Saline-alkali soil desalination improvement system and method based on brackish water recycling

Abstract

The invention relates to the technical field of agricultural engineering and ecological environment protection, in particular to a saline-alkali soil desalination improvement system and method based on brackish water recycling. The system comprises a data collection module which collects soil salinity data, meteorological data and crop growth data of the saline-alkali land in real time; the data processing and analyzing module is used for receiving and preprocessing the soil salinity data, the meteorological data and the crop growth data of the saline-alkali land, constructing a soil salinity prediction model and outputting a soil salinity prediction result; the irrigation strategy decision module is used for generating an optimal irrigation strategy by adopting a reinforcement learning algorithm according to the soil salinity prediction result and the preprocessed crop growth data; and the irrigation control system is used for performing brackish water irrigation operation on the saline-alkali soil according to the optimal irrigation strategy. Precise regulation and control of saline-alkali soil salinity can be achieved, brackish water is adopted as irrigation water, water resources are saved to a certain degree, and therefore the improvement cost of saline-alkali soil is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of agricultural engineering and ecological environment protection, and particularly relates to a saline-alkali soil desalination and improvement system and method based on micro-saline water recycling. BACKGROUND

[0002] As a widely distributed low-yield field type in the world, the improvement and utilization of saline-alkali soil is an important research direction to ensure food security and promote sustainable agricultural development. At present, the global saline-alkali soil area is more than 950 million hectares, and the total area of saline-alkali soil in China has reached about 1.5 billion mu. Traditional improvement technologies such as fresh water irrigation, chemical improvement agent improvement, and biological improvement have many drawbacks. Fresh water irrigation consumes a large amount of water, chemical improvement pollutes the soil, and biological improvement has a long cycle. With the increasing scarcity of fresh water resources, the use of micro-saline water for saline-alkali soil improvement has become a research hotspot. However, the existing micro-saline water improvement technology lacks precise control of soil salt dynamics and intelligent irrigation strategies based on crop characteristics, resulting in soil water-salt imbalance and difficulty in meeting the differentiated needs of different crops at different growth stages.

[0003] At present, some micro-saline water improvement technologies simply use fixed thresholds to judge the soil salt state, lack in-depth analysis of the spatial and temporal variation of soil salt, and cannot accurately predict the salt evolution trend. The irrigation strategy is mostly empirical or fixed mode, without fully considering the differences and dynamic changes of crop salt tolerance, making it difficult to achieve precise irrigation. For example, some systems use regular and quantitative irrigation, which cannot adjust the irrigation scheme in time when the soil salt fluctuates or the crop needs water at the critical period, resulting in the aggravation of soil salinization or the impact on crop growth. Therefore, there is an urgent need to develop a saline-alkali soil micro-saline water improvement system and method that can accurately perceive the spatial and temporal variation of salt and intelligently adapt to crop needs. SUMMARY

[0004] In order to solve the problems existing in the prior art, the present application provides a saline-alkali soil desalination and improvement system based on micro-saline water recycling, which comprises: a data collection module for real-time collection of soil salt data, meteorological data and crop growth data of saline-alkali soil; a data processing and analysis module for receiving and preprocessing the soil salt data, meteorological data and crop growth data of saline-alkali soil, and constructing a soil salt prediction model to output soil salt prediction results; an irrigation strategy decision module for generating an optimal irrigation strategy by using a reinforcement learning algorithm according to the soil salt prediction results and the preprocessed crop growth data; and an irrigation control system for micro-saline water irrigation operation of saline-alkali soil according to the optimal irrigation strategy. The present application can realize precise regulation and control of saline-alkali soil salt, and use micro-saline water as irrigation water to a certain extent to save water resources, thereby reducing the improvement cost of saline-alkali soil.

[0005] The application adopts the technical solutions below: a saline-alkali soil desalination and improvement system based on micro-saline water recycling, which comprises a data collection module, a data processing and analysis module, an irrigation strategy decision module and an irrigation control system; The data collection module is used for collecting soil salt data, meteorological data and crop growth data of the saline-alkali soil in real time and sending them to the data processing and analysis module; The data processing and analysis module is used for receiving and preprocessing the soil salt data, meteorological data and crop growth data of the saline-alkali soil, constructing a soil salt prediction model according to the preprocessed soil salt data, meteorological data and crop growth data and outputting a soil salt prediction result of the saline-alkali soil; The irrigation strategy decision module is used for generating an optimal irrigation strategy by using a reinforcement learning algorithm according to the soil salt prediction result output by the data processing and analysis module and the preprocessed crop growth data; The irrigation control system is used for micro-saline water irrigation of the saline-alkali soil according to the optimal irrigation strategy generated by the irrigation strategy decision module.

