道包括倾斜向上的上坡轨道、倾斜向下的下坡轨道和水平轨道。
发明内容
针对现有技术中的缺陷，本发明的目的是提供一种光照精准预测的苗床调度方法、系统及介质。
根据本发明提供的一种光照精准预测的苗床调度方法，包括：
光照预测模型建立步骤：采集并处理历史天气数据及历史光照数据，建立光照强度预测模型；
光照强度预测步骤：采集实时的天气数据，并根据获得的光照强度预测模型，预测光照强度，输出光照预测数据；
调度决策获取步骤：根据获得的光照预测数据，对苗床进行调度。
优选地，所述光照预测模型建立步骤包括：
数据采集步骤：采集历史第一区域天气数据及历史第二区域天气数据，输出第一及第二区域数据集；进一步地，所述第一区域指得是该地区最小能获得的天气预报的地方，比如上海市闵行区，第二区域指的是需要预测光照的某个温室或者某个地标；
数据处理步骤：剔除第一及第二区域数据集中的大于预设时长的连续空白数据段，并将连续空白数据段用时间尺度相邻的两个值的平均值进行填充，将填充后的第一及第二区域数据集以小时为单位进行平均操作，将第一及第二区域数据集中的数据以时间相同为标准合并组成样本数据，根据获得的样本数据构造并输出模型训练特征；
模型建立步骤：根据获得的模型训练数据，以光照强度为预测目标，采用集成学习模型进行训练，选择误差函数，交叉验证进行参数调整，获得光照强度预测模型。
优选地，所述历史第一区域天气数据包括以下所述任一种或任多种：
温度、相对湿度、降雨量、天气、风速、风向；
所述历史第二区域天气数据包括：传感器获取的光照强度；
所述模型训练特征包括：所有样本数据；对应于样本数据的时间；对应于样本数据的时间当日的预报温度、湿度的最大值和最小值；对应于样本数据的时间当日的前一日相同时刻的所有天气要素，包括：温度、相对湿度、降雨量、天气、风速、风向、本地光照强度；
所述模型建立步骤：
所述集成学习模型为渐进梯度回归树算法，包括：
训练一个基础回归树，用它对训练集进行预测，计算出决策树的预测残差，接着使用这个残差对第二个回归树进行训练，然后再次计算残差，继续训练第三个回归树，并不断循环向前，最后将所有树的预测残差相加，从而对新实例进行预测；
所述误差函数为均方根误差，公式如下：
Figure PCTCN2019100359-appb-000001
RMSE表示均方根误差，即Root Mean Squared Error；
m表示样本总数；
Figure PCTCN2019100359-appb-000002
表示模型预测值；
y i表示真实值；
所述交叉验证方法为K-fold验证方法：将训练集随机分割成K个不同的子集，每个子集称为一个折叠，然后对决策树模型进行K次训练和评估，即每次挑选一个折叠进行评估，使用另外的K-1个折叠进行训练，产出的结果是一个包含K次评估分数的数组；
所述参数调整包括：将模型学习速率设置为第一预设值，每次训练时选择输入特征随机子集的比例设置为第二预设值，每个回归树的叶节点数量设置为第三预设值，训练迭代次数设置为第四预设值。
优选地，所述光照强度预测步骤包括：
采集实时的天气数据，使用获得的光照强度预测模型，预测以一小时为时间粒度的光照强度值l pred(t)，并输出光照预测数据。
优选地，所述调度决策模型建立步骤：
预设只有在顶层苗床的作物才会接收到光照，在下层苗床的作物接收到的光照强度为0，在一天中，苗床共轮换j次，第一次轮换满足每层苗床作物生长光量需求最低量L min；
苗床作物在t 1到t 2时间段内光照总量L的计算公式如下：
Figure PCTCN2019100359-appb-000003
其中，
l表示作物接收到的光照强度值；
初始时，t＝T start,j＝1，其中j为轮换次数；
在t时刻每层作物已经获得的光照总量为：
Figure PCTCN2019100359-appb-000004
其中，
i表示苗床的层数；
t表示时刻；
L i(t)表示第i层作物截至t时刻已经获得的光照总量；
l i(t)表示第i层苗床的实时光照强度，当第i层苗床被调度到顶层接收光照时，l i(t)等于所测得的实时光照，当第i层苗床不在顶层时，l i(t)等于0；
T start表示当天开始有光照的时刻；
在t时刻，根据已测得光照和预测光照强度l pred(t)计算一整天的光照总量L pred：
Figure PCTCN2019100359-appb-000005
其中，
L pred表示根据已测得光照和预测光照强度计算出的一整天的光照总量；
l pred(t)表示预测的以一小时为时间粒度的光照强度值；
T end表示一天光照结束的时间；
N表示苗床的总层数；
确定每一次轮换结束的时间，满足：
Figure PCTCN2019100359-appb-000006
K＝c/(c+1)  (5)
Figure PCTCN2019100359-appb-000007
其中，
i表示苗床的层数；
L min表示苗床作物生长光量需求最低量；
T 1表示苗床第一次轮换结束的时间；
T 2表示苗床第二次轮换结束的时间；
c表示第二次轮换中的光照总量对于第三次轮换接收的光照总量的倍数；进一步地，为避免在第三次轮换时间内光照强度变化过快导致每层苗床接收到的总光量不均，c设置为1至5之间。
K表示由c计算得到的第二次轮换中的光量占第二、三次轮换的总光量的比例系数；
根据预测的光照强度计算t时刻到下一次轮换结束时可以获得的光照总量L jpred(t)为：
Figure PCTCN2019100359-appb-000008
L jpred(t)表示t时刻到下一次轮换结束时可以获得的光照总量；
T j表示第j次轮换结束的时间，j表示轮换次数；
可以得出在t时刻时第i层苗床在第j次轮换中的从顶层调度下来的时间T ij，推导公式如下：
Figure PCTCN2019100359-appb-000009
L ij(t)＝L j(t)-L i(t)  (9)
Figure PCTCN2019100359-appb-000010
其中，
L j(t)表示在t时刻预测的第j次轮换后每层苗床所受到的总光量；
L i(t)表示第i层苗床在t时刻已经获得的光量；
m表示在该次轮换中未接受光照的总层数，m≤N；
N表示苗床的总层数；
L ij(t)表示第i层苗床在第j次次轮换之前应该补足的光量；
T ij表示第i层苗床在第j次轮换中的从顶层调度下来的时间
由上述公式计算出T ij；
判断是否t+△t≤T j：
若t+△t≤T j，则t更新为t+△t，返回公式(2)重新计算；
若t+△t>T j，则判断轮换次数j是否大于预设次数：若不大于，则j＝j+1,t＝t+△t,返回公式(2)重新计算；若大于，则结束计算；
其中，△t表示预设间隔时间；
根据获得的在t时刻时第i层苗床在第j次轮换中的从顶层调度下来的时间T ij，对苗床进行调度。
根据本发明提供的一种光照精准预测的苗床调度系统，包括：
光照预测模型建立模块：采集并处理历史天气数据及历史光照数据，建立光照强度预测模型；
光照强度预测模块：采集实时的天气数据，并根据获得的光照强度预测模型，预测光照强度，输出光照预测数据；
调度决策获取模块：根据获得的光照预测数据，对苗床进行调度。
优选地，所述光照预测模型建立模块包括：
数据采集模块：采集历史第一区域天气数据及历史第二区域天气数据，输出第一及第二区域数据集；
数据处理模块：剔除第一及第二区域数据集中的大于预设时长的连续空白数据段，并将连续空白数据段用时间尺度相邻的两个值的平均值进行填充，将填充后的第一及第二区域数据集以小时为单位进行平均操作，将第一及第二区域数据集中的数据以时间相同为标准合并组成样本数据，根据获得的样本数据构造并输出模型训练特征；
模型建立模块：根据获得的模型训练数据，以光照强度为预测目标，采用集成学习模型进行训练，选择误差函数，交叉验证进行参数调整，获得光照强度预测模型。
优选地，所述历史第一区域天气数据包括以下所述任一种或任多种：
温度、相对湿度、降雨量、天气、风速、风向；
所述历史第二区域天气数据包括：传感器获取的光照强度；
所述模型训练特征包括：所有样本数据；对应于样本数据的时间；对应于样本数据的时间当日的预报温度、湿度的最大值和最小值；对应于样本数据的时间当日的前一日相同时刻的所有天气要素，包括：温度、相对湿度、降雨量、天气、风速、风向、本地光照强度；
