Plateau multi-energy complementary agricultural energy management system and method

Abstract

The invention discloses a plateau multi-energy complementary agricultural energy management system and a plateau multi-energy complementary agricultural energy management method, and belongs to the technical field of new energy. Wind, light and terrestrial heat multi-energy complementation is cooperated with a battery energy storage system and a phase change heat storage system, and a central intelligent management system is used for carrying out priority scheduling and real-time monitoring and prediction on a load; efficient energy supply of irrigation, processing and greenhouse heating is achieved, the renewable energy utilization rate and the load satisfaction degree are remarkably improved, the operation cost and environmental pollution are reduced, safe and efficient operation of the system is ensured, and good economic and environment-friendly effects are achieved.

Description

Technical Field

[0001] The present invention belongs to the field of new energy technology, and specifically relates to a plateau multi-energy complementary agricultural energy management system and method. Background Art

[0002] In the current global energy transformation process, how to efficiently integrate and manage multiple renewable energy sources is an important technical challenge. Renewable energy sources such as wind, photovoltaic and geothermal have significant supply intermittency, which makes it difficult for a single energy source to meet continuous and stable demand. In order to achieve stable supply and efficient utilization of energy, multi-energy complementarity has become a key solution. The multi-energy complementarity method aims to achieve dynamic energy distribution and storage by integrating renewable energy sources from different sources (such as wind, photovoltaic, geothermal, etc.) and real-time scheduling and optimization through intelligent management technology. This method can not only solve the intermittent problem of various energy sources, but also flexibly schedule according to actual needs, cut peaks and fill valleys, and ensure the stability of energy supply. In addition, the intelligent energy management method optimizes the use strategy of each energy source by analyzing data on energy supply and demand, combining meteorological conditions, real-time power load and other factors. Through precise scheduling algorithms, this method can achieve the complementary effect of different energy sources, maximize energy utilization efficiency, and reduce dependence on traditional energy. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a high-altitude multi-energy complementary agricultural energy management system and method that can achieve efficient energy supply for irrigation, processing and greenhouse heating, significantly improve the utilization rate of renewable energy and load satisfaction, reduce operating costs and environmental pollution, ensure safe and efficient operation of the system, and have good economic and environmental protection effects.

[0004] The technical solution of the present invention is:

[0005] A plateau multi-energy complementary agricultural energy management system, comprising: a central intelligent management system, an energy supply module, an energy storage module and a load module, wherein the central intelligent management system performs two-way communication with the energy supply module, the energy storage module and the load module through control signals, wherein the energy supply module comprises: a wind power generation module, a photovoltaic power generation module and a geothermal heating module, wherein the energy storage module comprises: a battery energy storage system and a phase change material thermal storage system, wherein the load module comprises: an irrigation system, grape processing equipment and a greenhouse heating system; wherein the wind power generation module supplies energy to the irrigation system, the grape processing equipment and the greenhouse heating system. The system directly supplies power, and the wind power generation module sends the remaining power after supplying power to the battery energy storage system. The photovoltaic power generation module directly supplies power to the irrigation system, grape processing equipment and greenhouse heating system, and the photovoltaic power generation module sends the remaining power after supplying power to the battery energy storage system. The geothermal heating module supplies heat to the greenhouse heating system through a thermal pipeline, and the geothermal heating module transmits the remaining heat energy to the phase change material thermal storage system after heating; the battery energy storage system is bidirectionally connected with the energy supply module and supplies power to the load module, and the phase change material thermal storage system is coupled with the thermal load of the greenhouse heating system.

[0006] Furthermore, the geothermal heating module includes a geothermal well and a geothermal heat pump, the geothermal well is connected to the geothermal heat pump through a circulation pipeline, and the geothermal heat pump extracts heat energy and supplies heat to the greenhouse heating system.

[0007] Furthermore, the battery energy storage system includes multiple battery modules and a matching battery management system, and the battery management system controls the charging paths and power distribution of different battery modules through a central intelligent management system.

[0008] Furthermore, the phase change material thermal energy storage system is connected to the greenhouse heating system via a heat exchanger and a circulation pump.

[0009] Furthermore, the installed capacity of the wind power generation module is 500kW, the installed capacity of the photovoltaic power generation module is 1000 kW, the installed capacity of the geothermal heating module is 1000kWth, the power storage capacity of the battery energy storage system is 1000kWh, and the heat storage capacity of the phase change material thermal energy storage system is 500kWth.