[0006] Further, the saline-alkali soil desalination and improvement system further comprises a water recycling module; The water recycling module comprises a drainage collection system, a water quality purification unit and a water storage and adjustment device; The drainage collection system collects micro-saline water seeping to the underground of the saline-alkali soil after irrigation by laying a drainage pipe network underground; The water quality purification unit is used for purifying the micro-saline water collected by the drainage collection system; The water storage and adjustment device is used for storing the purified micro-saline water and providing the irrigation control system with micro-saline water for irrigation at the next irrigation.

[0007] Further, the data collection module comprises a salt-tolerant characteristic data collection module and a soil salt monitoring module; The salt-tolerant characteristic data collection module is used for collecting crop growth data in the saline-alkali soil in real time; The soil salt monitoring module is used for collecting soil salt data and meteorological data in the saline-alkali soil in real time; Further, the method for receiving and preprocessing the soil salt data, meteorological data and crop growth data of the saline-alkali soil by the data processing and analysis module comprises: standardizing the soil salt data, meteorological data and crop growth data of the saline-alkali soil; constructing a multi-dimensional feature vector according to the standardized soil salt data, meteorological data and crop growth data; performing time-space correlation processing, heterogeneous data processing and abnormal data detection operation on the multi-dimensional feature vector; A soil salinity prediction model is established by using an LSTM network based on an attention mechanism, and a multi-dimensional feature vector is input into the soil salinity prediction model for training. A soil salinity prediction result of the saline-alkali soil is output according to the trained soil salinity prediction model.

[0008] Further, the irrigation strategy decision module generates an optimal irrigation strategy by using a reinforcement learning algorithm, and specifically: The preprocessed crop growth data and the soil salinity prediction result output by the data processing and analysis module are received to construct a state vector. A plurality of irrigation operation combinations are established according to executable irrigation operations. The constructed state vector and the plurality of irrigation operation combinations are input into a decision model established based on a DQN algorithm calculation to calculate Q values of each irrigation operation corresponding to the state vector. The irrigation operation combination with the maximum Q value is taken as the optimal irrigation strategy corresponding to the current state vector.

[0009] Further, the executable irrigation operations include an irrigation amount, an irrigation frequency, and an irrigation time.

[0010] Further, the irrigation strategy decision module further includes: The optimal irrigation strategy corresponding to the current state vector is used to control the irrigation control system to irrigate the saline-alkali soil. The score of the optimal irrigation strategy corresponding to the current state vector is calculated according to the re-collected soil salinity data, weather data, and crop growth data in the saline-alkali soil after irrigation. If the score of the optimal irrigation strategy corresponding to the current state vector is less than zero more than a set threshold, the decision model is retrained.

[0011] Further, the irrigation control system converts the optimal irrigation strategy transmitted by the irrigation strategy decision module into a control instruction, and controls the water pump and the electromagnetic valve to work through the control instruction; specifically, the flow rate and the pressure of the water pump are controlled according to the control instruction, and the electromagnetic valve is controlled to be opened or closed according to the control instruction.

[0012] The application further provides a saline-alkali soil desalination and improvement method based on micro-saline water recycling, which is applied to the saline-alkali soil desalination and improvement system based on micro-saline water recycling, and the method comprises the following steps: Real-time acquisition of soil salinity data, weather data, and crop growth data of the saline-alkali soil is performed. The soil salinity data, weather data, and crop growth data of the saline-alkali soil are preprocessed. A soil salinity prediction model is constructed according to the preprocessed soil salinity data, weather data, and crop growth data, and a soil salinity prediction result of the saline-alkali soil is output. According to the soil salt content prediction result and the pretreated crop growth data, an optimal irrigation strategy is generated by using a reinforcement learning algorithm; According to the optimal irrigation strategy generated by the irrigation strategy decision module, the saline-alkali soil is subjected to brackish water irrigation operation.