所述模型建立模块：
所述集成学习模型为渐进梯度回归树算法，包括：
训练一个基础回归树，用它对训练集进行预测，计算出决策树的预测残差，接着使用这个残差对第二个回归树进行训练，然后再次计算残差，继续训练第三个回归树，并不断循环向前，最后将所有树的预测残差相加，从而对新实例进行预测；
所述误差函数为均方根误差，公式如下：
Figure PCTCN2019100359-appb-000011
RMSE表示均方根误差，即Root Mean Squared Error；
m表示样本总数；
Figure PCTCN2019100359-appb-000012
表示模型预测值；
y i表示真实值；
所述交叉验证方法为K-fold验证方法：将训练集随机分割成K个不同的子集，每个子集称为一个折叠，然后对决策树模型进行K次训练和评估，即每次挑选一个折叠进行评估，使用另外的K-1个折叠进行训练，产出的结果是一个包含K次评估分数的数组；
所述参数调整包括：将模型学习速率设置为第一预设值，每次训练时选择输入特征随机子集的比例设置为第二预设值，每个回归树的叶节点数量设置为第三预设值，训练迭代次数设置为第四预设值。
优选地，所述光照强度预测模块包括：
采集实时的天气数据，使用获得的光照强度预测模型，预测以一小时为时间粒度的光照强度值l pred(t)，并输出光照预测数据；
所述调度决策模型建立模块：
预设只有在顶层苗床的作物才会接收到光照，在下层苗床的作物接收到的光照强度为0，在一天中，苗床共轮换j次，第一次轮换满足每层苗床作物生长光量需求最低量L min；
苗床作物在t 1到t 2时间段内光照总量L的计算公式如下：
Figure PCTCN2019100359-appb-000013
其中，
l表示作物接收到的光照强度值；
初始时，t＝T start,j＝1，其中j为轮换次数；
在t时刻每层作物已经获得的光照总量为：
Figure PCTCN2019100359-appb-000014
其中，
i表示苗床的层数；
t表示时刻；
L i(t)表示第i层作物截至t时刻已经获得的光照总量；
l i(t)表示第i层苗床的实时光照强度，当第i层苗床被调度到顶层接收光照时，l i(t)等于所测得的实时光照，当第i层苗床不在顶层时，l i(t)等于0；
T start表示当天开始有光照的时刻；
在t时刻，根据已测得光照和预测光照强度l pred(t)计算一整天的光照总量L pred：
Figure PCTCN2019100359-appb-000015
其中，
L pred表示根据已测得光照和预测光照强度计算出的一整天的光照总量；
l pred(t)表示预测的以一小时为时间粒度的光照强度值；
T end表示一天光照结束的时间；
N表示苗床的总层数；
确定每一次轮换结束的时间，满足：
Figure PCTCN2019100359-appb-000016
K＝c/(c+1)  (5)
Figure PCTCN2019100359-appb-000017
其中，
i表示苗床的层数；
L min表示苗床作物生长光量需求最低量；
T 1表示苗床第一次轮换结束的时间；
T 2表示苗床第二次轮换结束的时间；
c表示第二次轮换中的光照总量对于第三次轮换接收的光照总量的倍数；
K表示由c计算得到的第二次轮换中的光量占第二、三次轮换的总光量的比例系数；
根据预测的光照强度计算t时刻到下一次轮换结束时可以获得的光照总量L jpred(t)为：
Figure PCTCN2019100359-appb-000018
L jpred(t)表示t时刻到下一次轮换结束时可以获得的光照总量；
T j表示第j次轮换结束的时间，j表示轮换次数；
可以得出在t时刻时第i层苗床在第j次轮换中的从顶层调度下来的时间T ij，推导公式如下：
Figure PCTCN2019100359-appb-000019
L ij(t)＝L j(t)-L i(t)  (9)
Figure PCTCN2019100359-appb-000020
其中，
L j(t)表示在t时刻预测的第j次轮换后每层苗床所受到的总光量；
L i(t)表示第i层苗床在t时刻已经获得的光量；
m表示在该次轮换中未接受光照的总层数，m≤N；
N表示苗床的总层数；
L ij(t)表示第i层苗床在第j次次轮换之前应该补足的光量；
T ij表示第i层苗床在第j次轮换中的从顶层调度下来的时间
由上述公式计算出T ij；
判断是否t+△t≤T j：
若t+△t≤T j，则t更新为t+△t，返回公式(2)重新计算；
若t+△t>T j，则判断轮换次数j是否大于预设次数：若不大于，则j＝j+1,t＝t+△t,返回公式(2)重新计算；若大于，则结束计算；
其中，△t表示预设间隔时间；
根据获得的在t时刻时第i层苗床在第j次轮换中的从顶层调度下来的时间T ij，对苗床进行调度。
根据本发明提供的一种存储有计算机程序的计算机可读存储介质，所述计算机程序被处理器执行时实现上述中任一项所述的光照精准预测的苗床调度方法的步骤。
与现有技术相比，本发明具有如下的有益效果：
1、本发明针对天气预报数据对小区域天气预测不精确，地表光照强度预测困难的问题。本发明建立了基于Web预报信息的小区域光照预测模型，实现了小区域光照强度预测；
2、本发明解决了人工光源的耗能问题和太阳光利用型温室的土地利用率低下问题，实现了小区域光照预测和立体苗床的动态优化部署，从而提高了植物工厂作物栽培的科学化、精确化和智能化程度；
2、本发明针对立体苗床栽培系统自然光利用率不高，而人工光能耗较高的问题，本发明根据太阳光利用型立体栽培装备的特性，建立根据光照预测数据动态实时驱动的立体栽培苗床调度决策模型。提高了立体苗床的自然光利用率和植物工厂栽培系统的智能化。
附图说明
通过阅读参照以下附图对非限制性实施例所作的详细描述，本发明的其它特征、目的和优点将会变得更明显：
图1为本发明提供的光照预测的流程示意图。
图2是本发明提供的立体栽培苗床调度决策算法示意图。
图3是本发明提供的立体栽培苗床结构示意图。
[根据细则91更正 21.10.2019]　
图4是本发明提供的光照精准预测的苗床调度方法流程示意图。
具体实施方式
下面结合具体实施例对本发明进行详细说明。以下实施例将有助于本领域的技术人员进一步理解本发明，但不以任何形式限制本发明。应当指出的是，对本领域的普通技术人员来说，在不脱离本发明构思的前提下，还可以做出若干变化和改进。这些都属于本发明的保护范围。
根据本发明提供的一种光照精准预测的苗床调度方法，包括：
光照预测模型建立步骤：采集并处理历史天气数据及历史光照数据，建立光照强度预测模型；
光照强度预测步骤：采集实时的天气数据，并根据获得的光照强度预测模型，预测光照强度，输出光照预测数据；
调度决策获取步骤：根据获得的光照预测数据，对苗床进行调度。
优选地，所述光照预测模型建立步骤包括：
数据采集步骤：采集历史第一区域天气数据及历史第二区域天气数据，输出第一及第二区域数据集；进一步地，所述第一区域指得是该地区最小能获得的天气预报的地方，比如上海市闵行区，第二区域指的是需要预测光照的某个温室或者某个地标；
数据处理步骤：剔除第一及第二区域数据集中的大于预设时长的连续空白数据段，并将连续空白数据段用时间尺度相邻的两个值的平均值进行填充，将填充后的第一及第二区域数据集以小时为单位进行平均操作，将第一及第二区域数据集中的数据以时间相同为标准合并组成样本数据，根据获得的样本数据构造并输出模型训练特征；
模型建立步骤：根据获得的模型训练数据，以光照强度为预测目标，采用集成学习模型进行训练，选择误差函数，交叉验证进行参数调整，获得光照强度预测模型。
优选地，所述历史第一区域天气数据包括以下所述任一种或任多种：
温度、相对湿度、降雨量、天气、风速、风向；
所述历史第二区域天气数据包括：传感器获取的光照强度；
所述模型训练特征包括：所有样本数据；对应于样本数据的时间；对应于样本数据的时间当日的预报温度、湿度的最大值和最小值；对应于样本数据的时间当日的前一日相同时刻的所有天气要素，包括：温度、相对湿度、降雨量、天气、风速、风向、本地 光照强度；
所述模型建立步骤：
所述集成学习模型为渐进梯度回归树算法，包括：