[0010] An energy management method for a plateau multi-energy complementary agricultural energy management system comprises the following steps: Step 1: System startup and initialization phase, including the following steps: Step S1, equipment and monitoring platform startup, including: starting the wind turbine, photovoltaic inverter, geothermal well circulation pump, and ground source heat pump unit in sequence; checking whether the battery energy storage system meets the initial safety power requirements, checking the valves, temperature sensors, and insulation components of the phase change material heat storage system; starting the on-site remote terminal unit, data acquisition and monitoring control system or Internet of Things platform to confirm the real-time collection and transmission of key data; Step S2: Reading system configuration and restrictions, including: Capacity limits: wind turbine rated power, maximum grid-connected power of photovoltaic inverters, and maximum heating power of geothermal units; Energy storage upper and lower limits: the minimum and maximum state of charge of the battery, the maximum heat storage capacity and operating temperature range of the phase change heat storage system; Load priority and safety threshold: the load priority is: irrigation system > grape processing equipment > greenhouse heating system, and the safety threshold is the minimum heating temperature of the greenhouse heating system; Protection thresholds, including: battery temperature threshold, geothermal well flow or pressure threshold, and phase change material temperature threshold; Step S3: system self-check and error handling, including: Equipment fault detection, including: fan fault alarm, photovoltaic module communication interruption, geothermal water pump startup failure; Safety protection: If it is detected that the current state of charge of the battery is lower than the safety value and there is a risk of leakage in the phase change heat storage tank, the system will alarm and shut down or switch to safety mode; Initialization log: record the startup time, device initial status, battery charge status, and phase change material temperature information for subsequent reference; Step 2: Data collection and preprocessing, including the following steps: Step S4, collecting real-time operation data, including: power generation data, energy storage data and load data, the power generation data includes: wind power real-time power, photovoltaic power, voltage and current information, geothermal real-time output, the geothermal real-time output is: heating power or flow rate × temperature difference; the energy storage data includes: battery current charge state, battery temperature, phase change heat storage tank internal temperature, heat storage and valve switch state; the load data includes: irrigation pump power, grape processing equipment operating power or state, greenhouse fan, auxiliary heating, lighting electricity and heating demand; Step S5, collecting external environment data, including: meteorological information, historical records and electricity price information, the meteorological information includes: wind speed and direction, irradiance, outdoor temperature, relative humidity; the historical records include: wind power and photovoltaic power generation module output records, load module power and heat consumption records and energy storage state changes of energy storage modules in the previous cycle or longer; the electricity price information includes: current electricity price or peak and valley electricity price period; Step S6, data preprocessing and cleaning, including: filtering outliers, including: instantaneous spikes, offline data; interpolating or filling missing samples; statistically analyzing key quantities needed for subsequent prediction or scheduling decisions, including: mean value, variance, and change trend; Step 3: Short-term forecast and load priority confirmation, including the following steps: Step S7, wind, solar and geothermal forecast, including: based on current and historical meteorological data, using a machine learning model to make a short-term forecast of wind power and photovoltaic power in the next period, the machine learning model is a random forest regression or a multi-layer perceptron neural network; making a basic estimate of geothermal heating capacity: obtaining the upper limit of thermal power based on the product of geothermal well flow and water temperature; Step S8, load demand forecasting, comprising; Irrigation system: predict the pumping load for the next period based on crop planting plans and seasonal water demand; Grape processing equipment: If it is at the peak of brewing or processing, the load is stable or at a peak during a specific period; Greenhouse heating system load: predict the heating demand of the greenhouse heating system based on outdoor temperature and crop growth requirements; Step S9, load priority strategy, including: determining power supply priority, determining heating priority, and if the function is insufficient, limiting or postponing low priority loads, if there is excess energy supply, storing excess electricity and heat in batteries and phase change materials; the power supply priority is: irrigation system>grape processing equipment>greenhouse heating system; the heating priority is: greenhouse heating system heating required amount>excess heat entering phase change heat storage; Step 4: Rolling scheduling and decision making, including the following steps, which are performed periodically: Step S10, constructing the state required for scheduling, including: sorting out the wind power generation, photovoltaic power generation, geothermal heat, remaining capacity of the energy storage system and load demand at the current time and in the forecast period; Step S11, execution of the scheduling algorithm, including: inputting the current state into the scheduling algorithm model; giving the next stage of power allocation or energy storage charging and discharging instructions according to the set load priority and equipment constraints; Step S12, power distribution and energy storage operation, including: Wind and solar power generation: Calculate the total power demand of the irrigation system, grape processing equipment and greenhouse heating system; when the power generation of wind and solar power generation meets the power demand of the load module, the load module power demand is met first, and the remaining power is used to charge the battery energy storage system; when the battery energy storage system is fully charged, it is connected to the grid or limited; Geothermal heating: Prioritize heat supply to the greenhouse heating system, and the excess heat enters the phase change material thermal energy storage system. If the phase change material thermal energy storage system is full, geothermal extraction is reduced or the excess heat is discharged; Energy storage dispatch, including: battery energy storage system: discharge support when load peaks or wind and solar power is insufficient; phase change material thermal storage energy storage system: release heat to supplement when greenhouse heating system load increases and geothermal supply is insufficient; Step S13, outputting a scheduling instruction, includes: Wind power and photovoltaic control: Set the output power limit or start-stop strategy of the grid-connected inverter to give priority to high-priority loads; Geothermal control: Control the heating power of geothermal wells or the circulating water flow to provide basic heating for the greenhouse heating system; Battery energy storage: set the charge and discharge power or working mode; Phase change heat storage: Set the valve opening or heat exchanger flow to determine whether the phase change material heat storage system absorbs or releases heat; Step S14, execution recording and updating, including: writing the scheduling results into the log, including the power or heat obtained by each load, the state change of the energy storage device and whether there is surplus grid connection or waste; preparing the data input and state update of the next cycle; Step 5: Real-time monitoring and security control, including the following steps: Step S15, real-time monitoring: compare the dispatching instruction with the actual power, battery SOC and phase change material temperature, and issue a warning if the deviation is too large or an abnormality occurs. The abnormality includes: rapid power loss of the energy storage module, abnormal temperature of the geothermal heating module; monitor whether the key load actually obtains the required energy, that is, whether the irrigation system pump operates normally and whether the power consumption of the grape processing equipment meets the requirements; Step S16, fault detection and protection, includes: Battery safety: If the battery's current state of charge exceeds the threshold or the temperature is too high, charging will automatically stop and an alarm will sound; Phase change heat storage safety: if the internal temperature or pressure exceeds the standard, the pressure will be automatically released or the heat source input will be stopped; Energy supply module: If the wind turbine speed or photovoltaic inverter temperature is abnormal, the load will be automatically reduced or the system will be shut down for protection; Load module: If the load is short-circuited or over-current occurs, the corresponding circuit will be cut off and an alarm will be sounded; Step S17, emergency dispatch switching: If a major fault is found, including: battery failure and inability to store energy and geothermal well failure and inability to provide heat, quickly switch to the emergency strategy: When the power supply is insufficient, temporarily sacrifice low priority loads and retain the irrigation system load; When the heat is insufficient, the phase change material heat storage system is activated to ensure that the greenhouse temperature is not lower than the minimum heating temperature; Step 6: Data recording and performance analysis, including the following steps: Step S18, data recording, including: The power generation, load, energy storage charge and discharge, system dispatch instructions, fault and alarm information of each cycle are written into the database in a unified manner; At the same time, weather and market electricity price information are saved to provide reference for subsequent energy optimization, performance evaluation and model training; The performance indicators of the performance evaluation include: Renewable utilization rate: measures the proportion of wind, solar and geothermal energy that is actually utilized, stored and connected to the grid; Load satisfaction: the ratio of the demands of each priority load being met during the scheduling period; Energy storage efficiency: battery charging and discharging loss, phase change heat storage and heat absorption and release loss; Economic indicators: whether the amount of electricity purchased from outside has been reduced or the operating cost has been lowered; Environmental indicators: greenhouse gas emission reduction and clean energy use ratio; Step S19, fault statistics and maintenance, includes: Analyze the cause and frequency of failures and assess whether inspection or maintenance is required; Count the lifespan impacts of key equipment and develop maintenance plans. The lifespan impacts include: battery charge and discharge cycles and fan operating hours; Step 7: Rolling iteration and optimization, including the following steps: Step S20, periodic scheduling: repeat the above steps 2-6 in each period and make dynamic adjustments; Step S21, long-term experience accumulation: each scheduling result and reward feedback are included in the experience replay pool, and the scheduling algorithm model is continuously trained and iteratively optimized; Step S22, adjusting objectives and strategies: adjusting load priority, energy allocation strategy or multi-objective weights according to seasonal changes, market electricity prices, and crop growth stage factors.

