Photovoltaic area ecological restoration intelligent irrigation method based on automatic machine learning

Abstract

The invention relates to the technical field of desert photovoltaic area ecological restoration and irrigation, and discloses a photovoltaic area ecological restoration intelligent irrigation method based on automatic machine learning, and the method comprises the steps: obtaining photovoltaic area environment monitoring data; performing data format unified processing, abnormal data standardization processing and redundant data cleaning on the photovoltaic area environment monitoring data, and establishing a database; performing feature selection, feature extraction and feature construction on the photovoltaic area environment monitoring data in the database, and establishing a feature set; performing model training, model evaluation and model management on the intelligent model to obtain an irrigation decision model; and inputting the environmental monitoring data of the photovoltaic area into the irrigation decision-making model, and when the output result of the irrigation decision-making model is that irrigation is needed, generating a water supply pump opening instruction and a valve opening degree adjusting instruction for irrigation. And water resources are efficiently utilized.

Description

Technical Field

[0001] The present invention relates to the technical field of ecological restoration and irrigation of desert photovoltaic areas, and in particular to an intelligent irrigation method for ecological restoration of photovoltaic areas based on automatic machine learning. Background Art

[0002] With the rapid development of the new energy industry, the scale of photovoltaic power station construction has continued to expand. However, the construction of photovoltaic areas may have a certain impact on the local ecological environment, such as soil disturbance and vegetation damage. In order to achieve the sustainable development of photovoltaic areas, ecological restoration has become an important task, and the smart irrigation system is one of the key technologies to ensure vegetation growth.

[0003] Currently, photovoltaic projects in arid or semi-arid areas face problems such as difficulty in vegetation restoration and water shortage. Traditional irrigation methods often lack precision, easily lead to waste of water resources, and it is difficult to adjust irrigation strategies in real time according to soil and meteorological conditions. At the same time, there is a lack of effective monitoring methods to comprehensively evaluate the effects of ecological restoration, which is not conducive to the long-term healthy growth of vegetation. Summary of the invention

[0004] In view of this, the present invention provides an intelligent irrigation method for ecological restoration of photovoltaic areas based on automatic machine learning to solve the problem that traditional irrigation methods cannot irrigate according to soil and meteorological conditions.

[0005] In a first aspect, the present invention provides a photovoltaic area ecological restoration smart irrigation method based on automatic machine learning, the method comprising:

[0006] Obtaining environmental monitoring data of the photovoltaic area, including soil temperature, humidity and salinity monitoring data, meteorological parameter monitoring data, and vegetation monitoring data;

[0007] Unify the data format, normalize abnormal data, and clean redundant data of the photovoltaic area environmental monitoring data to establish a database;

[0008] Perform feature selection, feature extraction, and feature construction on the photovoltaic area environmental monitoring data in the database to establish a feature set;

[0009] Conduct model training, model evaluation and model management on the intelligent model to obtain the irrigation decision model;

[0010] The environmental monitoring data of the photovoltaic area is input into the irrigation decision model. When the output result of the irrigation decision model is that irrigation is needed, a water supply pump start instruction and a valve opening adjustment instruction are generated for irrigation.

[0011] The present invention monitors the photovoltaic area environment by integrating multiple monitoring devices, preprocesses the photovoltaic area environment monitoring data, performs feature processing, and constructs a feature set to prepare data for the irrigation decision model. The irrigation decision model outputs whether irrigation is needed, realizes automatic irrigation in combination with soil and meteorological conditions, and realizes efficient use of water resources and scientific management of vegetation.

[0012] In an optional implementation, the photovoltaic area environmental monitoring data is subjected to unified data format processing, abnormal data normalization processing, and redundant data cleaning, including:

[0013] Unify the data format of the photovoltaic area environmental monitoring data collected by different monitoring equipment;

[0014] Normalize abnormal data of PV area environmental monitoring data collected by abnormal equipment;

[0015] The compressed, converted and offline stored photovoltaic area environmental monitoring data are cleaned.

[0016] The present invention unifies the data formats of photovoltaic area environmental monitoring data collected by different monitoring devices to avoid the situation where the collected data formats are not unified due to differences in monitoring devices, normalizes the abnormal data of photovoltaic area environmental monitoring data collected by abnormal devices to solve the problem of abnormal collected data due to damage, failure or other failure conditions of monitoring equipment, and cleans the compressed, converted and offline stored photovoltaic area environmental monitoring data to solve the problems of compression, conversion and offline storage due to the large space occupied by the original files.

[0017] In an optional implementation, establishing a database includes:

[0018] MySQL is used as the main database to index soil temperature, humidity, salinity and meteorological parameters, optimize the format, encode and compress vegetation monitoring data, and divide them into different libraries and tables to establish a database.

[0019] The present invention uses MySQL as the main database to process soil temperature, humidity and salinity, meteorological parameters and vegetation monitoring data respectively to establish a database, thereby preparing data for model training.

[0020] In an optional implementation, feature selection, feature extraction, and feature construction are performed on the photovoltaic area environmental monitoring data in the database, including:

[0021] Perform null value processing, feature importance analysis, and feature selection on the photovoltaic area environmental monitoring data in the database;

[0022] Dimensionality reduction is used to extract the effective subspace of features, and convolution is used to extract feature expressions;

[0023] Feature construction is performed by transforming fields within features and combining features.

[0024] The present invention performs feature selection on the photovoltaic area environmental monitoring data in the database to remove invalid features, retain high-contribution features, perform dimensionality reduction and convolution processing on the features to obtain more expressive features, and perform feature construction to adapt to different model selections.