训练一个基础回归树，用它对训练集进行预测，计算出决策树的预测残差，接着使用这个残差对第二个回归树进行训练，然后再次计算残差，继续训练第三个回归树，并不断循环向前，最后将所有树的预测残差相加，从而对新实例进行预测；
所述误差函数为均方根误差，公式如下：
Figure PCTCN2019100359-appb-000021
RMSE表示均方根误差，即Root Mean Squared Error；
m表示样本总数；
Figure PCTCN2019100359-appb-000022
表示模型预测值；
y i表示真实值；
所述交叉验证方法为K-fold验证方法：将训练集随机分割成K个不同的子集，每个子集称为一个折叠，然后对决策树模型进行K次训练和评估，即每次挑选一个折叠进行评估，使用另外的K-1个折叠进行训练，产出的结果是一个包含K次评估分数的数组；
所述参数调整包括：将模型学习速率设置为第一预设值，每次训练时选择输入特征随机子集的比例设置为第二预设值，每个回归树的叶节点数量设置为第三预设值，训练迭代次数设置为第四预设值。
优选地，所述光照强度预测步骤包括：
采集实时的天气数据，使用获得的光照强度预测模型，预测以一小时为时间粒度的光照强度值l pred(t)，并输出光照预测数据。
优选地，所述调度决策模型建立步骤：
预设只有在顶层苗床的作物才会接收到光照，在下层苗床的作物接收到的光照强度为0，在一天中，苗床共轮换j次，第一次轮换满足每层苗床作物生长光量需求最低量L min；
苗床作物在t 1到t 2时间段内光照总量L的计算公式如下：
Figure PCTCN2019100359-appb-000023
其中，
l表示作物接收到的光照强度值；
初始时，t＝T start,j＝1，其中j为轮换次数；
在t时刻每层作物已经获得的光照总量为：
Figure PCTCN2019100359-appb-000024
其中，
i表示苗床的层数；
t表示时刻；
L i(t)表示第i层作物截至t时刻已经获得的光照总量；
l i(t)表示第i层苗床的实时光照强度，当第i层苗床被调度到顶层接收光照时，l i(t)等于所测得的实时光照，当第i层苗床不在顶层时，l i(t)等于0；
T start表示当天开始有光照的时刻；
在t时刻，根据已测得光照和预测光照强度l pred(t)计算一整天的光照总量L pred：
Figure PCTCN2019100359-appb-000025
其中，
L pred表示根据已测得光照和预测光照强度计算出的一整天的光照总量；
l pred(t)表示预测的以一小时为时间粒度的光照强度值；
T end表示一天光照结束的时间；
N表示苗床的总层数；
确定每一次轮换结束的时间，满足：
Figure PCTCN2019100359-appb-000026
K＝c/(c+1)  (5)
Figure PCTCN2019100359-appb-000027
其中，
i表示苗床的层数；
L min表示苗床作物生长光量需求最低量；
T 1表示苗床第一次轮换结束的时间；
T 2表示苗床第二次轮换结束的时间；
c表示第二次轮换中的光照总量对于第三次轮换接收的光照总量的倍数；进一步地，为避免在第三次轮换时间内光照强度变化过快导致每层苗床接收到的总光量不均，c设置为1至5之间。
K表示由c计算得到的第二次轮换中的光量占第二、三次轮换的总光量的比例系数；
根据预测的光照强度计算t时刻到下一次轮换结束时可以获得的光照总量L jpred(t)为：
Figure PCTCN2019100359-appb-000028
L jpred(t)表示t时刻到下一次轮换结束时可以获得的光照总量；
T j表示第j次轮换结束的时间，j表示轮换次数；
可以得出在t时刻时第i层苗床在第j次轮换中的从顶层调度下来的时间T ij，推导公式如下：
Figure PCTCN2019100359-appb-000029
L ij(t)＝L j(t)-L i(t)  (9)
Figure PCTCN2019100359-appb-000030
其中，
L j(t)表示在t时刻预测的第j次轮换后每层苗床所受到的总光量；
L i(t)表示第i层苗床在t时刻已经获得的光量；
m表示在该次轮换中未接受光照的总层数，m≤N；
N表示苗床的总层数；
L ij(t)表示第i层苗床在第j次次轮换之前应该补足的光量；
T ij表示第i层苗床在第j次轮换中的从顶层调度下来的时间
由上述公式计算出T ij；
判断是否t+△t≤T j：
若t+△t≤T j，则t更新为t+△t，返回公式(2)重新计算；
若t+△t>T j，则判断轮换次数j是否大于预设次数：若不大于，则j＝j+1,t＝t+△t,返回公式(2)重新计算；若大于，则结束计算；
其中，△t表示预设间隔时间；
根据获得的在t时刻时第i层苗床在第j次轮换中的从顶层调度下来的时间T ij，对苗床进行调度。
本发明提供的光照精准预测的苗床调度系统，可以通过本发明给的光照精准预测的苗床调度方法的步骤流程实现。本领域技术人员可以将所述光照精准预测的苗床调度方法，理解为所述光照精准预测的苗床调度系统的一个优选例。
根据本发明提供的一种光照精准预测的苗床调度系统，包括：
光照预测模型建立模块：采集并处理历史天气数据及历史光照数据，建立光照强度 预测模型；
光照强度预测模块：采集实时的天气数据，并根据获得的光照强度预测模型，预测光照强度，输出光照预测数据；
调度决策获取模块：根据获得的光照预测数据，对苗床进行调度。
优选地，所述光照预测模型建立模块包括：
数据采集模块：采集历史第一区域天气数据及历史第二区域天气数据，输出第一及第二区域数据集；
数据处理模块：剔除第一及第二区域数据集中的大于预设时长的连续空白数据段，并将连续空白数据段用时间尺度相邻的两个值的平均值进行填充，将填充后的第一及第二区域数据集以小时为单位进行平均操作，将第一及第二区域数据集中的数据以时间相同为标准合并组成样本数据，根据获得的样本数据构造并输出模型训练特征；
模型建立模块：根据获得的模型训练数据，以光照强度为预测目标，采用集成学习模型进行训练，选择误差函数，交叉验证进行参数调整，获得光照强度预测模型。
优选地，所述历史第一区域天气数据包括以下所述任一种或任多种：
温度、相对湿度、降雨量、天气、风速、风向；
所述历史第二区域天气数据包括：传感器获取的光照强度；
所述模型训练特征包括：所有样本数据；对应于样本数据的时间；对应于样本数据的时间当日的预报温度、湿度的最大值和最小值；对应于样本数据的时间当日的前一日相同时刻的所有天气要素，包括：温度、相对湿度、降雨量、天气、风速、风向、本地光照强度；
所述模型建立模块：
所述集成学习模型为渐进梯度回归树算法，包括：
训练一个基础回归树，用它对训练集进行预测，计算出决策树的预测残差，接着使用这个残差对第二个回归树进行训练，然后再次计算残差，继续训练第三个回归树，并不断循环向前，最后将所有树的预测残差相加，从而对新实例进行预测；
所述误差函数为均方根误差，公式如下：
Figure PCTCN2019100359-appb-000031
RMSE表示均方根误差，即Root Mean Squared Error；
m表示样本总数；
Figure PCTCN2019100359-appb-000032
表示模型预测值；
y i表示真实值；
所述交叉验证方法为K-fold验证方法：将训练集随机分割成K个不同的子集，每个子集称为一个折叠，然后对决策树模型进行K次训练和评估，即每次挑选一个折叠进行评估，使用另外的K-1个折叠进行训练，产出的结果是一个包含K次评估分数的数组；
所述参数调整包括：将模型学习速率设置为第一预设值，每次训练时选择输入特征随机子集的比例设置为第二预设值，每个回归树的叶节点数量设置为第三预设值，训练迭代次数设置为第四预设值。
优选地，所述光照强度预测模块包括：
采集实时的天气数据，使用获得的光照强度预测模型，预测以一小时为时间粒度的光照强度值l pred(t)，并输出光照预测数据；
所述调度决策模型建立模块：
预设只有在顶层苗床的作物才会接收到光照，在下层苗床的作物接收到的光照强度为0，在一天中，苗床共轮换j次，第一次轮换满足每层苗床作物生长光量需求最低量L min；