[0011] Furthermore, the machine learning model in step S7 is a random forest regression, comprising the following steps: Step 1: Data preparation and preprocessing, including the following steps: Step S1, collecting and integrating data, including: Collect historical meteorological data, including: wind speed v, wind direction , irradiance G, ambient temperature T, and the real power / thermal power value y of the corresponding period; Form the initial training set: , , in, represents the input feature vector of the i-th sample, Indicates wind speed, Indicates wind direction. represents the irradiance, Indicates the ambient temperature, represents the target output of the i-th sample, N represents the number of samples, Represents the transpose operation; Step S2, data cleaning and feature engineering: remove missing values ​​or use interpolation or mean method to fill in; remove obvious outliers; Step S3, divide the training, validation, and test sets: split the data in chronological order or randomly; Step S4: Output a clean and representative training set: ; Step 2: Random forest training and Bootstrap sampling, including the following steps: Step S5: Set the random forest hyperparameters, including: the number of trees K, the maximum depth of each tree max_depth, the minimum number of leaf node samples min_samples_leaf, and the number of features randomly extracted when splitting ; Step S6, Bootstrap sampling, including: Randomly extract N samples with replacement from the original training set to obtain a sub-training set , corresponding to the kth tree; Repeat the above operation K times to form ; The samples not selected for sampling are OBB samples, which are used for internal evaluation; Output K sub-training sets containing repeated samples ; Step 3: Single regression tree training: For the kth sub-training set , train a regression decision tree , including the following steps: Step S7, initialization: let the current node be the root node, including All samples in; Step S8: Random selection of feature subsets: At each split, randomly select from all d features Features participate in splitting candidates; Step S9, splitting criterion: For regression tree, use the variance minimization criterion to select the split point and The fth eigencomponent of , so that the splitting gain is maximized, that is: , in, Indicates the sample set contained in the current node, the number of samples is n, and They represent the sample sets of the left and right child nodes after splitting, and the number of samples is and , the larger the split gain Δ, the more effective the split is in reducing the variance and improving the prediction ability of the tree; Step S10, recursive growth, includes: If the splitting gain Δ is greater than the threshold and the maximum depth has not been reached, the current node is split into left and right child nodes, and the above process is repeated on the child nodes; When the number of node samples is less than the minimum number of leaf node samples or the maximum depth is reached, growth stops and the node is set as a leaf node. The output of the leaf node is the mean of the target value of the node samples. Step S11: Output a single regression tree , the expression is: , in, Indicates that the kth tree is input The output, represents the input feature vector, represents the number of leaf nodes of the kth tree, represents the feature space sub-region corresponding to the jth leaf node in the kth tree, Represents the constant prediction value corresponding to the leaf node, that is, the mean or weighted mean of the target values ​​of the training samples in the region; Step 4: Combine and generate random forests: complete K regression trees After training, the random forest model is obtained: , The model predicts the input sample as the mean of all tree outputs: , in, Represents the input feature vector The predicted value of, K represents the number of regression trees in the forest, Indicates that the kth tree is input Output: Step 5: Model prediction and evaluation, including the following steps: Step S12: Model prediction: for new samples Input to each tree one by one ,get , the final output of random forest is: , in, represents the input sample to be predicted, Represents the random forest model for the input The final prediction output, K represents the number of regression trees in the forest, represents the output of the kth regression tree; Step S13: Evaluation indicators include mean square error and mean absolute error. The expression of the mean square error is: , Among them, MSE represents the mean square error. The lower the MSE, the better the model prediction effect. represents the model prediction value of the i-th test sample, represents the true value of the i-th test sample, Indicates the sample number of the test sample; The expression of the mean absolute error is: , Among them, MAE represents the mean absolute error. The smaller the MAE, the closer the model prediction value is to the true value; Step S14, adjusting the participants for iterative optimization, including: According to the validation set or OBB error, adjust the number of trees K, the maximum depth of each tree max_depth, and the minimum number of leaf node samples min_samples_leaf; If overfitting occurs, reduce the maximum depth or increase the minimum number of leaf node samples. If the model accuracy is insufficient, increase K or perform more feature engineering.