[0013] The salt-alkali soil desalination and improvement system provided by the present application can realize intelligentization of the whole process from soil salt content monitoring and prediction to irrigation strategy decision and execution, the data processing and analysis module in the system adopts an LSTM network based on an attention mechanism to predict soil salt content, which can accurately capture the change law of soil salt content in the time and space dimensions, thereby improving the accuracy of soil salt content prediction, providing accurate input for the subsequent irrigation strategy decision module, and enabling the irrigation strategy obtained by using the DQN algorithm to accurately regulate and control soil salt content, effectively avoiding soil water-salt imbalance, reducing the dependence on fresh water resources during irrigation, avoiding the use of fresh water and chemical agents, and further improving the utilization efficiency of water resources by using the recycling module during irrigation, thereby greatly reducing the improvement cost. BRIEF DESCRIPTION OF DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0015] Figure 1 The structure schematic diagram of the salt-alkali soil desalination and improvement system based on brackish water recycling according to the embodiments of the present application is shown in Figure 2 The flowchart of the salt-alkali soil desalination and improvement method based on brackish water recycling according to the embodiments of the present application is shown in DETAILED DESCRIPTION

[0016] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments only constitute some embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0017] The structure schematic diagram of the salt-alkali soil desalination and improvement system based on brackish water recycling according to the embodiments of the present application is shown in Figure 1 The structure schematic diagram of the salt-alkali soil desalination and improvement system based on brackish water recycling according to the embodiments of the present application is shown in The data collection module is configured to collect soil salinity data, meteorological data and crop growth data in real time and send the data to the data processing and analysis module. In the embodiment of the present application, the data collection module comprises a salt-tolerant characteristic data collection module and a soil salinity monitoring module.

[0018] Specifically, the data collection module comprises a plurality of devices for collecting different data of the saline-alkali soil.

[0019] The data processing and analysis module is configured to receive and pre-process the soil salinity data, meteorological data and crop growth data of the saline-alkali soil, construct a soil salinity prediction model according to the pre-processed soil salinity data, meteorological data and crop growth data, and output a soil salinity prediction result of the saline-alkali soil. In the embodiment of the present application, the data processing and analysis module receives the soil salinity data, meteorological data and crop growth data collected by the data collection module through the LoRa wireless communication network, and uses the related algorithms and models built-in the module to preprocess the collected data, including: standardizing the soil salinity data, meteorological data and crop growth data of the saline-alkali soil; constructing a multi-dimensional feature vector according to the standardized soil salinity data, meteorological data and crop growth data; performing operations including spatio-temporal correlation processing, heterogeneous data processing and abnormal data detection on the multi-dimensional feature vector; using an LSTM network based on an attention mechanism to establish a soil salinity prediction model, inputting the multi-dimensional feature vector into the soil salinity prediction model for training; and outputting the soil salinity prediction result of the saline-alkali soil according to the trained soil salinity prediction model.

[0020] Specifically, due to the large difference in the dimension and value range of the soil salinity data, meteorological data and crop growth data, a standardization method is needed to unify the data scale. In the embodiment of the present application, the soil salinity data and meteorological data of the saline-alkali soil are standardized using the Z-score standardization formula, and the crop growth data is converted using the one-hot encoding method, which encodes the crop growth data into a binary vector, so that different types of data are initially comparable.

[0021] According to the standardized soil salinity data, meteorological data and crop growth data, a multi-dimensional feature vector is constructed, and the specific method is: in the embodiment of the present application, a multi-dimensional feature vector is constructed, so as to integrate different soil layer concentrations and change rates of the soil salinity data, rainfall, evaporation, air temperature, humidity in the meteorological data, and growth stage, plant height, leaf area index and other features of the crop growth data. After constructing the multi-dimensional feature vector, principal component analysis method is used to reduce the dimension of the multi-dimensional feature vector, so as to reduce feature redundancy and improve data processing efficiency on the premise of retaining more than 95% of data information.