[0025] In an optional implementation, the intelligent model includes a machine learning model and a neural network model, and model training of the intelligent model includes:

[0026] When the intelligent model is a machine learning model, the basic classification model and the integrated learning model are used for model training, and hyperparameter optimization is used for tuning;

[0027] When the intelligent model is a neural network model, FastAI pre-defined neural network architecture is used, and PyTorch is used to provide the underlying neural network construction capabilities and perform automatic training.

[0028] The present invention selects machine learning models and neural network models to train the models, so as to select models for different model spaces, and adapts corresponding hyperparameter optimization and architecture search techniques to improve the performance of the models.

[0029] In an optional implementation, the intelligent model is evaluated and managed, including:

[0030] Use cross-validation, integrated validation, and evaluation indicators to evaluate the intelligent model;

[0031] The intelligent model is managed through experimental material management, training log management, model backup management, and resource scheduling management.

[0032] The present invention performs model management on the intelligent model by performing model evaluation on the intelligent model to improve the generalization ability of the model, utilize the advantages of different models, adjust the model training method, and manage various intermediate files and resource scheduling during the model training and model evaluation process.

[0033] In an optional implementation, after generating a water supply pump start instruction and a valve opening adjustment instruction for irrigation, the method further includes:

[0034] Monitor the actual growth of vegetation and adjust the parameters of the irrigation decision model when there is a deviation between the actual growth of vegetation and the predicted irrigation demand.

[0035] After irrigation is completed, the present invention adjusts the parameters of the irrigation decision model according to the actual growth status of the vegetation monitoring situation, optimizes and adjusts the irrigation decision model to generate a more appropriate irrigation strategy, improve the vegetation coverage rate of the photovoltaic area, and achieve efficient use of water resources and scientific management of vegetation.

[0036] In a second aspect, the present invention provides a photovoltaic area ecological restoration smart irrigation device based on automatic machine learning, the device comprising:

[0037] The acquisition module is used to acquire the environmental monitoring data of the photovoltaic area, which includes soil temperature, humidity and salinity monitoring data, meteorological parameter monitoring data and vegetation monitoring data;

[0038] The data processing module is used to unify the data format of the photovoltaic area environmental monitoring data, normalize abnormal data, clean redundant data, and establish a database;

[0039] The feature processing module is used to perform feature selection, feature extraction, and feature construction on the photovoltaic area environmental monitoring data in the database to establish a feature set;

[0040] The model processing module is used to train, evaluate and manage the intelligent model to obtain the irrigation decision model;

[0041] The irrigation module is used to input the environmental monitoring data of the photovoltaic area into the irrigation decision model. When the output result of the irrigation decision model is that irrigation is needed, a water supply pump start instruction and a valve opening adjustment instruction are generated for irrigation.

[0042] In a third aspect, the present invention provides a computer device comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to execute the smart irrigation method for ecological restoration of photovoltaic areas based on automatic machine learning according to the above-mentioned first aspect or any corresponding embodiment thereof.

[0043] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, and the computer instructions are used to enable a computer to execute the intelligent irrigation method for ecological restoration of photovoltaic areas based on automatic machine learning according to the above-mentioned first aspect or any corresponding embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0045] Figure 1 It is a flow chart of a smart irrigation method for ecological restoration of photovoltaic areas based on automatic machine learning according to an embodiment of the present invention;

[0046] Figure 2 is a schematic diagram of a smart irrigation platform according to an embodiment of the present invention;

[0047] Figure 3 is a structural schematic diagram of a smart irrigation model according to an embodiment of the present invention;

[0048] Figure 4 is a schematic diagram of a workflow of a smart irrigation system according to an embodiment of the present invention;

[0049] Figure 5 is a structural block diagram of a photovoltaic area ecological restoration smart irrigation device based on automatic machine learning according to an embodiment of the present invention;

[0050] Figure 6 It is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0052] According to an embodiment of the present invention, an embodiment of an intelligent irrigation method for ecological restoration of photovoltaic areas based on automatic machine learning is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in an order different from that shown here.

[0053] Automatic machine learning is an automated artificial intelligence technology that automatically builds machine learning models or neural network models based on input data features, which can minimize human intervention and achieve autonomous decision-making of the system. The embodiment of the present invention provides a smart irrigation method for ecological restoration of photovoltaic areas based on automatic machine learning. By integrating multiple monitoring devices, intelligent control technology and automatic machine learning technology, precise irrigation and scientific management in the process of ecological restoration of photovoltaic areas can be achieved, water resource utilization efficiency can be improved, vegetation growth can be promoted, and the balance and sustainable development of the photovoltaic area ecosystem can be achieved.

[0054] In this embodiment, a photovoltaic area ecological restoration smart irrigation method based on automatic machine learning is provided. Figure 1 is a flow chart of a photovoltaic area ecological restoration smart irrigation method based on automatic machine learning according to an embodiment of the present invention, such as Figure 1 As shown, the process includes the following steps:

[0055] Step S101, obtaining photovoltaic area environmental monitoring data.

[0056] In the embodiment of the present invention, the photovoltaic area environmental monitoring data includes soil temperature, humidity and salinity monitoring data, meteorological parameter monitoring data and vegetation monitoring data.