[0012] Furthermore, the scheduling algorithm in step S11 includes the following steps: Step 1, initialization, includes the following steps: Step S1, parameter setting: setting discount factor , set the initial weights of multiple objectives: load satisfaction weight , waste penalty weight , cost penalty weight , security breach penalty weight , set the weight of the irrigation system in the reward function calculation , set the weight of grape processing equipment in the reward function calculation , set the weight of the greenhouse heating system in the reward function calculation ; Step S2: Network initialization: Initialize the online Q network and the target Q network ,in, represents the action-value function, Indicates the system status. represents the executable action, θ represents the trainable parameters of the online Q network, and θ — represents the trainable parameters of the target Q network; Step 2: Prediction and status update, including the following steps:

[0013] Step S3: Obtain the real-time power of wind power generation in the next period , real-time power of photovoltaic power generation , Geothermal real-time thermal power 、The current state of charge of the battery energy storage system , Current heat storage capacity of phase change material thermal energy storage system , Power supply requirements for irrigation systems , Power supply requirements for grape processing equipment , Power demand for greenhouse heating systems , heating demand of greenhouse heating system and historical scheduling strategies and performance indicators ; Step S4: Form a state vector , the state vector The expression is: , Step 3: Define the action space. The expression of the action space is: , in, represents the decision vector at time t, i.e., the action, represents the output power of wind power, Represents the photovoltaic output power, Indicates the charge / discharge power of the battery energy storage system in the current cycle, Represents the heat absorption / release power of the phase change material thermal energy storage system, Indicates the priority of power supply when power is insufficient; Step 4: Multi-objective weighted reward function, the expression of the multi-objective weighted reward function is: , , in, represents the immediate reward obtained at time t+1, represents the weighted sum of load satisfaction, It indicates the amount of available wind and solar power and geothermal heat that is not utilized and wasted. Represents the economic cost of purchasing electricity or fossil fuels. Indicates safety or capacity violations, including penalties caused by battery overcharge and discharge, phase change heat storage exceeding the upper limit, and equipment overload. represents the load satisfaction weight, represents the waste penalty weight, represents the cost penalty weight, represents the security breach penalty weight, It represents the electricity demand satisfaction rate of the irrigation system in this cycle, It indicates the electricity demand satisfaction rate of grape processing equipment in this cycle, It indicates the electricity demand satisfaction rate of the greenhouse heating system in this cycle, represents the weight of the irrigation system in the reward function calculation, represents the weight of grape processing equipment in the reward function calculation, represents the weight of the greenhouse heating system in the reward function calculation, ; Step 5: RL core loop, including the following steps: Step S5, state reading and action selection: At time t, according to the current state ,pass Greedy or noisy exploration strategy, from Q network Select the optimal or near-optimal action ,Right now: , in, represents the optimal action, represents the action space, that is, the set of all possible actions, represents the action-value function, represents the environmental state at time t, represents the executable action, θ represents the trainable parameters of the Q network, represents the exploration rate, Indicates in status Next, the action that maximizes the Q value; Step S6, execution scheduling and feedback, including; System Execution , update wind, solar and geothermal power distribution, battery charge state, phase change material heat storage and load module energy distribution; At the next scheduling time t+1, the actual operation data is measured, and the reward is calculated and output , and the new state ; Step S7, storing experience and neural network training, includes: Will Deposit into the experience replay pool; Every few steps, randomly draw small batches from the replay pool Perform training to minimize the DQN objective function, namely: , , in, represents the loss function, B represents the sample size of the mini-batch, represents the target Q value of the i-th sample, Represents the online Q network for the sample The predicted Q value of represents the immediate reward of the i-th sample in the environment, γ represents the discount factor, Indicates the next moment state , among all possible actions Find the Q value corresponding to the action that maximizes the Q value, θ represents the trainable parameters of the online Q network, θ — represents the trainable parameters of the target Q network, represents the environmental state of the i-th sample, Indicates in status The actions performed below, Indicates the state of the environment at the next moment. Indicates in status All possible candidate actions; Step S8: Regularly update the target network , that is: a complete copy of the current online network parameter θ is given to the target network parameter θ — ; Step 6: After the current round of training is completed, enter the next round of training and repeat the above steps 1-5 until the set number of training rounds or a satisfactory operation strategy is reached.