苗床作物在t 1到t 2时间段内光照总量L的计算公式如下：
Figure PCTCN2019100359-appb-000033
其中，
l表示作物接收到的光照强度值；
初始时，t＝T start,j＝1，其中j为轮换次数；
在t时刻每层作物已经获得的光照总量为：
Figure PCTCN2019100359-appb-000034
其中，
i表示苗床的层数；
t表示时刻；
L i(t)表示第i层作物截至t时刻已经获得的光照总量；
l i(t)表示第i层苗床的实时光照强度，当第i层苗床被调度到顶层接收光照时，l i(t)等于所测得的实时光照，当第i层苗床不在顶层时，l i(t)等于0；
T start表示当天开始有光照的时刻；
在t时刻，根据已测得光照和预测光照强度l pred(t)计算一整天的光照总量L pred：
Figure PCTCN2019100359-appb-000035
其中，
L pred表示根据已测得光照和预测光照强度计算出的一整天的光照总量；
l pred(t)表示预测的以一小时为时间粒度的光照强度值；
T end表示一天光照结束的时间；
N表示苗床的总层数；
确定每一次轮换结束的时间，满足：
Figure PCTCN2019100359-appb-000036
K＝c/(c+1)  (5)
Figure PCTCN2019100359-appb-000037
其中，
i表示苗床的层数；
L min表示苗床作物生长光量需求最低量；
T 1表示苗床第一次轮换结束的时间；
T 2表示苗床第二次轮换结束的时间；
c表示第二次轮换中的光照总量对于第三次轮换接收的光照总量的倍数；
K表示由c计算得到的第二次轮换中的光量占第二、三次轮换的总光量的比例系数；
根据预测的光照强度计算t时刻到下一次轮换结束时可以获得的光照总量L jpred(t)为：
Figure PCTCN2019100359-appb-000038
L jpred(t)表示t时刻到下一次轮换结束时可以获得的光照总量；
T j表示第j次轮换结束的时间，j表示轮换次数；
可以得出在t时刻时第i层苗床在第j次轮换中的从顶层调度下来的时间T ij，推导公式如下：
Figure PCTCN2019100359-appb-000039
L ij(t)＝L j(t)-L i(t)  (9)
Figure PCTCN2019100359-appb-000040
其中，
L j(t)表示在t时刻预测的第j次轮换后每层苗床所受到的总光量；
L i(t)表示第i层苗床在t时刻已经获得的光量；
m表示在该次轮换中未接受光照的总层数，m≤N；
N表示苗床的总层数；
L ij(t)表示第i层苗床在第j次次轮换之前应该补足的光量；
T ij表示第i层苗床在第j次轮换中的从顶层调度下来的时间
由上述公式计算出T ij；
判断是否t+△t≤T j：
若t+△t≤T j，则t更新为t+△t，返回公式(2)重新计算；
若t+△t>T j，则判断轮换次数j是否大于预设次数：若不大于，则j＝j+1,t＝t+△t,返回公式(2)重新计算；若大于，则结束计算；
其中，△t表示预设间隔时间；
根据获得的在t时刻时第i层苗床在第j次轮换中的从顶层调度下来的时间T ij，对苗床进行调度。
根据本发明提供的一种存储有计算机程序的计算机可读存储介质，所述计算机程序被处理器执行时实现上述中任一项所述的光照精准预测的苗床调度方法的步骤。
下面通过优选例，对本发明进行更为具体地说明。
优选例1：
本发明提供一种新型的基于数据驱动的小区域光照精准预测的苗床调度方法。如图1所示，为本发明提供的光照预测的流程示意图，包括以下步骤：
步骤一：数据采集与处理，构造特征，为模型训练做准备。
1.1获取跟踪收集得到的历史大区域预报天气数据。获取采集的小区域天气数据。
1.2对数据进行初步的筛选和处理。包括空白值、异常值的处理，大区域和小区域数据的合并。
1.3构造特征。为模型训练做数据准备。
步骤二：建立光照预测模型。以光照强度为预测目标，采用集成学习模型进行训练，选择误差函数，交叉验证进行调参，得到最优模型。
步骤三：调用模型预测光照，建立立体栽培苗床调度决策模型
3.1实时跟踪收集天气预报权威机构发布的天气预报数据，使用光照预测模型预测以一小时为时间粒度的光照强度值l pred(t)。
3.2根据太阳光利用型立体栽培装备的特性，建立动态实时光照预测数据驱动的立体栽培苗床调度决策算法。
优选地，所述步骤1.1中历史大区域预报天气数据包括从互联网发布天气预报中获 取的五个变量，包括温度，相对湿度，降雨量，天气，风速，风向。所述小区域天气数据包括本地传感器获取的光照强度。
优选地，所述步骤1.2中对所有数据进行的初步的筛选和处理包括：剔除原始大小区域数据集中的大于三小时的连续空白数据段。将经过后的大小区域数据集中的空白值和异常值用时间尺度相邻的两个值得平均值进行填充。将经过后的大小区域数据集以小时为单位进行平均操作。将大区域和小区域的数据以时间相同为标准合并组成数据样本。
优选地，所述步骤1.3中构造的特征包括：经过步骤1.2处理过后的所有样本数据。当日预报温度、湿度的最大值和最小值。前一日相同时刻的所有天气要素。
优选地，所述步骤二中的集成学习模型是渐进梯度回归树(Gradient Boosting Regression Tree,GBRT)。首先，先训练一个基础回归树，用它对训练集进行预测。计算出决策树的预测残差，接着使用这个残差对第二个回归树进行训练。然后再次计算残差，继续训练第三个回归树，并不断循环向前。最后将所有树的预测相加，从而对新实例进行预测。
优选地，所述步骤二中的误差函数为均方根误差，公式为
Figure PCTCN2019100359-appb-000041
优选地，所述步骤二中的交叉验证方法是K-fold验证方法。将训练集随机分割成K个不同的子集，每个子集称为一个折叠(fold)，然后对决策树模型进行K次训练和评估——每次挑选一个折叠进行评估，使用另外的K-1个折叠进行训练。产出的结果是一个包含K次评估分数的数组；
优选地，所述步骤二中的参数调整包括：将模型学习速率设置为0.02，每次训练时选择输入特征随机子集的比例设置为0.7，每个回归树的叶节点数量设置为60，训练迭代次数设置为1500。
优选地，如图2所示，为本发明提供的立体栽培苗床调度决策算法示意图，所述步骤3.2中的立体栽培苗床调度决策算法如下：
首先，假设只有在顶层苗床的作物才会接收到光照，在下层苗床的作物接收到的光照强度为0。在一天中，苗床共轮换三次，第一次轮换满足每层苗床生长需求最低量L min。
作物在t1到t2时间段内光照总量L的计算方法如下：
Figure PCTCN2019100359-appb-000042
(1)初始时，t＝T start,j＝1，其中j为轮换次数。
(2)在t时刻每层作物已经获得的光照总量为：
Figure PCTCN2019100359-appb-000043
其中，l i(t)表示第i层苗床的实时光照强度，当第i层苗床被调度到顶层接收光照时，l i(t)等于所测得的实时光照，当第i层苗床不在顶层时，l i(t)等于0。T start表示当天开始有光照的时刻。L i(t)表示第i层作物截至t时刻已经获得的光照总量。如图3所示，为本发明的立体栽培苗床结构示意图。
(3)在t时刻，根据已测得光照和预测光照强度计算一整天的光照总量L pred
Figure PCTCN2019100359-appb-000044
(4)确定每一次轮换结束的时间，满足：
Figure PCTCN2019100359-appb-000045
Figure PCTCN2019100359-appb-000046
其中K＝2/3，即：第一次轮换结束满足每层苗床生长需求最低光照量，第二次轮换结束满足在第二次轮换中的光照总量是第三次轮换接收的光照总量的2倍。
(5)根据预测的光照强度计算t时刻到下一次轮换结束时可以获得的光照总量L jpred(t)为：
Figure PCTCN2019100359-appb-000047
(6)可以得出在t时刻时第i层苗床在第j次轮换中的从顶层调度下来的时间T ij为,推导公式如下：
Figure PCTCN2019100359-appb-000048