[0057] Specifically, soil temperature, humidity and salinity monitoring: 3 sets of monitoring equipment are set up, each set corresponds to a soil within the monitoring range, and each location is equipped with three layers of temperature, humidity and salt probes. The sensor range and accuracy requirements are humidity monitoring range 0-100%, accuracy 0-50% ±2%, 50%-100% ±5%; temperature monitoring range -50-80℃, accuracy ±0.5℃; conductivity monitoring 10-20000us / cm, accuracy 2%. The equipment includes sensors, data collectors, power supply systems, mounting brackets, etc. The data is transmitted and stored in the computer in the local cabinet and uploaded to the cloud platform at the same time.

[0058] Meteorological parameter monitoring: The measured parameters include air temperature, relative humidity, atmospheric pressure, wind speed, wind direction, rainfall, total radiation, photosynthetically active radiation, and net radiation. The collection frequency is half an hour, equipped with a mounting bracket, including the installation of the foundation and bracket, data transmission and storage to the computer in the local cabinet, and upload to the cloud platform at the same time.

[0059] Vegetation monitoring: A vegetation image monitoring system is used with an industrial camera to take pictures of the restoration area every hour. The acquisition frequency is once an hour during the day, with automatic focus and a resolution of no less than 1080p (1920×1080). The camera is installed on the top of the weather station, with an operating temperature of -40-60°C and an operating humidity of <90%. The data is transmitted and stored in a computer in an on-site cabinet and uploaded to the cloud platform at the same time.

[0060] Various monitoring devices collect data such as soil temperature, humidity, salinity, meteorological parameters and vegetation growth conditions in real time, and transmit them to the computer in the on-site control cabinet.

[0061] Among them, the functions of the local control cabinet include manually controlling the opening and closing of valves, providing power distribution for all equipment, collecting all monitoring data and uploading them to the cloud platform. It is equipped with a 27-inch (16:9) smart screen, which can display monitoring and irrigation data in real time, and a supporting computer for data storage and transmission. The control cabinet is installed outdoors, and the surface of the cabinet is treated with anti-corrosion, which has good airtightness. It is also equipped with cables and network cables: 300 meters of cables and 300 meters of network cables, all of which are buried underground.

[0062] Step S102, the photovoltaic area environmental monitoring data is processed in a unified data format, abnormal data is normalized, and redundant data is cleaned to establish a database.

[0063] In the embodiment of the present invention, the computer in the local control cabinet performs preliminary processing on the collected data and then uploads it to the cloud platform. The data processing platform receives various field monitoring data from the collection equipment, further processes the collected photovoltaic area environmental monitoring data, manages real-time data and historical data through data collection, data cleaning, data storage and other links, and completes data preprocessing. Data cleaning specifically includes unified processing of data format, normalization of abnormal data, and cleaning of redundant data. The processed data is stored in the database to establish a database.

[0064] Step S103, performing feature selection, feature extraction, and feature construction on the photovoltaic area environment monitoring data in the database to establish a feature set.

[0065] In the embodiment of the present invention, the feature processing platform mainly completes the conversion process from collected and cleaned data to feature sets, and further processes the original data, mainly including feature selection, feature extraction, and feature construction. According to the data characteristics and model characteristics, effective features are selected to form feature sets in preparation for the next step of model training. Feature selection is to remove invalid features and retain high-contribution features through a series of methods. Feature extraction is to transform the original features through mathematical methods to obtain more expressive features. Feature construction is to transform the features within the features and combine the features between them.

[0066] Step S104, performing model training, model evaluation and model management on the intelligent model to obtain an irrigation decision model.

[0067] In the embodiment of the present invention, the model processing platform mainly completes the training, evaluation and management of each model to build a smart irrigation decision-making model. Model training includes model space configuration (machine learning model configuration, neural network model configuration) and parameter space configuration (hyperparameter optimization configuration, neural network architecture configuration); model prediction includes cross-validation, integrated validation and evaluation index analysis; model management includes: experimental material management, training log management, model backup management and resource scheduling management.

[0068] Model training selects models for different model spaces and adapts the corresponding hyperparameter optimization and construction search technology. Model evaluation is to evaluate the effect of feature set data after model training. Model management manages various intermediate files and resource scheduling during model training and evaluation.

[0069] Step S105, inputting the photovoltaic area environmental monitoring data into the irrigation decision model, and when the output result of the irrigation decision model is that irrigation is needed, generating a water supply pump start instruction and a valve opening adjustment instruction for irrigation.

[0070] In an embodiment of the present invention, the input parameters of the irrigation decision model include soil temperature and humidity data, meteorological parameter data (temperature, humidity, rainfall, radiation, etc.) and vegetation data, and the output is an irrigation decision, including whether to irrigate, irrigation water volume, etc. Based on the irrigation decision model, the cloud platform combines real-time data to analyze the soil moisture deficit, meteorological conditions and vegetation water demand rules, determine whether irrigation is needed and determine the irrigation water volume. When the output result of the irrigation decision model is that irrigation is needed, that is, when the soil moisture is lower than the set threshold and the meteorological conditions are suitable, the cloud platform sends instructions to the water supply pump and valve control system to automatically start the water pump and adjust the valve opening for irrigation, and the water supply pump and valve control system perform irrigation operations. At the same time, the smart water meter records the irrigation water volume to achieve precise irrigation, and the user can also manually control the irrigation operation through the cloud platform.

[0071] Specifically, the water supply pump and valve control system is equipped with water pumps, valves and control systems to achieve remote automatic control or manual control. An electric control valve is set with adjustable opening and remote control. The working temperature is 0-60℃. A set of smart water meters is set to support cloud transmission and data storage, which is installed on the water inlet pipe.