[0014] Beneficial effects of the present invention: 1. Multiple energy sources complement each other and improve the utilization rate of clean energy: The present invention integrates three clean energy sources, wind power, photovoltaic power and geothermal power, and coordinates battery energy storage and phase change heat storage system for coordinated scheduling, which greatly improves the actual utilization rate of renewable energy in plateau areas and reduces the waste of wind power, photovoltaic power and geothermal waste heat; 2. Priority scheduling and guarantee of key loads: The present invention establishes a load priority strategy (irrigation system > grape processing equipment > greenhouse heating system) and performs real-time monitoring and prediction, which can give priority to guaranteeing the demand of important loads when energy supply is insufficient, thereby improving the reliability and stability of agricultural production; 3. Rolling optimization and adaptation to dynamic environment: The present invention adopts rolling scheduling and combines machine learning models for short-term prediction. According to real-time changes in wind speed, irradiance, load demand, etc., the scheduling instructions are updated periodically to ensure that the system can flexibly respond to environmental fluctuations and changes in energy demand; 4. Energy storage synergy and efficient energy management: The battery energy storage system of the present invention works in synergy with the phase change material thermal energy storage system to achieve "dual energy storage" management of electricity and thermal energy. It can store electricity and geothermal energy when there is excess wind and solar energy, and can also provide electricity and thermal support during peak energy consumption, thereby improving the overall energy utilization efficiency of the system. 5. Safety protection and fault emergency response: The present invention can quickly identify fan failure, battery overcharge or thermal storage system abnormality through multiple threshold monitoring (battery charge state, geothermal well flow, phase change material temperature, etc.) and fault detection mechanism, and switch to safety mode or emergency strategy to enhance system safety and sustainable operation capability; 6. Multi-objective optimization, balancing economy and environmental protection: The present invention introduces a multi-objective weighted reward function (load satisfaction, waste penalty, cost penalty, safety breach penalty) into the scheduling algorithm, which can maximize the use of renewable energy and reduce the consumption of purchased electricity or fossil fuels while ensuring agricultural production and safe operation, balancing economy and environmental protection; 7. Data-driven, continuous iterative improvement: The present invention continuously accumulates scheduling decision and environmental feedback data through rolling iteration and experience replay pool mechanism, continuously trains and optimizes machine learning models and reinforcement learning scheduling algorithms, and improves the long-term operating performance and adaptability of the system; 8. Applicable to increasing income and green development of plateau agriculture: The present invention can provide a stable and clean energy supply for crop planting, grape processing and greenhouses in plateau areas, reduce dependence on fossil energy, have both high economic returns and good environmental benefits, and contribute to the green and sustainable development of agriculture in plateau areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a structural schematic diagram of a plateau multi-energy complementary agricultural energy management system of the present invention. DETAILED DESCRIPTION

[0016] like Figure 1As shown, a plateau multi-energy complementary agricultural energy management system includes: a central intelligent management system, an energy supply module, an energy storage module and a load module. The central intelligent management system communicates bidirectionally with the energy supply module, the energy storage module and the load module through control signals. The energy supply module includes: a wind power generation module, a photovoltaic power generation module and a geothermal heating module. The energy storage module includes: a battery energy storage system and a phase change material thermal storage system. The load module includes: an irrigation system, grape processing equipment and a greenhouse heating system; the wind power generation module supplies energy to the irrigation system, the grape processing equipment and the greenhouse. The heating system is directly powered, and the wind power generation module sends the remaining power after power supply to the battery energy storage system. The photovoltaic power generation module directly supplies power to the irrigation system, grape processing equipment and greenhouse heating system, and the photovoltaic power generation module sends the remaining power after power supply to the battery energy storage system. The geothermal heating module supplies heat to the greenhouse heating system through a thermal pipeline, and the geothermal heating module transmits the remaining heat energy to the phase change material thermal storage system after heating; the battery energy storage system is bidirectionally connected with the energy supply module and supplies power to the load module, and the phase change material thermal storage system is coupled with the thermal load of the greenhouse heating system.

[0017] Preferably, the geothermal heating module includes a geothermal well and a geothermal heat pump, the geothermal well is connected to the geothermal heat pump through a circulation pipeline, and the geothermal heat pump extracts heat energy and supplies heat to the greenhouse heating system.

[0018] Preferably, the battery energy storage system includes multiple groups of battery modules and a matching battery management system, and the battery management system controls the charging paths and power distribution of different battery modules through a central intelligent management system.

[0019] Preferably, the phase change material thermal energy storage system is connected to the greenhouse heating system via a heat exchanger and a circulation pump.

[0020] Preferably, the installed capacity of the wind power generation module is 500kW, the installed capacity of the photovoltaic power generation module is 1000 kW, the installed capacity of the geothermal heating module is 1000kWth, the power storage capacity of the battery energy storage system is 1000kWh, and the heat storage capacity of the phase change material thermal energy storage system is 500kWth.