[0022] The multi-dimensional feature vector is subjected to operations including spatio-temporal correlation processing, heterogeneous data processing and abnormal data detection, specifically: The spatio-temporal correlation operation is a process of correlating the spatial position information of the soil salt content data with the meteorological data and the crop growth data by using a geographic information system (GIS) technology. For the meteorological data, according to the latitude and longitude of the saline-alkali land, the meteorological station data can be mapped to each salt content data collection point arranged in the saline-alkali land through a spatial difference algorithm. For the crop growth data, the crop is matched with the soil salt content data and the meteorological data in the corresponding planting area of the crop through division of the crop planting area. Meanwhile, the data at the same time point are aligned and combined with the time sequence as an axis, so that the multi-dimensional feature vector is processed into a fusion data vector with unified spatio-temporal information, thereby providing comprehensive input data for subsequent algorithms.

[0023] The heterogeneous data processing is to establish a unified data interface specification processing process for the differences in data structure and format. For example, the soil salt content data collected by the sensor is usually binary data, the meteorological data output by the meteorological station is in the form of words, and the processed growth data is in the form of an encoded binary vector. In the embodiment of the present application, the above data is converted into a unified data format through heterogeneous data processing, and the optional data format is CSV or Parquet. A data dictionary is defined to ensure the meaning of each data field and the consistency of data from different sources at the semantic level, thereby eliminating the obstacles caused by data heterogeneity.

[0024] The abnormal data detection is an operation of removing abnormal data from each data in the multi-dimensional feature vector. In the embodiment of the present application, an integrated learning abnormal detection method is adopted, which combines the isolation forest algorithm, the local outlier factor and the density peak value clustering method to detect abnormal data of the soil salt content data, the meteorological data and the crop growth data. Specifically, for the soil salt content data, if it is detected that the salt content concentration of a certain monitoring point arranged appears a sharp fluctuation in a short time and exceeds the range of 3 times the standard deviation of the historical data, the soil salt content data monitored by the point is marked as suspected abnormal. For the meteorological data, when the rainfall, air temperature and other indicators appear data that do not conform to the seasonal rule or exceed the normal value range, the abnormal data is also marked. For the crop growth data, if the growth indicators such as plant height and leaf area index appear unreasonable mutation, the data is regarded as abnormal data. After all the abnormal data is marked and removed, a multi-dimensional feature vector containing more accurate data can be obtained.

[0025] The LSTM network based on the attention mechanism is adopted to establish a soil salt content prediction model, and the multi-dimensional feature vector is input into the soil salt content prediction model for training. In the embodiment of the present application, a deep learning model combining long short-term memory network (LSTM) and attention mechanism is used for dynamic prediction of soil salinity, a neural network model containing multiple LSTM layers is constructed, each LSTM layer contains multiple memory cells, the memory cells control the flow of information through the forget gate, input gate and output gate, and can effectively capture the long-term dependence relationship in the time series data of soil salinity, and a multi-dimensional feature vector is used as the input of the LSTM network, and the network learns the law of change of soil salinity with time and space in the training process; in the embodiment of the present application, an attention mechanism is further introduced in the output layer of the LSTM network, the attention mechanism calculates the weight of each time step in the input sequence, and highlights the time step and feature that has greater influence on the prediction of soil salinity.

[0026] In the specific calculation process, the attention score of each time step is calculated according to the output of the LSTM network hidden layer, for example, the dot product attention calculation method is used: Among them, represents the correlation degree score between time step i and time step j when calculating attention, which is a scalar value; and are the hidden states of the LSTM network time step i and time step j respectively, is a trainable weight matrix, is the dimension of the hidden state, which is a scaling factor (to prevent gradient disappearance), the attention score is normalized by the softmax function to obtain the attention weight: Among them, is the attention weight obtained after normalization by the softmax function, which represents the relative importance of time step j to time step i in all time steps, The value of is between 0 and 1, and for each time step i, the sum of the attention weights of all time steps j is 1, that is, ; T is the total number of time steps, and the exponential function is used to convert the attention score into a positive number, because the input of the softmax function usually requires a positive number, which can better reflect the difference between different scores; reflects the dependence of time step i on each time step k in the sequence when generating the output, when represents the attention of time step i to j, when k traverses all time steps, The attention weight is used for normalizing all possible association scores, and finally, the hidden states of the LSTM network are weighted and summed according to the attention weight to obtain an output containing attention information, which is used for dynamic prediction of soil salinity.