[0072] In order to achieve precise irrigation, multi-parameter monitoring and intelligent analysis are carried out to avoid waste of water resources and improve the utilization rate of irrigation water.

[0073] The intelligent irrigation method for ecological restoration of photovoltaic areas based on automatic machine learning provided in this embodiment monitors the environment of the photovoltaic area by integrating multiple monitoring devices, preprocesses and features the photovoltaic area environmental monitoring data, and constructs a feature set to prepare data for the irrigation decision model. The irrigation decision model outputs whether irrigation is needed, realizes automatic irrigation in combination with soil and meteorological conditions, and realizes efficient use of water resources and scientific management of vegetation.

[0074] In this embodiment, a photovoltaic area ecological restoration smart irrigation method based on automatic machine learning is provided, and the process includes the following steps:

[0075] Step S201, obtaining photovoltaic area environmental monitoring data.

[0076] For details, please see Figure 1 Step S101 of the illustrated embodiment will not be described in detail here.

[0077] Step S202 , the photovoltaic area environmental monitoring data is processed in a unified data format, abnormal data is normalized, and redundant data is cleaned to establish a database.

[0078] Specifically, the above step S202 includes:

[0079] Step S2021, unifying the data format of the photovoltaic area environmental monitoring data collected by different monitoring devices.

[0080] Step S2022, normalizing the abnormal data of the photovoltaic area environmental monitoring data collected by the abnormal equipment.

[0081] Step S2023, cleaning the compressed, converted and offline stored photovoltaic area environmental monitoring data.

[0082] In the embodiment of the present invention, the data processing platform D mainly completes the data conversion work from the original acquisition device to the control platform to facilitate further data analysis, and specifically includes the following functions:

[0083] D1 Data acquisition. The data acquisition module implements independent acquisition adapter modules for soil monitoring equipment, meteorological monitoring equipment, and vegetation monitoring equipment, and obtains equipment data regularly and regularly through the scheduling module to complete the original data acquisition. According to the number and working status of monitoring equipment in each link, the acquisition adapter module performs concurrent management and multi-threaded scheduling to cope with multi-source, heterogeneous and possible peak data (adjusting the acquisition frequency according to climate conditions or time).

[0084] D2 Data cleaning. The data cleaning module further processes the data obtained by the acquisition module. Specifically, it includes data format unification (time, integer, floating point, string, image, etc.), abnormal data processing, redundant data cleaning and other links. Among them, data format unification is mainly for data format standardization processing due to equipment differences between different monitoring equipment (soil, meteorology, vegetation); abnormal data processing is mainly for abnormal data standardization processing due to possible equipment damage, failure or other extreme situations; redundant data cleaning is mainly for compression, conversion and offline storage of original files that occupy a large space (such as image data to be base64 encoded, offline indexing, etc.).

[0085] The data formats of the photovoltaic area environmental monitoring data collected by different monitoring devices are unified to avoid the situation where the collected data formats are not unified due to differences in monitoring equipment. The abnormal data of the photovoltaic area environmental monitoring data collected by abnormal equipment are normalized to solve the problem of abnormal collected data due to damage, failure or other failure conditions of the monitoring equipment. The compressed, converted and offline stored photovoltaic area environmental monitoring data are cleaned to solve the problems of compression, conversion and offline storage due to the large space occupied by the original files.

[0086] Step S2024, using MySQL as the main database, indexing and format optimization of soil temperature, humidity, salinity and meteorological parameters, encoding and compressing vegetation monitoring data and dividing them into libraries and tables to establish a database.

[0087] In the embodiment of the present invention, the processed data is stored in the database and a database is constructed. The data processing platform D also includes the following functions:

[0088] D3 data storage. After data collection and cleaning, data storage processing must be completed. Data storage involves database write operations, which are sensitive to performance and format, and are prone to abnormal errors or insufficient performance. In this system, data storage completes batch asynchronous writing, abnormal reconnection, slow query detection and other mechanisms.

[0089] D4 database. This system uses MySQL8.0 as the main database for monitoring equipment data management. Among them, the main fields such as soil data and meteorological data are indexed and format optimized, and the vegetation data (mainly images) are encoded, compressed, and divided into databases and tables. According to the different uses of the model, the data output includes offline data and real-time data.

[0090] D41 offline data: According to the frequency of data collection, the data of every other day is stored as historical data to provide data accumulation for feature processing.

[0091] D42 Real-time data. According to the frequency of data collection, the data of the day is stored as real-time data. After each data update, the feature processing platform updates the real-time features for model prediction.

[0092] DF1 offline data feature processing. Offline feature processing is mainly aimed at historical retained data. According to a fixed time period (week, month, quarter, year), a feature set is established for model training and model evaluation.

[0093] DF2 performs feature processing on real-time data. Real-time feature processing mainly targets real-time data and dynamically generates real-time features based on the system initialization strategy or the features required by the optimal model.

[0094] By using MySQL as the main database, soil temperature, humidity, salinity, meteorological parameters and vegetation monitoring data are processed separately to establish a database, thereby preparing data for model training.

[0095] Step S203, performing feature selection, feature extraction, and feature construction on the photovoltaic area environment monitoring data in the database to establish a feature set.

[0096] Specifically, the above step S203 includes:

[0097] Step S2031, performing null value processing and feature importance analysis on the photovoltaic area environmental monitoring data in the database, and performing feature selection.

[0098] Step S2032, using dimensionality reduction to extract the effective subspace of features, and using convolution to extract feature expressions.