L ij(t)＝L j(t)-L i(t)
Figure PCTCN2019100359-appb-000049
其中L j(t)表示在t时刻预测的第j次轮换后每层作物所受到的总光量。m表示在该次轮换中未接受光照的总层数(m≤N)。L ij(t)表示第i层苗床在第j次次轮换之前应该补足的光量，由上面两个公式可以计算出T ij。
(7)判断t+△t≤T j，其中△t可自定义，若t+△t≤T j，则t＝t+△t，返回(2)；若t+△t>T j，判断j+1>4，若为假，则j＝j+1,t＝t+△t,返回(2)；若为真，则结束计算。
优选例2：
本发明所述地一种新型的基于数据驱动的小区域光照精准预测的苗床调度方法，其特征在于，具体步骤如下：
步骤一：数据采集与处理，构造特征，为模型训练做准备。
1.1获取跟踪收集得到的历史大区域预报天气数据。获取采集的小区域天气数据。
1.2对数据进行初步的筛选和处理。包括空白值、异常值的处理，大区域数据和小区域数据的合并。
1.3构造特征。为模型训练做数据准备。
步骤二：建立光照预测模型。以光照强度为预测目标，采用集成学习模型进行训练，选择误差函数，交叉验证进行调参，得到最优模型。
步骤三：调用模型预测光照，建立立体栽培苗床调度决策模型
3.1实时跟踪收集天气预报权威机构发布的天气预报数据，使用光照预测模型预测以一小时为时间粒度的光照强度值l pred(t)。
3.2根据太阳光利用型立体栽培装备的特性，建立动态实时光照预测数据驱动的立体栽培苗床调度决策模型。
所述的步骤一，其特征在于：
所述1.1中历史大区域预报天气数据包括从互联网发布天气预报中获取的五个变量，包括温度，相对湿度，降雨量，天气，风速，风向。所述小区域天气数据包括本地传感器获取的光照强度。
所述1.2中对数据进行的初步的筛选和处理包括：剔除原始大小区域数据集中的大于三小时的连续空白数据段。将经过后的大小区域数据集中的空白值和异常值用时间尺度相邻的两个值得平均值进行填充。将经过后的大小区域数据集以小时为单位进行平均操作。将大区域和小区域的数据以时间相同为标准合并组成数据样本。
所述1.3中构造的特征包括：经过步骤1.2处理过后的所有样本数据。对应于每个处理后样本数据的时间。对应于每个处理后样本的时间当日的预报温度、湿度的最大值和最小值。对应于每个处理后样本的时间当日的前一日相同时刻的所有天气要素，包括温度，相对湿度，降雨量，天气，风速，风向，本地光照强度。
步骤二，其特征在于：
所述集成学习模型是渐进梯度回归树(Gradient Boosting Regression Tree,GBRT)算法。首先，先训练一个基础回归树，用它对训练集进行预测。计算出决策树的预测残差，接着使用这个残差对第二个回归树进行训练。然后再次计算残差，继续训练第三个 回归树，并不断循环向前。最后将所有树的预测相加，从而对新实例进行预测。
所述误差函数为均方根误差，公式为
Figure PCTCN2019100359-appb-000050
RMSE表示均方根误差(Root Mean Squared Error，RMSE)
m表示样本总数
Figure PCTCN2019100359-appb-000051
表示模型预测值
y i表示真实值
所述交叉验证方法是K-fold验证方法。将训练集随机分割成K个不同的子集，每个子集称为一个折叠(fold)，然后对决策树模型进行K次训练和评估——每次挑选一个折叠进行评估，使用另外的K-1个折叠进行训练。产出的结果是一个包含K次评估分数的数组；
所述参数调整包括：将模型学习速率设置为0.02，每次训练时选择输入特征随机子集的比例设置为0.7，每个回归树的叶节点数量设置为60，训练迭代次数设置为1500。
所述的步骤3.1，其特征在于，所述的动态实时光照预测数据驱动的立体栽培苗床调度决策算法如下：
首先，假设只有在顶层苗床的作物才会接收到光照，在下层苗床的作物接收到的光照强度为0。在一天中，苗床共轮换三次，第一次轮换满足每层苗床作物生长光量需求最低量L min。
作物在t1到t2时间段内光照总量L的计算方法如下：
Figure PCTCN2019100359-appb-000052
其中，
l表示作物接收到的光照强度值
(1)初始时，t＝T start,j＝1，其中j为轮换次数。
(2)在t时刻每层作物已经获得的光照总量为：
Figure PCTCN2019100359-appb-000053
其中，
i表示苗床的层数；
t表示时刻；
L i(t)表示第i层作物截至t时刻已经获得的光照总量；
l i(t)表示第i层苗床的实时光照强度，当第i层苗床被调度到顶层接收光照时，l i(t)等于所测得的实时光照，当第i层苗床不在顶层时，l i(t)等于0；
T start表示当天开始有光照的时刻；
(3)在t时刻，根据已测得光照和预测光照强度计算一整天的光照总量L pred
Figure PCTCN2019100359-appb-000054
其中，
L pred表示根据已测得光照和预测光照强度计算出的一整天的光照总量；
l pred表示预测的光照强度
T end表示一天光照结束的时间
(4)确定每一次轮换结束的时间，满足：
Figure PCTCN2019100359-appb-000055
Figure PCTCN2019100359-appb-000056
其中，
i表示苗床的层数
L min表示苗床作物生长光量需求最低量
T 1表示苗床第一次轮换结束的时间
T 2表示苗床第二次轮换结束的时间
K＝2/3，即：第一次轮换结束满足每层苗床生长需求最低光照量，第二次轮换结束满足在第二次轮换中的光照总量是第三次轮换接收的光照总量的2倍。
(5)根据预测的光照强度计算t时刻到下一次轮换结束时可以获得的光照总量L jpred(t)为：
Figure PCTCN2019100359-appb-000057
L jpred(t)表示t时刻到下一次轮换结束时可以获得的光照总量
T j表示第j次轮换结束的时间，j表示轮换次数
(6)可以得出在t时刻时第i层苗床在第j次轮换中的从顶层调度下来的时间T ij为,推导公式如下：
Figure PCTCN2019100359-appb-000058
L ij(t)＝L j(t)-L i(t)
Figure PCTCN2019100359-appb-000059
其中，
L j(t)表示在t时刻预测的第j次轮换后每层苗床所受到的总光量；
L i(t)表示第i层苗床在t时刻已经获得的光量；
m表示在该次轮换中未接受光照的总层数，m≤N；
N表示苗床的总层数
L ij(t)表示第i层苗床在第j次次轮换之前应该补足的光量；
T ij表示第i层苗床在第j次轮换中的从顶层调度下来的时间
由上面两个公式可以计算出T ij。
(7)判断t+△t≤T j，其中△t可自定义，若t+△t≤T j，则t＝t+△t，返回(2)；若t+△t>T j，判断j+1>4，若为假，则j＝j+1,t＝t+△t,返回(2)；若为真，则结束计算。
在本申请的描述中，需要理解的是，术语“上”、“下”、“前”、“后”、“左”、“右”、“竖直”、“水平”、“顶”、“底”、“内”、“外”等指示的方位或位置关系为基于附图所示的方位或位置关系，仅是为了便于描述本申请和简化描述，而不是指示或暗示所指的装置或元件必须具有特定的方位、以特定的方位构造和操作，因此不能理解为对本申请的限制。
本领域技术人员知道，除了以纯计算机可读程序代码方式实现本发明提供的系统、装置及其各个模块以外，完全可以通过将方法步骤进行逻辑编程来使得本发明提供的系统、装置及其各个模块以逻辑门、开关、专用集成电路、可编程逻辑控制器以及嵌入式微控制器等的形式来实现相同程序。所以，本发明提供的系统、装置及其各个模块可以被认为是一种硬件部件，而对其内包括的用于实现各种程序的模块也可以视为硬件部件内的结构；也可以将用于实现各种功能的模块视为既可以是实现方法的软件程序又可以是硬件部件内的结构。
以上对本发明的具体实施例进行了描述。需要理解的是，本发明并不局限于上述特定实施方式，本领域技术人员可以在权利要求的范围内做出各种变化或修改，这并不影响本发明的实质内容。在不冲突的情况下，本申请的实施例和实施例中的特征可以任意 相互组合。