[0027] Through the cooperation of the above algorithms, the data processing and analysis module can realize accurate processing and accurate prediction of soil salinity data, and provide solid data and algorithm support for saline-alkali soil slightly saline water irrigation regulation.

[0028] The irrigation strategy decision module is used for generating an optimal irrigation strategy by using a reinforcement learning algorithm according to the soil salinity prediction result output by the data processing and analysis module and the preprocessed crop growth data. In the embodiment of the present application, the irrigation strategy decision module receives the preprocessed crop growth data and the soil salinity prediction result output by the data processing and analysis module, and constructs a state vector; a plurality of irrigation operation combinations are established according to executable irrigation operations; the constructed state vector and the plurality of irrigation operation combinations are input into a decision model established based on a DQN algorithm, and the Q value of each irrigation operation corresponding to the state vector is calculated; and the irrigation operation combination with the maximum Q value is taken as the optimal irrigation strategy corresponding to the current state vector.

[0029] Specifically, the state vector constructed in the embodiment of the present application is used to describe the current environmental information of the saline-alkali land, and the soil salinity prediction result includes the salinity concentration values of the surface layer, the middle layer and the bottom layer of the saline-alkali land soil, and the deviation value of each layer of soil salinity concentration from the appropriate salinity for the corresponding crop growth stage can be obtained, for example, for the wheat in the seedling stage, the appropriate soil surface salinity range is 0.8%-1.2%, and if the current surface salinity concentration is 1.3%, the deviation value is 0.1%, and all these data can be used as state variables. In addition, the change rate of the salinity concentration of different soil layers, such as the change amount of the soil surface salinity concentration in the past 1 hour, is also included, which is used to reflect the dynamic trend of soil salinity. The preprocessed crop growth state includes the current growth stage of the crop in the saline-alkali land, such as the germination stage, the seedling stage, the growth stage and the mature stage, and the growth indicators of the crop, such as the plant height and the leaf area index, and different growth stages and growth indicators correspond to different salt tolerance, thereby affecting the decision of the irrigation strategy; all the above data are integrated into the state vector to completely describe the current saline-alkali land environmental state, thereby providing a basis for decision-making.

[0030] The executable irrigation operation refers to all the irrigation methods that can be selected in the irrigation strategy, which mainly includes the irrigation amount, the irrigation frequency and the irrigation time; wherein the irrigation amount: a plurality of discrete irrigation amount levels can be set, such as 5m 3 / acre, 10m 3 / acre, 15m 3 / mu, 20m 3 / mu, etc. to meet the irrigation needs of different soil water and salt conditions; irrigation frequency: including 1 time per day, 2 times per day, 1 time every two days, 2 times per week, etc. Different irrigation time interval options; irrigation time: divided into morning (6-12), afternoon (12-18), evening (18-24) three time periods, taking into account the differences in water evaporation, soil water infiltration and crop absorption under different irrigation times. Through different combinations of irrigation amount, irrigation frequency and irrigation time, a rich irrigation operation combination can be formed, so as to select the appropriate action to perform the irrigation operation.

[0031] The constructed state vector and the plurality of irrigation operation combinations are input into the decision model calculated and established based on the DQN algorithm, and the Q value of each irrigation operation corresponding to the state vector is calculated; the irrigation operation combination with the maximum Q value is taken as the optimal irrigation strategy corresponding to the current state vector.

[0032] In the embodiment of the present application, the decision model calculated and established based on the DQN algorithm can be applied to the present application by referring to any one of the prior art, and the principle is to calculate the maximum Q value for each irrigation operation combination through the input state vector, so as to output the irrigation operation combination corresponding to the maximum Q value as the optimal irrigation strategy; in another embodiment of the present application, the decision model can also be established by using a policy gradient algorithm, and the establishment method is disclosed in the prior art, which will not be described here.

[0033] For the established decision model, the optimal irrigation strategy output by the embodiment of the present application is used to control the irrigation control system to irrigate the saline-alkali soil; according to the re-collected soil salt data, meteorological data and crop growth data in the saline-alkali soil after irrigation, the score of the optimal irrigation strategy corresponding to the current state vector is calculated; if the number of times when the score of the optimal irrigation strategy corresponding to the current state vector is less than zero exceeds the set threshold, the decision model is retrained.