[0099] Step S2033, constructing features by transforming fields within features and combining features.

[0100] In the embodiment of the present invention, the feature processing platform F mainly completes the conversion process from collected and cleaned data to feature sets, and specifically includes the following functions:

[0101] F1 feature selection. Feature selection mainly uses a series of methods to remove invalid features and retain high-contribution features. In this system, the original data features are screened through null value processing, feature importance analysis (collinearity) and other links to complete feature selection. Specifically, null value processing is the removal of null value fields in the original data (including row removal and column removal), and feature importance analysis is calculated through collinearity (Pearson correlation coefficient) to remove high collinearity features (such as high collinearity between air humidity and rainfall).

[0102] F2 Feature extraction. Feature extraction is a method of converting original features through mathematical methods (dimensionality reduction, convolution, etc.) to obtain more expressive features. Specifically, dimension reduction is to obtain the effective subspace of the original features through matrix calculation; convolution is to perform time shift or window calculation on speech or images to obtain effective feature expression. In this system, the soil and meteorological data are trained using the PCA method for dimensionality reduction when using the SVM (support vector machine) model. Vegetation monitoring data (images) are convolved using CNN (convolutional neural network) to extract feature expressions.

[0103] F3 Feature construction. Feature construction is a processing method for transforming the fields within a feature and combining the features between them. Field transformation within a feature refers to calculation operations such as logarithmic transformation of large values, feature difference, feature ratio, and feature aggregation; combining features between features is operations such as subsetting and merging features. In this system, when adapting to different model selections, the above feature construction work will be automatically completed.

[0104] F4 feature set. The feature set includes training set, test set, and validation set. Depending on the specific usage scenario (model training or evaluation), the feature set uses different splitting methods and ratios. In this system, the default ratio of training set, test set, and validation set is 7:2:1, and they are randomly shuffled (shuffle). During the model training and evaluation phases, the feature set will be further dynamically processed (cross mode, integration mode, etc.). Depending on the purpose of the model, feature construction includes offline features and real-time features.

[0105] F41 offline features. Offline features are mainly used to build different feature sets for large-capacity historical data for model optimization (training and evaluation).

[0106] F42 Real-time features: Real-time features are mainly based on the system initialization strategy or the features required by the optimal model, dynamically construct feature data, and are directly used for model prediction.

[0107] FM1 offline features are used for model training and model evaluation.

[0108] FS1 real-time features for model prediction.

[0109] By performing feature selection on the photovoltaic area environmental monitoring data in the database to remove invalid features and retain high-contribution features, the features are subjected to dimensionality reduction and convolution processing to obtain more expressive features, and feature construction is performed to adapt to different model selections.

[0110] Step S204, performing model training, model evaluation and model management on the intelligent model to obtain an irrigation decision model.

[0111] Specifically, the above step S204 includes:

[0112] Step S2041, when the intelligent model is a machine learning model, the basic classification model and the integrated learning model are used for model training, and hyperparameter optimization is used for tuning.

[0113] Step S2042: When the intelligent model is a neural network model, the FastAI predefined neural network architecture is used, and PyTorch is used to provide the underlying neural network construction capability and perform automatic training.

[0114] In the embodiment of the present invention, the model processing platform M mainly completes the automated machine learning process of training, evaluating, and managing each model, and specifically includes the following functions:

[0115] M1 model training. Model training selects models for different model spaces and adapts corresponding hyperparameter optimization and architecture search techniques.

[0116] M11 model space. The mainstream model space mainly includes machine learning models and neural network models.

[0117] M111 machine learning model. In this system, the machine learning model mainly selects basic classification models such as logistic regression, decision tree, support vector machine, random forest, etc. for model training. The model call is based on the Scikit-Learn framework. In order to enhance the model training effect, several types of integrated model frameworks are also added for model training, including: XGBoost, LightGBM, CatBoost. Different models and frameworks are called and managed through configuration files. The basic classification model has a fast training speed and low machine performance requirements, and the effect is slightly worse than the integrated model; the integrated model has a slow training speed and high machine performance requirements, and the effect is better.

[0118] M112 neural network model. In this system, the neural network model is mainly generated by calling FastAI and PyTorch interfaces. FastAI provides a variety of predefined neural network architectures, which is convenient for rapid selection according to the characteristics; PyTorch provides the ability to build the underlying neural network, which is convenient for customizing the network structure. Soil data and meteorological data are both tabular structured data. The network structure is mainly based on FCNN (Full Connected Neural Network) layer + nonlinear activation layer (ReLU is used as hidden layer nonlinearity, Sigmoid is used as output layer nonlinearity). Hyperparameters such as the number of model layers and weight dimensions are obtained through automatic training and tuning.

[0119] M12 parameter space. The combination of parameters that can be configured in the automatic machine learning platform.

[0120] M121 hyperparameter optimization. Hyperparameter configuration adapted to the above model selection, such as num_leaves (number of leaf nodes) and max_depth (maximum tree depth) used by tree models, num_layers (number of neural network layers) and num_units (number of neurons) used by neural network models, and learning_rate (learning rate) used by most models. In the process of hyperparameter optimization, there are methods such as grid search and Bayesian search. This system uses grid search to find the optimal hyperparameters.

[0121] M122 network architecture search. Mainly for neural network models, traverse the neural network combination that adapts the data. In this system, the FastAI framework is used to traverse the preset neural network architecture; PyTorch is used to automatically generate an adapted neural network based on the hyperparameter configuration and train it.