Claims (10)

  1. 一种光照精准预测的苗床调度方法，其特征在于，包括：
    光照预测模型建立步骤：采集并处理历史天气数据及历史光照数据，建立光照强度预测模型；
    光照强度预测步骤：采集实时的天气数据，并根据获得的光照强度预测模型，预测光照强度，输出光照预测数据；
    调度决策获取步骤：根据获得的光照预测数据，对苗床进行调度。
  2. 根据权利要求1所述的光照精准预测的苗床调度方法，其特征在于，所述光照预测模型建立步骤包括：
    数据采集步骤：采集历史第一区域天气数据及历史第二区域天气数据，输出第一及第二区域数据集；
    数据处理步骤：剔除第一及第二区域数据集中的大于预设时长的连续空白数据段，并将连续空白数据段用时间尺度相邻的两个值的平均值进行填充，将填充后的第一及第二区域数据集以小时为单位进行平均操作，将第一及第二区域数据集中的数据以时间相同为标准合并组成样本数据，根据获得的样本数据构造并输出模型训练特征；
    模型建立步骤：根据获得的模型训练数据，以光照强度为预测目标，采用集成学习模型进行训练，选择误差函数，交叉验证进行参数调整，获得光照强度预测模型。
  3. 根据权利要求2所述的光照精准预测的苗床调度方法，其特征在于，所述历史第一区域天气数据包括以下所述任一种或任多种：
    温度、相对湿度、降雨量、天气、风速、风向；
    所述历史第二区域天气数据包括：传感器获取的光照强度；
    所述模型训练特征包括：所有样本数据；对应于样本数据的时间；对应于样本数据的时间当日的预报温度、湿度的最大值和最小值；对应于样本数据的时间当日的前一日相同时刻的所有天气要素，包括：温度、相对湿度、降雨量、天气、风速、风向、本地光照强度；
    所述模型建立步骤：
    所述集成学习模型为渐进梯度回归树算法，包括：
    训练一个基础回归树，用它对训练集进行预测，计算出决策树的预测残差，接着使用这个残差对第二个回归树进行训练，然后再次计算残差，继续训练第三个回归树，并 不断循环向前，最后将所有树的预测残差相加，从而对新实例进行预测；
    所述误差函数为均方根误差，公式如下：
    Figure PCTCN2019100359-appb-100001
    RMSE表示均方根误差，即Root Mean Squared Error；
    m表示样本总数；
    Figure PCTCN2019100359-appb-100002
    表示模型预测值；
    y i表示真实值；
    所述交叉验证方法为K-fold验证方法：将训练集随机分割成K个不同的子集，每个子集称为一个折叠，然后对决策树模型进行K次训练和评估，即每次挑选一个折叠进行评估，使用另外的K-1个折叠进行训练，产出的结果是一个包含K次评估分数的数组；
    所述参数调整包括：将模型学习速率设置为第一预设值，每次训练时选择输入特征随机子集的比例设置为第二预设值，每个回归树的叶节点数量设置为第三预设值，训练迭代次数设置为第四预设值。
  4. 根据权利要求3所述的光照精准预测的苗床调度方法，其特征在于，所述光照强度预测步骤包括：
    采集实时的天气数据，使用获得的光照强度预测模型，预测以一小时为时间粒度的光照强度值l pred(t)，并输出光照预测数据。
  5. 根据权利要求4所述的光照精准预测的苗床调度方法，其特征在于，所述调度决策模型建立步骤：
    预设只有在顶层苗床的作物才会接收到光照，在下层苗床的作物接收到的光照强度为0，在一天中，苗床共轮换j次，第一次轮换满足每层苗床作物生长光量需求最低量L min；
    苗床作物在t 1到t 2时间段内光照总量L的计算公式如下：
    Figure PCTCN2019100359-appb-100003
    其中，
    l表示作物接收到的光照强度值；
    初始时，t＝T start,j＝1，其中j为轮换次数；
    在t时刻每层作物已经获得的光照总量为：
    Figure PCTCN2019100359-appb-100004
    其中，
    i表示苗床的层数；
    t表示时刻；
    L i(t)表示第i层作物截至t时刻已经获得的光照总量；
    l i(t)表示第i层苗床的实时光照强度，当第i层苗床被调度到顶层接收光照时，l i(t)等于所测得的实时光照，当第i层苗床不在顶层时，l i(t)等于0；
    T start表示当天开始有光照的时刻；
    在t时刻，根据已测得光照和预测光照强度l pred(t)计算一整天的光照总量L pred：
    Figure PCTCN2019100359-appb-100005
    其中，
    L pred表示根据已测得光照和预测光照强度计算出的一整天的光照总量；
    l pred(t)表示预测的以一小时为时间粒度的光照强度值；
    T end表示一天光照结束的时间；
    N表示苗床的总层数；
    确定每一次轮换结束的时间，满足：
    Figure PCTCN2019100359-appb-100006
    K＝c/(c+1) (5)
    Figure PCTCN2019100359-appb-100007
    其中，
    i表示苗床的层数；
    L min表示苗床作物生长光量需求最低量；
    T 1表示苗床第一次轮换结束的时间；
    T 2表示苗床第二次轮换结束的时间；
    c表示第二次轮换中的光照总量对于第三次轮换接收的光照总量的倍数；
    K表示由c计算得到的第二次轮换中的光量占第二、三次轮换的总光量的比例系数；
    根据预测的光照强度计算t时刻到下一次轮换结束时可以获得的光照总量L jpred(t)为：
    Figure PCTCN2019100359-appb-100008
    L jpred(t)表示t时刻到下一次轮换结束时可以获得的光照总量；
    T j表示第j次轮换结束的时间，j表示轮换次数；
    可以得出在t时刻时第i层苗床在第j次轮换中的从顶层调度下来的时间T ij，推导公式如下：
    Figure PCTCN2019100359-appb-100009
    L ij(t)＝L j(t)-L i(t) (9)
    Figure PCTCN2019100359-appb-100010
    其中，
    L j(t)表示在t时刻预测的第j次轮换后每层苗床所受到的总光量；
    L i(t)表示第i层苗床在t时刻已经获得的光量；
    m表示在该次轮换中未接受光照的总层数，m≤N；
    N表示苗床的总层数；
    L ij(t)表示第i层苗床在第j次次轮换之前应该补足的光量；
    T ij表示第i层苗床在第j次轮换中的从顶层调度下来的时间
    由上述公式计算出T ij；
    判断是否t+△t≤T j：
    若t+△t≤T j，则t更新为t+△t，返回公式(2)重新计算；
    若t+△t>T j，则判断轮换次数j是否大于预设次数：若不大于，则j＝j+1,t＝t+△t,返回公式(2)重新计算；若大于，则结束计算；
    其中，△t表示预设间隔时间；
    根据获得的在t时刻时第i层苗床在第j次轮换中的从顶层调度下来的时间T ij，对苗床进行调度。
  6. 一种光照精准预测的苗床调度系统，其特征在于，包括：
    光照预测模型建立模块：采集并处理历史天气数据及历史光照数据，建立光照强度预测模型；
    光照强度预测模块：采集实时的天气数据，并根据获得的光照强度预测模型，预测光照强度，输出光照预测数据；