[0034] Specifically, the embodiment of the present application designs a reward function to evaluate the accuracy of the decision model, and considers factors such as soil salt control, crop growth demand and irrigation cost in the design, which includes the following contents: Soil salt control reward: After irrigating the saline-alkali soil, the soil salt data in the saline-alkali soil is collected again. When the salt concentration in each layer of the soil salt data is maintained within the appropriate range for the corresponding growth stage of the crops, a positive reward is given. For example, if the salt concentration of the top layer of soil during the wheat seedling stage is between 0.8% and 1.2%, a reward of +5 is given for every 1 hour of maintenance. If the salt concentration exceeds the appropriate range, a negative reward of -10 is given for every 1% deviation, prompting the adjustment of the irrigation strategy to restore the salt balance. At the same time, an additional reward is given for the trend of the salt concentration approaching the appropriate range. Crop growth reward: rewards are given according to the changes in crop growth indicators, such as the growth of crop height, leaf area index, etc. If the growth is as expected, a reward of +8 is given. If the growth is inhibited, such as yellowing of leaves, growth stagnation, etc., a negative reward of -15 is given. Irrigation cost reward: to avoid excessive irrigation and waste of water resources and increase in cost, an irrigation cost penalty mechanism is introduced. According to the actual amount of brackish water consumed and the cost, a reward of -2 is given for every 1 m 3 of water consumed if it exceeds the reasonable irrigation amount range. If it is within the reasonable range, a positive reward of +3 is given to optimize the irrigation water amount while ensuring soil salt and crop growth.

[0035] According to the content of the above reward function, if the score of the optimal irrigation strategy calculated in a period of time is less than zero more than a set threshold, it is considered that the accuracy of the current decision model is not enough, and data needs to be collected for training.

[0036] An irrigation control system for irrigating the saline-alkali soil with brackish water according to the optimal irrigation strategy generated by the irrigation strategy decision module.

[0037] The irrigation control system receives the optimal irrigation strategy generated by the irrigation strategy decision model and converts the strategy into specific irrigation instructions to control the irrigation equipment to work, for example, the optimal irrigation strategy gives an irrigation amount of 10 m 3 / acre, an irrigation frequency of 1 time per day, and an irrigation time in the afternoon. It is converted into specific irrigation instructions, i.e. adjusting the flow rate and pressure of the water pump, and starting the electromagnetic valve to begin irrigating brackish water in the afternoon and controlling the irrigation time.

[0038] In another specific embodiment of the present application, the present application also proposes a method applied to the above-mentioned saline-alkali soil desalination and improvement system based on the recycling of brackish water, which comprises: Real-time collection of soil salt data, weather data and crop growth data of the saline-alkali soil; The soil salt data, the meteorological data and the crop growth data of the saline-alkali soil are preprocessed, a soil salt prediction model is constructed according to the preprocessed soil salt data, the meteorological data and the crop growth data, and a soil salt prediction result of the saline-alkali soil is output; According to the soil salt prediction result and the preprocessed crop growth data, an optimal irrigation strategy is generated by using a reinforcement learning algorithm; According to the optimal irrigation strategy generated by the irrigation strategy decision module, a brackish water irrigation operation is performed on the saline-alkali soil.

[0039] The above merely describes the preferred embodiments of the present application and is not used to limit the present application, and any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A saline-alkali soil desalination and improvement system based on brackish water recycling, characterized in that, The system comprises a data collection module, a data processing and analysis module, an irrigation strategy decision module, and an irrigation control system. The data collection module is configured to collect soil salinity data, meteorological data, and crop growth data of the saline-alkali soil in real time and send them to the data processing and analysis module. The data processing and analysis module is configured to receive and preprocess the soil salinity data, meteorological data, and crop growth data of the saline-alkali soil, construct a soil salinity prediction model based on the preprocessed data, and output the soil salinity prediction result of the saline-alkali soil. The irrigation strategy decision module is configured to generate an optimal irrigation strategy based on the soil salinity prediction result output by the data processing and analysis module and the preprocessed crop growth data using a reinforcement learning algorithm. The irrigation control system is configured to perform micro-saline water irrigation on the saline-alkali soil based on the optimal irrigation strategy generated by the irrigation strategy decision module. The saline-alkali soil desalination and improvement system further comprises a water recycling module.