[0122] Select machine learning models and neural network models, train the models to select models for different model spaces, and adapt the corresponding hyperparameter optimization and architecture search techniques to improve the performance of the models.

[0123] Step S2043, use cross-validation, integrated validation, and evaluation indicators to evaluate the intelligent model.

[0124] Step S2044, manage the intelligent model using experimental material management, training log management, model backup management, and resource scheduling management.

[0125] In the embodiment of the present invention, the model processing platform M specifically also includes the following functions:

[0126] M2 Model evaluation: Model evaluation is the stage of evaluating the effect of feature set data after model training.

[0127] M21 cross validation. Cross validation performs folding operations on feature subsets, such as K-Fold, and traverses the training and evaluation of feature subsets to achieve the best model generalization effect. This system uses the 5-Fold method.

[0128] M22 Integration verification. Integration verification is to combine trained models to maximize the advantages of different models. Integration methods include gradient boosting integration of tree models and multi-layer merging integration of neural network models. In this system, frameworks such as XGBoost complete gradient boosting integration when called, and frameworks such as FastAI complete neural network integration when called.

[0129] M23 evaluation indicators. Evaluation indicators are mainly used to reflect the model effect. Commonly used evaluation indicators include accuracy, precision, recall, F1-Score, etc. In this system, in addition to the above evaluation indicators, the confusion matrix is ​​also used to calculate sample results such as TP (True Positive), TN (True Negative), FP (False Positive), and FN (False Negative), which are used to analyze the error cases more specifically to adjust the model training method.

[0130] M3 Model Management: The model management module is used to manage various intermediate files and resource scheduling during model training and evaluation.

[0131] M31 Experimental material management. Experimental materials include CheckPoints generated during model training, training configurations, model files, monitoring data, etc. By performing version management on experimental materials, it is convenient to conduct comparative experimental analysis on different training stages and different models.

[0132] M32 training log management. The training log includes the system logs generated during the training process, including system resource monitoring, operation status monitoring, model changes, etc., to facilitate traceability analysis.

[0133] M33 model backup management. Automatic machine learning will generate a large number of model results during the training process, and the model files must be backed up regularly.

[0134] M34 resource scheduling management. Model training has high requirements for hardware resources, and dynamic adjustments need to be made according to system resources during training, including distributed scheduling and GPU management.

[0135] MS1 model update: The optimal model obtained in the model evaluation phase is used to predict the real-time data and obtain irrigation instructions.

[0136] By evaluating the intelligent model, we can improve the generalization ability of the model, utilize the advantages of different models, adjust the model training method, and manage the intelligent model to manage various intermediate files and resource scheduling during the model training and model evaluation process.

[0137] Step S205, inputting the photovoltaic area environmental monitoring data into the irrigation decision model, and when the output result of the irrigation decision model is that irrigation is needed, generating a water supply pump start instruction and a valve opening adjustment instruction for irrigation.

[0138] For details, please see Figure 1 Step S105 of the illustrated embodiment will not be described in detail here.

[0139] In some optional embodiments, the method further comprises:

[0140] Step S206, monitoring the actual growth status of vegetation, and adjusting the parameters of the irrigation decision model when there is a deviation between the actual growth status of vegetation and the predicted irrigation demand.

[0141] In the embodiment of the present invention, after the irrigation is completed, the actual growth of the vegetation is monitored, the soil moisture changes are recorded, the irrigation effect is evaluated, and the model is optimized and adjusted regularly to provide a reference for the next irrigation decision. When there is a deviation between the actual growth of the vegetation and the predicted irrigation demand, that is, the vegetation growth has not improved significantly, the data is re-analyzed, the parameters of the irrigation decision model are adjusted, and the irrigation strategy is adjusted to increase the irrigation water volume or adjust the irrigation frequency to improve the accuracy and adaptability of the model.

[0142] Users can remotely monitor the system operation status and vegetation growth anytime and anywhere through the cloud platform, making it easier to make management decisions. The system can automatically adjust irrigation strategies according to different seasons, climate conditions and vegetation growth stages to adapt to complex and changing environments. Based on big data analysis and smart irrigation models, it provides a scientific basis for irrigation management, promotes healthy vegetation growth, and improves ecological restoration effects.

[0143] The intelligent irrigation method for ecological restoration of photovoltaic areas based on automatic machine learning provided in this embodiment adjusts the parameters of the irrigation decision-making model according to the actual growth status of vegetation monitoring after irrigation is completed, optimizes and adjusts the irrigation decision-making model to generate a more appropriate irrigation strategy, improve the vegetation coverage rate in the photovoltaic area, and achieve efficient use of water resources and scientific management of vegetation.

[0144] like Figure 2 As shown, the smart irrigation platform receives data from meteorological monitoring, vegetation monitoring, and soil monitoring, and automatically controls irrigation equipment.

[0145] like Figure 3 As shown, Figure 3 This is the structure diagram of the smart irrigation model, including the data processing platform D, feature processing platform F, model processing platform M and model service platform.

[0146] The model service platform is used in the model output stage to host the optimal model and provide standardized output, including the following functions:

[0147] S01 Model release. The model release module selects the model with the best effect for loading and packaging, and provides a unified model prediction output format.

[0148] S02 model interface: An online model reasoning interface in the form of RestFul, which is convenient for online access and operation.

[0149] S03 Model SDK. A development library for offline calling, supporting cross-platform calling (Linux, Windows platforms, etc.).

[0150] S04 prediction output. Provide a unified output format. This system uses Json as the model output format.