    调度决策获取模块：根据获得的光照预测数据，对苗床进行调度。
  7. 根据权利要求6所述的光照精准预测的苗床调度系统，其特征在于，所述光照预测模型建立模块包括：
    数据采集模块：采集历史第一区域天气数据及历史第二区域天气数据，输出第一及第二区域数据集；
    数据处理模块：剔除第一及第二区域数据集中的大于预设时长的连续空白数据段，并将连续空白数据段用时间尺度相邻的两个值的平均值进行填充，将填充后的第一及第二区域数据集以小时为单位进行平均操作，将第一及第二区域数据集中的数据以时间相同为标准合并组成样本数据，根据获得的样本数据构造并输出模型训练特征；
    模型建立模块：根据获得的模型训练数据，以光照强度为预测目标，采用集成学习模型进行训练，选择误差函数，交叉验证进行参数调整，获得光照强度预测模型。
  8. 根据权利要求7所述的光照精准预测的苗床调度系统，其特征在于，所述历史第一区域天气数据包括以下所述任一种或任多种：
    温度、相对湿度、降雨量、天气、风速、风向；
    所述历史第二区域天气数据包括：传感器获取的光照强度；
    所述模型训练特征包括：所有样本数据；对应于样本数据的时间；对应于样本数据的时间当日的预报温度、湿度的最大值和最小值；对应于样本数据的时间当日的前一日相同时刻的所有天气要素，包括：温度、相对湿度、降雨量、天气、风速、风向、本地光照强度；
    所述模型建立模块：
    所述集成学习模型为渐进梯度回归树算法，包括：
    训练一个基础回归树，用它对训练集进行预测，计算出决策树的预测残差，接着使用这个残差对第二个回归树进行训练，然后再次计算残差，继续训练第三个回归树，并不断循环向前，最后将所有树的预测残差相加，从而对新实例进行预测；
    所述误差函数为均方根误差，公式如下：
    Figure PCTCN2019100359-appb-100011
    RMSE表示均方根误差，即Root Mean Squared Error；
    m表示样本总数；
    Figure PCTCN2019100359-appb-100012
    表示模型预测值；
    y i表示真实值；
    所述交叉验证方法为K-fold验证方法：将训练集随机分割成K个不同的子集，每个子集称为一个折叠，然后对决策树模型进行K次训练和评估，即每次挑选一个折叠进行评估，使用另外的K-1个折叠进行训练，产出的结果是一个包含K次评估分数的数组；
    所述参数调整包括：将模型学习速率设置为第一预设值，每次训练时选择输入特征 随机子集的比例设置为第二预设值，每个回归树的叶节点数量设置为第三预设值，训练迭代次数设置为第四预设值。
  9. 根据权利要求8所述的光照精准预测的苗床调度系统，其特征在于，所述光照强度预测模块包括：
    采集实时的天气数据，使用获得的光照强度预测模型，预测以一小时为时间粒度的光照强度值l pred(t)，并输出光照预测数据；
    所述调度决策模型建立模块：
    预设只有在顶层苗床的作物才会接收到光照，在下层苗床的作物接收到的光照强度为0，在一天中，苗床共轮换j次，第一次轮换满足每层苗床作物生长光量需求最低量L min；
    苗床作物在t 1到t 2时间段内光照总量L的计算公式如下：
    Figure PCTCN2019100359-appb-100013
    其中，
    l表示作物接收到的光照强度值；
    初始时，t＝T start,j＝1，其中j为轮换次数；
    在t时刻每层作物已经获得的光照总量为：
    Figure PCTCN2019100359-appb-100014
    其中，
    i表示苗床的层数；
    t表示时刻；
    L i(t)表示第i层作物截至t时刻已经获得的光照总量；
    l i(t)表示第i层苗床的实时光照强度，当第i层苗床被调度到顶层接收光照时，l i(t)等于所测得的实时光照，当第i层苗床不在顶层时，l i(t)等于0；
    T start表示当天开始有光照的时刻；
    在t时刻，根据已测得光照和预测光照强度l pred(t)计算一整天的光照总量L pred：
    Figure PCTCN2019100359-appb-100015
    其中，
    L pred表示根据已测得光照和预测光照强度计算出的一整天的光照总量；
    l pred(t)表示预测的以一小时为时间粒度的光照强度值；
    T end表示一天光照结束的时间；
    N表示苗床的总层数；
    确定每一次轮换结束的时间，满足：
    Figure PCTCN2019100359-appb-100016
    K＝c/(c+1) (5)
    Figure PCTCN2019100359-appb-100017
    其中，
    i表示苗床的层数；
    L min表示苗床作物生长光量需求最低量；
    T 1表示苗床第一次轮换结束的时间；
    T 2表示苗床第二次轮换结束的时间；
    c表示第二次轮换中的光照总量对于第三次轮换接收的光照总量的倍数；
    K表示由c计算得到的第二次轮换中的光量占第二、三次轮换的总光量的比例系数；
    根据预测的光照强度计算t时刻到下一次轮换结束时可以获得的光照总量L jpred(t)为：
    Figure PCTCN2019100359-appb-100018
    L jpred(t)表示t时刻到下一次轮换结束时可以获得的光照总量；
    T j表示第j次轮换结束的时间，j表示轮换次数；
    可以得出在t时刻时第i层苗床在第j次轮换中的从顶层调度下来的时间T ij，推导公式如下：
    Figure PCTCN2019100359-appb-100019
    L ij(t)＝L j(t)-L i(t) (9)
    Figure PCTCN2019100359-appb-100020
    其中，
    L j(t)表示在t时刻预测的第j次轮换后每层苗床所受到的总光量；
    L i(t)表示第i层苗床在t时刻已经获得的光量；
    m表示在该次轮换中未接受光照的总层数，m≤N；
    N表示苗床的总层数；
    L ij(t)表示第i层苗床在第j次次轮换之前应该补足的光量；
    T ij表示第i层苗床在第j次轮换中的从顶层调度下来的时间
    由上述公式计算出T ij；
    判断是否t+△t≤T j：
    若t+△t≤T j，则t更新为t+△t，返回公式(2)重新计算；
    若t+△t>T j，则判断轮换次数j是否大于预设次数：若不大于，则j＝j+1,t＝t+△t,返回公式(2)重新计算；若大于，则结束计算；
    其中，△t表示预设间隔时间；
    根据获得的在t时刻时第i层苗床在第j次轮换中的从顶层调度下来的时间T ij，对苗床进行调度。
  10. 一种存储有计算机程序的计算机可读存储介质，其特征在于，所述计算机程序被处理器执行时实现权利要求1至5中任一项所述的光照精准预测的苗床调度方法的步骤。
PCT/CN2019/100359 2019-03-29 2019-08-13 光照精准预测的苗床调度方法、系统及介质 WO2020199453A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910250048.0A CN110100620B (zh) 2019-03-29 2019-03-29 光照精准预测的苗床调度方法、系统及介质
CN201910250048.0 2019-03-29