[0021] An energy management method for a plateau multi-energy complementary agricultural energy management system comprises the following steps: Step 1: System startup and initialization phase, including the following steps: Step S1, equipment and monitoring platform startup, including: starting the wind turbine, photovoltaic inverter, geothermal well circulation pump, and ground source heat pump unit in sequence; checking whether the battery energy storage system meets the initial safety power requirements, checking the valves, temperature sensors, and insulation components of the phase change material heat storage system; starting the on-site remote terminal unit, data acquisition and monitoring control system or Internet of Things platform to confirm the real-time collection and transmission of key data; Step S2: Reading system configuration and restrictions, including: Capacity limits: wind turbine rated power, maximum grid-connected power of photovoltaic inverters, and maximum heating power of geothermal units; Energy storage upper and lower limits: the minimum and maximum state of charge of the battery, the maximum heat storage capacity and operating temperature range of the phase change heat storage system; Load priority and safety threshold: the load priority is: irrigation system > grape processing equipment > greenhouse heating system, and the safety threshold is the minimum heating temperature of the greenhouse heating system; Protection thresholds, including: battery temperature threshold, geothermal well flow or pressure threshold, and phase change material temperature threshold; Step S3: system self-check and error handling, including: Equipment fault detection, including: fan fault alarm, photovoltaic module communication interruption, geothermal water pump startup failure; Safety protection: If it is detected that the current state of charge of the battery is lower than the safety value and there is a risk of leakage in the phase change heat storage tank, the system will alarm and shut down or switch to safety mode; Initialization log: record the startup time, device initial status, battery charge status, and phase change material temperature information for subsequent reference; Step 2: Data collection and preprocessing, including the following steps: Step S4, collecting real-time operation data, including: power generation data, energy storage data and load data, the power generation data includes: wind power real-time power, photovoltaic power, voltage and current information, geothermal real-time output, the geothermal real-time output is: heating power or flow rate × temperature difference; the energy storage data includes: battery current state of charge (SOC%), battery temperature, phase change heat storage tank internal temperature, heat storage and valve switch status; the load data includes: irrigation pump power, grape processing equipment operating power or status, greenhouse fan, auxiliary heating, lighting electricity and heating demand; Step S5, collecting external environment data, including: meteorological information, historical records and electricity price information, the meteorological information includes: wind speed and direction, irradiance, outdoor temperature, relative humidity; the historical records include: wind power and photovoltaic power generation module output records, load module power and heat consumption records and energy storage state changes of energy storage modules in the previous cycle or longer; the electricity price information includes: current electricity price or peak and valley electricity price period; Step S6, data preprocessing and cleaning, including: filtering outliers, including: instantaneous spikes, offline data; interpolating or filling missing samples; statistically analyzing key quantities needed for subsequent prediction or scheduling decisions, including: mean value, variance, and change trend; Step 3: Short-term forecast and load priority confirmation, including the following steps: Step S7, wind, solar and geothermal forecast, including: based on current and historical meteorological data, using a machine learning model to make a short-term forecast of wind power and photovoltaic power in the next period, the machine learning model is a random forest regression or a multi-layer perceptron neural network; making a basic estimate of geothermal heating capacity: obtaining the upper limit of thermal power based on the product of geothermal well flow and water temperature; Step S8, load demand forecasting, comprising; Irrigation system: predict the pumping load for the next period based on crop planting plans and seasonal water demand; Grape processing equipment: If it is at the peak of brewing or processing, the load is stable or at a peak during a specific period; Greenhouse heating system load: predict the heating demand of the greenhouse heating system based on outdoor temperature and crop growth requirements; Step S9, load priority strategy, including: determining power supply priority, determining heating priority, and if the function is insufficient, limiting or postponing low priority loads, if there is excess energy supply, storing excess electricity and heat in batteries and phase change materials; the power supply priority is: irrigation system>grape processing equipment>greenhouse heating system; the heating priority is: greenhouse heating system heating required amount>excess heat entering phase change heat storage; Step 4: Rolling scheduling and decision making, including the following steps, which are performed periodically: Step S10, constructing the state required for scheduling, including: sorting out the wind power generation, photovoltaic power generation, geothermal heat, remaining capacity of the energy storage system and load demand at the current time and in the forecast period; Step S11, execution of the scheduling algorithm, including: inputting the current state into the scheduling algorithm model; giving the next stage of power allocation or energy storage charging and discharging instructions according to the set load priority and equipment constraints; Step S12, power distribution and energy storage operation, including: Wind and solar power generation: Calculate the total power demand of the irrigation system, grape processing equipment and greenhouse heating system; when the power generation of wind and solar power generation meets the power demand of the load module, the load module power demand is met first, and the remaining power is used to charge the battery energy storage system; when the battery energy storage system is fully charged, it is connected to the grid or limited; Geothermal heating: Prioritize heat supply to the greenhouse heating system, and the excess heat enters the phase change material thermal energy storage system. If the phase change material thermal energy storage system is full, geothermal extraction is reduced or the excess heat is discharged; Energy storage dispatch, including: battery energy storage system: discharge support when load peaks or wind and solar power is insufficient; phase change material thermal storage energy storage system: release heat to supplement when greenhouse heating system load increases and geothermal supply is insufficient; Step S13, outputting a scheduling instruction, includes: Wind power and photovoltaic control: Set the output power limit or start-stop strategy of the grid-connected inverter to give priority to high-priority loads; Geothermal control: Control the heating power of geothermal wells or the circulating water flow to provide basic heating for the greenhouse heating system; Battery energy storage: set the charge and discharge power or working mode (peak shaving, power replenishment); Phase change heat storage: Set the valve opening or heat exchanger flow to determine whether the phase change material heat storage system absorbs or releases heat; Step S14, execution recording and updating, including: writing the scheduling results into the log, including the power or heat obtained by each load, the state change of the energy storage device and whether there is surplus grid connection or waste; preparing the data input and state update of the next cycle; Step 5: Real-time monitoring and security control, including the following steps: Step S15, real-time monitoring: compare the dispatching instruction with the actual power, battery SOC and phase change material temperature, and issue a warning if the deviation is too large or an abnormality occurs. The abnormality includes: rapid power loss of the energy storage module, abnormal temperature of the geothermal heating module; monitor whether the key load actually obtains the required energy, that is, whether the irrigation system pump operates normally and whether the power consumption of the grape processing equipment meets the requirements; Step S16, fault detection and protection, includes: Battery safety: If the battery's current state of charge exceeds the threshold or the temperature is too high, charging will automatically stop and an alarm will sound; Phase change heat storage safety: if the internal temperature or pressure exceeds the standard, the pressure will be automatically released or the heat source input will be stopped; Energy supply module: If the wind turbine speed or photovoltaic inverter temperature is abnormal, the load will be automatically reduced or the system will be shut down for protection; Load module: If the load is short-circuited or over-current occurs, the corresponding circuit will be cut off and an alarm will be sounded; Step S17, emergency dispatch switching: If a major fault is found, including: battery failure and inability to store energy and geothermal well failure and inability to provide heat, quickly switch to the emergency strategy: When the power supply is insufficient, temporarily sacrifice low priority loads and retain the irrigation system load; When the heat is insufficient, the phase change material heat storage system is activated to ensure that the greenhouse temperature is not lower than the minimum heating temperature; Step 6: Data recording and performance analysis, including the following steps: Step S18, data recording, including: The power generation, load, energy storage charge and discharge, system dispatch instructions, fault and alarm information of each cycle are written into the database in a unified manner; At the same time, weather and market electricity price information are saved to provide reference for subsequent energy optimization, performance evaluation and model training; The performance indicators of the performance evaluation include: Renewable utilization rate: measures the proportion of wind, solar and geothermal energy that is actually utilized, stored and connected to the grid; Load satisfaction: the ratio of the demands of each priority load being met during the scheduling period; Energy storage efficiency: battery charging and discharging loss, phase change heat storage and heat absorption and release loss; Economic indicators: whether the amount of electricity purchased from outside has been reduced or the operating cost has been lowered; Environmental indicators: greenhouse gas emission reduction and clean energy use ratio; Step S19, fault statistics and maintenance, includes: Analyze the cause and frequency of failures and assess whether inspection or maintenance is required; Count the lifespan impacts of key equipment and develop maintenance plans. The lifespan impacts include: battery charge and discharge cycles and fan operating hours; Step 7: Rolling iteration and optimization, including the following steps: Step S20, periodic scheduling: repeat the above steps 2-6 in each period (such as 15 minutes, 30 minutes or 1 hour) and make dynamic adjustments; Step S21, long-term experience accumulation: each scheduling result and reward feedback are included in the experience replay pool, and the scheduling algorithm model is continuously trained and iteratively optimized; Step S22, adjusting objectives and strategies: adjusting load priority, energy allocation strategy or multi-objective weights (such as reducing costs vs. increasing renewable energy utilization) according to seasonal changes, market electricity prices, and crop growth stage factors.