[0151] As a specific application example of the embodiment of the present invention, Figure 4 As shown in the figure, the workflow of the smart irrigation system includes the following steps:

[0152] 1. Equipment Installation

[0153] According to the actual situation of the project site, the installation locations of monitoring equipment, irrigation equipment and local control cabinets are determined. For example, the probes of soil temperature, humidity and salinity monitoring equipment are buried in the soil in layers according to design requirements, meteorological observation equipment is installed in an open and unobstructed location, and the camera is installed at a suitable angle on the top of the meteorological station.

[0154] Lay cables and network cables underground to ensure reliable connections without external interference. Install water supply pumps and valve control systems, connect pipelines and smart water meters to ensure smooth waterways and flexible control.

[0155] 2. System debugging

[0156] After the equipment is installed, each monitoring device is calibrated and debugged to ensure the accuracy of the collected data. For example, the soil moisture sensor is calibrated on site to keep its measurement error within the specified range.

[0157] Debug water supply pumps and valve control systems, test remote automatic and manual control functions, and check the accuracy of data recording of smart water meters. Perform functional tests on local control cabinets to ensure that they can normally collect data, upload to the cloud platform, and control equipment operation.

[0158] 3. Data Collection and Analysis

[0159] After the system is officially put into operation, data is collected in real time and uploaded to the cloud platform. The cloud platform stores and analyzes the data at set time intervals, such as analyzing meteorological data every half hour and analyzing vegetation growth image data every hour.

[0160] The data is processed according to the smart irrigation model to generate irrigation decision instructions. When the soil moisture is lower than the set threshold and the meteorological conditions are suitable, the irrigation instruction is triggered.

[0161] 4. Irrigation Implementation and Feedback

[0162] After receiving the irrigation instruction, the water supply pump and valve control system execute the irrigation operation, and the smart water meter records the irrigation water volume. During the irrigation process, the irrigation status can be monitored in real time through the local control cabinet or cloud platform.

[0163] After the irrigation is completed, the effect of this irrigation is evaluated based on the vegetation growth monitoring data and soil moisture changes, providing a reference for the next irrigation decision. For example, if the vegetation growth condition has not improved significantly, the irrigation strategy can be adjusted to increase the irrigation water volume or adjust the irrigation frequency.

[0164] In this embodiment, a photovoltaic area ecological restoration smart irrigation device based on automatic machine learning is also provided, which is used to implement the above-mentioned embodiments and preferred implementation modes, and will not be repeated hereafter. As used below, the term "module" can implement a combination of software and / or hardware of a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0165] This embodiment provides a photovoltaic area ecological restoration smart irrigation device based on automatic machine learning, such as Figure 5 As shown, including:

[0166] The acquisition module 501 is used to acquire the photovoltaic area environmental monitoring data, which includes soil temperature, humidity and salinity monitoring data, meteorological parameter monitoring data and vegetation monitoring data.

[0167] The data processing module 502 is used to perform unified data format processing, normalized abnormal data processing, and redundant data cleaning on the photovoltaic area environmental monitoring data to establish a database.

[0168] The feature processing module 503 is used to perform feature selection, feature extraction, and feature construction on the photovoltaic area environment monitoring data in the database to establish a feature set.

[0169] The model processing module 504 is used to perform model training, model evaluation and model management on the intelligent model to obtain an irrigation decision model.

[0170] The irrigation module 505 is used to input the photovoltaic area environmental monitoring data into the irrigation decision model, and when the output result of the irrigation decision model is that irrigation is needed, generate a water supply pump start instruction and a valve opening adjustment instruction for irrigation.

[0171] In some optional implementations, the data processing module 502 includes:

[0172] The first processing unit is used to unify the data format of the photovoltaic area environmental monitoring data collected by different monitoring devices.

[0173] The second processing unit is used to normalize the abnormal data of the photovoltaic area environmental monitoring data collected by the abnormal equipment.

[0174] The third processing unit is used to clean the compressed, converted and offline stored photovoltaic area environmental monitoring data.

[0175] In some optional implementations, the data processing module 502 further includes:

[0176] The database establishment unit is used to use MySQL as the main database, to index and optimize the format of soil temperature, humidity, salinity and meteorological parameters, to encode and compress the vegetation monitoring data, and to divide the data into different libraries and tables to establish a database.

[0177] In some optional implementations, the feature processing module 503 includes:

[0178] The feature selection unit is used to perform null value processing, feature importance analysis and feature selection on the photovoltaic area environmental monitoring data in the database.

[0179] The feature extraction unit is used to extract the effective subspace of features by using dimensionality reduction and to extract feature expressions by using convolution.

[0180] The feature construction unit is used to construct features by transforming fields within features and combining features.

[0181] In some optional implementations, the model processing module 504 includes:

[0182] The first model training unit is used when the intelligent model is a machine learning model. It uses the basic classification model and the integrated learning model to perform model training, and uses hyperparameter optimization to perform tuning.

[0183] The second model training unit, when the intelligent model is a neural network model, adopts the FastAI predefined neural network architecture, uses PyTorch to provide the underlying neural network construction capabilities, and performs automatic training.

[0184] In some optional implementations, the model processing module 504 further includes:

[0185] The model evaluation unit is used to evaluate the intelligent model using cross-validation, integrated validation, and evaluation indicators.

[0186] The model management unit is used to manage the intelligent model through experimental material management, training log management, model backup management, and resource scheduling management.

[0187] In some optional embodiments, the device further comprises:

[0188] The parameter adjustment module is used to monitor the actual growth status of vegetation and adjust the parameters of the irrigation decision model when there is a deviation between the actual growth status of vegetation and the predicted irrigation demand.