Publications (1)

Publication Number Publication Date
WO2020199453A1 true WO2020199453A1 (zh) 2020-10-08

Family

ID=67484752

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/100359 WO2020199453A1 (zh) 2019-03-29 2019-08-13 光照精准预测的苗床调度方法、系统及介质

Country Status (2)

Country Link
CN (1) CN110100620B (zh)
WO (1) WO2020199453A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115812477A (zh) * 2023-02-20 2023-03-21 华东交通大学 一种立体栽培温室的光照控制方法及光照控制系统

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110100620B (zh) * 2019-03-29 2021-06-29 上海交通大学 光照精准预测的苗床调度方法、系统及介质
CN110533256A (zh) * 2019-09-05 2019-12-03 天津华春智慧能源科技发展有限公司 一种需求侧供热负荷的预测方法
CN112800627B (zh) * 2021-02-19 2024-02-02 中国农业科学院农业环境与可持续发展研究所 一种立体栽培模式下栽培层直射光进光率分析方法及装置
CN113190064B (zh) * 2021-04-12 2022-06-14 中国农业大学 植物工厂光强最优控制方法

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103984980A (zh) * 2014-01-28 2014-08-13 中国农业大学 一种温室内温度极值的预测方法
CN104604598A (zh) * 2015-01-19 2015-05-13 同济大学 一种可调节透光率的自动移动立体苗床系统
CN106326191A (zh) * 2016-08-26 2017-01-11 河海大学 一种基于天气预报描述性数据量化日照时数的方法
CN108022023A (zh) * 2017-12-20 2018-05-11 深圳春沐源控股有限公司 一种农作物的种植产量预测方法、装置及存储介质
JP2019033720A (ja) * 2017-08-21 2019-03-07 コニカミノルタ株式会社 刈り取りスケジュール決定方法及び刈り取りスケジュール決定プログラム
CN110100620A (zh) * 2019-03-29 2019-08-09 上海交通大学 光照精准预测的苗床调度方法、系统及介质

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10728336B2 (en) * 2016-03-04 2020-07-28 Sabrina Akhtar Integrated IoT (Internet of Things) system solution for smart agriculture management
KR20180086738A (ko) * 2017-01-23 2018-08-01 한국과학기술연구원 작물 재배 가이드 정보 제공 장치, 및 그 방법

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103984980A (zh) * 2014-01-28 2014-08-13 中国农业大学 一种温室内温度极值的预测方法
CN104604598A (zh) * 2015-01-19 2015-05-13 同济大学 一种可调节透光率的自动移动立体苗床系统
CN106326191A (zh) * 2016-08-26 2017-01-11 河海大学 一种基于天气预报描述性数据量化日照时数的方法
JP2019033720A (ja) * 2017-08-21 2019-03-07 コニカミノルタ株式会社 刈り取りスケジュール決定方法及び刈り取りスケジュール決定プログラム
CN108022023A (zh) * 2017-12-20 2018-05-11 深圳春沐源控股有限公司 一种农作物的种植产量预测方法、装置及存储介质
CN110100620A (zh) * 2019-03-29 2019-08-09 上海交通大学 光照精准预测的苗床调度方法、系统及介质

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115812477A (zh) * 2023-02-20 2023-03-21 华东交通大学 一种立体栽培温室的光照控制方法及光照控制系统

Also Published As

Publication number Publication date
CN110100620B (zh) 2021-06-29
CN110100620A (zh) 2019-08-09

Similar Documents

Publication Publication Date Title
WO2020199453A1 (zh) 光照精准预测的苗床调度方法、系统及介质
Dong et al. A novel convolutional neural network framework based solar irradiance prediction method
Adenugba et al. Smart irrigation system for environmental sustainability in Africa: An Internet of Everything (IoE) approach
Stefenon et al. Photovoltaic power forecasting using wavelet neuro-fuzzy for active solar trackers
CN102736596B (zh) 基于作物信息融合的多尺度温室环境控制系统
CN108388956A (zh) 考虑辐射衰减的光伏功率预测方法
CN107390754B (zh) 基于物联网云平台的智能植物生长环境调节系统与方法
CN118247532B (zh) 一种用于植物生长环境的智能监测调控方法及系统
CN106026084B (zh) 一种基于虚拟发电部落的agc功率动态分配方法
CN113966680B (zh) 一种植物补光方法、系统、装置、设备及存储介质
CN107219759A (zh) 一种温室环境控制方法及装置
CN113554522A (zh) 基于动态神经网络的葡萄园精准滴灌控制系统
Syrotiuk et al. A hybrid system with intelligent control for the processes of resource and energy supply of a greenhouse complex with application of energy renewable sources
Saraswathy et al. Integrating intelligence in hydroponic farms
CN103984991A (zh) 一种基于微气象数据的分布式太阳能辐射预测方法及系统
Ibrahim et al. Smart irrigation system using a fuzzy logic method
Su et al. Greenhouse climate setpoint optimization: An online decision strategy
Sutarna et al. Hyperparameter Tuning Impact on Deep Learning Bi-LSTM for Photovoltaic Power Forecasting
Rodríguez et al. A hierarchical control system for maximizing profit in greenhouse crop production
Lyu et al. Design of irrigation control system for vineyard based on lora wireless communication and dynamic neural network
CN113160311A (zh) 太阳能电池板调节方法及调节装置、追日系统和存储设备
Santosh et al. Development of IoT based intelligent irrigation system using particle swarm optimization and XGBoost techniques
Weng et al. Greenhouse environment control based on computational experiments
Yu et al. Eco-climate intelligent monitoring system of an agricultural science-and-technology park based on internet of things
Anjaneyulu et al. Sustainable practices for plant nutrient control using IoT

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19922699

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1)EPC DATED 03.02.2022 (EPO FORM 1205A)

122 Ep: pct application non-entry in european phase

Ref document number: 19922699

Country of ref document: EP

Kind code of ref document: A1