[0022] Preferably, the machine learning model in step S7 is a random forest regression, comprising the following steps: Step 1: Data preparation and preprocessing, including the following steps: Step S1, collecting and integrating data, including: Collect historical meteorological data, including: wind speed v, wind direction , irradiance G, ambient temperature T, and the real power / thermal power value y of the corresponding period; Form the initial training set: , , in, represents the input feature vector of the i-th sample, Indicates wind speed, Indicates wind direction. represents the irradiance, Indicates the ambient temperature, represents the target output of the i-th sample, N represents the number of samples, Represents the transpose operation; Step S2, data cleaning and feature engineering: remove missing values ​​or use interpolation or mean method to fill in; remove obvious outliers; Step S3, divide the training, validation, and test sets: split the data in chronological order or randomly; Step S4: Output a clean and representative training set: ; Step 2: Random forest training and Bootstrap sampling, including the following steps: Step S5: Set the random forest hyperparameters, including: the number of trees K, the maximum depth of each tree max_depth, the minimum number of leaf node samples min_samples_leaf, and the number of features randomly extracted when splitting ; Step S6, Bootstrap sampling, including: Randomly extract N samples with replacement from the original training set to obtain a sub-training set , corresponding to the kth tree; Repeat the above operation K times to form ; The samples not selected for sampling are OBB samples, which are used for internal evaluation; Output K sub-training sets containing repeated samples ; Step 3: Single regression tree training: For the kth sub-training set , train a regression decision tree , including the following steps: Step S7, initialization: let the current node be the root node, including All samples in; Step S8: Random selection of feature subsets: At each split, randomly select from all d features Features participate in splitting candidates; Step S9, splitting criterion: For regression tree, use the variance minimization criterion to select the split point and The fth eigencomponent of , so that the splitting gain is maximized, that is: , in, Indicates the sample set contained in the current node, the number of samples is n, and They represent the sample sets of the left and right child nodes after splitting, and the number of samples is and , the larger the split gain Δ, the more effective the split is in reducing the variance and improving the prediction ability of the tree; Step S10, recursive growth, includes: If the splitting gain Δ is greater than the threshold and the maximum depth has not been reached, the current node is split into left and right child nodes, and the above process is repeated on the child nodes; When the number of node samples is less than the minimum number of leaf node samples or the maximum depth is reached, growth stops and the node is set as a leaf node. The output of the leaf node is the mean of the target value of the node samples. Step S11: Output a single regression tree , the expression is: , in, Indicates that the kth tree is input The output, represents the input feature vector, represents the number of leaf nodes of the kth tree, represents the feature space sub-region corresponding to the jth leaf node in the kth tree, Represents the constant prediction value corresponding to the leaf node, that is, the mean or weighted mean of the target values ​​of the training samples in the region; Step 4: Combine and generate random forests: complete K regression trees After training, the random forest model is obtained: , The model predicts the input sample as the mean of all tree outputs: , in, Represents the input feature vector The predicted value of, K represents the number of regression trees in the forest, Indicates that the kth tree is input Output: Step 5: Model prediction and evaluation, including the following steps: Step S12: Model prediction: for new samples Input to each tree one by one ,get , the final output of random forest is: , in, represents the input sample to be predicted, Represents the random forest model for the input The final prediction output, K represents the number of regression trees in the forest, represents the output of the kth regression tree; Step S13: Evaluation indicators include mean square error and mean absolute error. The expression of the mean square error is: , Among them, MSE represents the mean square error. The lower the MSE, the better the model prediction effect. represents the model prediction value of the i-th test sample, represents the true value of the i-th test sample, Indicates the sample number of the test sample; The expression of the mean absolute error is: , Among them, MAE represents the mean absolute error. The smaller the MAE, the closer the model prediction value is to the true value; Step S14, adjusting the participants for iterative optimization, including: According to the validation set or OBB error, adjust the number of trees K, the maximum depth of each tree max_depth, and the minimum number of leaf node samples min_samples_leaf; If overfitting occurs, reduce the maximum depth or increase the minimum number of leaf node samples. If the model accuracy is insufficient, increase K or perform more feature engineering.