[0189] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0190] The photovoltaic area ecological restoration intelligent irrigation device based on automatic machine learning in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0191] The embodiment of the present invention also provides a computer device having the above Figure 5 The figure shows the intelligent irrigation device for ecological restoration of photovoltaic areas based on automatic machine learning.

[0192] See also Figure 6 , Figure 6 is a schematic diagram of the structure of a computer device provided by an optional embodiment of the present invention, such as Figure 6 As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components are connected to each other using different buses for communication, and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 6 A processor 10 is taken as an example.

[0193] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0194] The memory 20 stores instructions executable by at least one processor 10, so that at least one processor 10 executes the method shown in the above embodiment.

[0195] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0196] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 20 may also include a combination of the above types of memory.

[0197] The computer device also includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30 and the output device 40 may be connected via a bus or other means. Figure 6 The example of connecting through bus is taken in the following.

[0198] The input device 30 can receive input digital or character information and generate key signal input related to user settings and function control of the computer device, such as a touch screen, etc. The output device 40 can include a display device, etc.

[0199] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0200] A part of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the existence of the computer program instruction in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc., and accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium accessible to the computer.

[0201] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations all fall within the scope of the present application.

Claims

1. A photovoltaic area ecological restoration smart irrigation method based on automatic machine learning, characterized in that: The method comprises: Acquire photovoltaic area environmental monitoring data, wherein the photovoltaic area environmental monitoring data includes soil temperature, humidity and salinity monitoring data, meteorological parameter monitoring data, and vegetation monitoring data; The photovoltaic area environmental monitoring data is processed in a unified data format, abnormal data is normalized, and redundant data is cleaned to establish a database; Perform feature selection, feature extraction, and feature construction on the photovoltaic area environmental monitoring data in the database to establish a feature set; Conduct model training, model evaluation and model management on the intelligent model to obtain the irrigation decision model; The photovoltaic area environmental monitoring data is input into the irrigation decision model. When the output result of the irrigation decision model is that irrigation is needed, a water supply pump start instruction and a valve opening adjustment instruction are generated to carry out irrigation.

2. The method according to claim 1, characterized in that The step of processing the photovoltaic area environmental monitoring data in a unified data format, normalizing abnormal data, and cleaning redundant data includes: Unify the data format of the photovoltaic area environmental monitoring data collected by different monitoring equipment; Normalize abnormal data of PV area environmental monitoring data collected by abnormal equipment; The compressed, converted and offline stored photovoltaic area environmental monitoring data are cleaned.

3. The method according to claim 1, characterized in that The establishing of the database comprises: MySQL is used as the main database, the soil temperature, humidity, salinity and meteorological parameters are indexed and format optimized, the vegetation monitoring data is encoded, compressed and divided into libraries and tables to establish a database.

4. The method according to claim 1, characterized in that: The feature selection, feature extraction and feature construction of the photovoltaic area environmental monitoring data in the database include: Perform null value processing, feature importance analysis, and feature selection on the photovoltaic area environmental monitoring data in the database; Dimensionality reduction is used to extract the effective subspace of features, and convolution is used to extract feature expressions; Feature construction is performed by transforming fields within features and combining features.

5. The method according to claim 1, characterized in that: The intelligent model includes a machine learning model and a neural network model, and the model training of the intelligent model includes: When the intelligent model is a machine learning model, the basic classification model and the integrated learning model are used for model training, and hyperparameter optimization is used for tuning; When the intelligent model is a neural network model, FastAI pre-defined neural network architecture is adopted, PyTorch is used to provide the underlying neural network construction capability, and automatic training is performed.

6. The method according to claim 1, characterized in that Conduct model evaluation and model management for intelligent models, including: Use cross-validation, integrated validation, and evaluation indicators to evaluate the intelligent model; The intelligent model is managed through experimental material management, training log management, model backup management, and resource scheduling management.

7. The method according to claim 1, characterized in that After generating a water supply pump start instruction and a valve opening adjustment instruction for irrigation, the method further includes: The actual growth status of vegetation is monitored, and when there is a deviation between the actual growth status of vegetation and the predicted irrigation demand, the parameters of the irrigation decision model are adjusted.

8. A photovoltaic area ecological restoration smart irrigation device based on automatic machine learning, characterized in that: The device comprises: An acquisition module is used to acquire photovoltaic area environmental monitoring data, wherein the photovoltaic area environmental monitoring data includes soil temperature, humidity and salinity monitoring data, meteorological parameter monitoring data and vegetation monitoring data; A data processing module is used to perform unified data format processing, normalized abnormal data processing, and redundant data cleaning on the photovoltaic area environmental monitoring data to establish a database; The feature processing module is used to perform feature selection, feature extraction, and feature construction on the photovoltaic area environmental monitoring data in the database to establish a feature set; The model processing module is used to train, evaluate and manage the intelligent model to obtain the irrigation decision model; The irrigation module is used to input the photovoltaic area environmental monitoring data into the irrigation decision model, and when the output result of the irrigation decision model is that irrigation is needed, generate a water supply pump start instruction and a valve opening adjustment instruction for irrigation.

9. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the intelligent irrigation method for ecological restoration of photovoltaic areas based on automatic machine learning as described in any one of claims 1 to 7 by executing the computer instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the intelligent irrigation method for ecological restoration of photovoltaic areas based on automatic machine learning as described in any one of claims 1 to 7.

Images