2. The system for saline soil desalination and reclamation based on brackish water recycling according to claim 1, characterized in that: The water recycling module comprises a drainage collection system, a water quality purification unit, and a water storage and adjustment device. The drainage collection system collects micro-saline water that seeps into the ground after irrigation of the saline-alkali soil by laying a drainage pipe network underground. The water quality purification unit is configured to purify the micro-saline water collected by the drainage collection system. The water storage and adjustment device is configured to store the purified micro-saline water and provide it for irrigation by the irrigation control system during the next irrigation. The data collection module comprises a salt-tolerant feature data collection module and a soil salinity monitoring module.

3. The system for saline soil desalination and reclamation based on brackish water recycling according to claim 1, characterized in that: The salt-tolerant feature data collection module is configured to collect crop growth data in real time. The soil salinity monitoring module is configured to collect soil salinity data and meteorological data in real time. The data processing and analysis module receives and preprocesses the soil salinity data, meteorological data, and crop growth data of the saline-alkali soil by:

4. The system for saline soil desalination and reclamation through brackish water recycling according to claim 1, characterized in that: standardizing the soil salinity data, meteorological data, and crop growth data of the saline-alkali soil; constructing a multi-dimensional feature vector based on the standardized soil salinity data, meteorological data, and crop growth data; performing spatio-temporal correlation processing, heterogeneous data processing, and abnormal data detection operations on the multi-dimensional feature vector; establishing a soil salinity prediction model based on an attention mechanism LSTM network and inputting the multi-dimensional feature vector into the soil salinity prediction model for training; outputting the soil salinity prediction result of the saline-alkali soil based on the trained soil salinity prediction model. The irrigation strategy decision module generates an optimal irrigation strategy using a reinforcement learning algorithm by:

5. The system for saline soil desalination and reclamation using brackish water recycling according to claim 1, wherein: receiving the preprocessed crop growth data and the soil salinity prediction result output by the data processing and analysis module, and constructing a state vector; establishing multiple irrigation operation combinations based on executable irrigation operations; inputting the constructed state vector and multiple irrigation operation combinations into a decision model established based on a DQN algorithm calculation, and calculating the Q value of each irrigation operation corresponding to the state vector; selecting the irrigation operation combination with the maximum Q value as the optimal irrigation strategy corresponding to the current state vector. ​ 6. The system for saline soil desalination and reclamation through brackish water recycling according to claim 4, characterized in that: The executable irrigation operation includes an irrigation amount, an irrigation frequency, and an irrigation time.

7. The system for saline soil desalination and reclamation through brackish water recycling according to claim 4, characterized in that: The irrigation strategy decision module further includes: According to the optimal irrigation strategy corresponding to the current state vector, the irrigation control system controls the irrigation of the saline-alkali soil; According to the re-acquired soil salt data, meteorological data and crop growth data in the saline-alkali soil after irrigation, the score of the optimal irrigation strategy corresponding to the current state vector is calculated. If the number of times that the score of the optimal irrigation strategy corresponding to the current state vector is less than zero exceeds a set threshold, the decision model is re-trained.

8. The system for saline soil desalination and reclamation using brackish water recycling according to claim 1, wherein: The irrigation control system converts the optimal irrigation strategy transmitted by the irrigation strategy decision module into control instructions, and controls the water pump and the electromagnetic valve to work through the control instructions; specifically, the flow rate and pressure of the water pump are controlled according to the control instructions; and the electromagnetic valve is controlled to be opened or closed according to the control instructions.

9. The method for improving saline-alkali soil desalination according to the brackish water recycling, applied to the system for improving saline-alkali soil desalination according to the brackish water recycling of any one of claims 1 to 8, characterized in that, It includes: Real-time acquisition of soil salt data, meteorological data and crop growth data of the saline-alkali soil; Pretreatment of the soil salt data, meteorological data and crop growth data of the saline-alkali soil, According to the pretreated soil salt data, meteorological data and crop growth data, a soil salt prediction model is constructed, and a soil salt prediction result of the saline-alkali soil is outputted; According to the soil salt prediction result and the pretreated crop growth data, an optimal irrigation strategy is generated by using a reinforcement learning algorithm; According to the optimal irrigation strategy generated by the irrigation strategy decision module, the saline-alkali soil is irrigated with brackish water.

Images