[0023] Preferably, the scheduling algorithm in step S11 includes the following steps: Step 1, initialization, includes the following steps: Step S1, parameter setting: setting discount factor , set the initial weights of multiple objectives: load satisfaction weight , waste penalty weight , cost penalty weight , security breach penalty weight , set the weight of the irrigation system in the reward function calculation , set the weight of grape processing equipment in the reward function calculation , set the weight of the greenhouse heating system in the reward function calculation ; Step S2: Network initialization: Initialize the online Q network and the target Q network ,in, represents the action-value function, Indicates the system status. represents the executable action, θ represents the trainable parameters of the online Q network, and θ — represents the trainable parameters of the target Q network; Step 2: Prediction and status update, including the following steps: Step S3: Obtain the real-time power of wind power generation in the next period , real-time power of photovoltaic power generation , Geothermal real-time thermal power 、The current state of charge of the battery energy storage system , Current heat storage capacity of phase change material thermal energy storage system , Power supply requirements for irrigation systems , Power supply requirements for grape processing equipment , Power demand for greenhouse heating systems , heating demand of greenhouse heating system and historical scheduling strategies and performance indicators ; Step S4: Form a state vector , the state vector The expression is: , Step 3: Define the action space. The expression of the action space is: , in, represents the decision vector at time t, i.e., the action, represents the output power of wind power, Represents the photovoltaic output power, Indicates the charge / discharge power of the battery energy storage system in the current cycle, Represents the heat absorption / release power of the phase change material thermal energy storage system, Indicates the priority of power supply when power is insufficient; Step 4: Multi-objective weighted reward function, the expression of the multi-objective weighted reward function is: , , in, represents the immediate reward obtained at time t+1, represents the weighted sum of load satisfaction, It indicates the amount of available wind and solar power and geothermal heat that is not utilized and wasted. Represents the economic cost of purchasing electricity or fossil fuels. Indicates safety or capacity violations, including penalties caused by battery overcharge and discharge, phase change heat storage exceeding the upper limit, and equipment overload. represents the load satisfaction weight, represents the waste penalty weight, represents the cost penalty weight, represents the security breach penalty weight, It represents the electricity demand satisfaction rate of the irrigation system in this cycle, It indicates the electricity demand satisfaction rate of grape processing equipment in this cycle, It indicates the electricity demand satisfaction rate of the greenhouse heating system in this cycle, represents the weight of the irrigation system in the reward function calculation, represents the weight of grape processing equipment in the reward function calculation, represents the weight of the greenhouse heating system in the reward function calculation, ; Step 5: RL core loop, including the following steps: Step S5, state reading and action selection: At time t, according to the current state ,pass Greedy or noisy exploration strategy, from Q network Select the optimal or near-optimal action ,Right now: , in, represents the optimal action, represents the action space, that is, the set of all possible actions, represents the action-value function, represents the environmental state at time t, represents the executable action, θ represents the trainable parameters of the Q network, represents the exploration rate, Indicates in status Next, the action that maximizes the Q value; Step S6, execution scheduling and feedback, including; System Execution , update wind, solar and geothermal power distribution, battery charge state, phase change material heat storage and load module energy distribution; At the next scheduling time t+1, the actual operation data is measured, and the reward is calculated and output , and the new state ; Step S7, storing experience and neural network training, includes: Will Deposit into the experience replay pool; Every few steps, randomly draw small batches from the replay pool Perform training to minimize the DQN objective function, namely: , , in, represents the loss function, B represents the sample size of the mini-batch, represents the target Q value of the i-th sample, Represents the online Q network for the sample The predicted Q value of represents the immediate reward of the i-th sample in the environment, γ represents the discount factor, Indicates the next moment state , among all possible actions Find the Q value corresponding to the action that maximizes the Q value, θ represents the trainable parameters of the online Q network, θ — represents the trainable parameters of the target Q network, represents the environmental state of the i-th sample, Indicates in status The actions performed below, Indicates the state of the environment at the next moment. Indicates in status All possible candidate actions; Step S8: Regularly update the target network , that is: a complete copy of the current online network parameter θ is given to the target network parameter θ — ; Step 6: After the current round of training is completed, enter the next round of training and repeat the above steps 1-5 until the set number of training rounds or a satisfactory operation strategy is reached.

[0024] The basic principles, main features and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not used to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which shall fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A plateau multi-energy complementary agricultural energy management system, characterized in that: include: A central intelligent management system, an energy supply module, an energy storage module and a load module. The central intelligent management system communicates bidirectionally with the energy supply module, the energy storage module and the load module through control signals. The energy supply module includes: a wind power generation module, a photovoltaic power generation module and a geothermal heating module. The energy storage module includes: a battery energy storage system and a phase change material thermal storage system. The load module includes: an irrigation system, grape processing equipment and a greenhouse heating system. The wind power generation module directly supplies power to the irrigation system, the grape processing equipment and the greenhouse heating system. The wind power generation module sends the remaining power after power supply to the battery energy storage system. The photovoltaic power generation module directly supplies power to the irrigation system, the grape processing equipment and the greenhouse heating system. The photovoltaic power generation module sends the remaining power after power supply to the battery energy storage system. The geothermal heating module supplies heat to the greenhouse heating system through a thermal pipeline. After heating, the geothermal heating module transmits the remaining heat energy to the phase change material thermal storage system. The battery energy storage system is bidirectionally connected with the energy supply module and supplies power to the load module. The phase change material thermal storage system is coupled with the thermal load of the greenhouse heating system.

2. A plateau multi-energy complementary agricultural energy management system according to claim 1, characterized in that: The geothermal heating module includes a geothermal well and a geothermal heat pump. The geothermal well is connected to the geothermal heat pump through a circulation pipeline. The geothermal heat pump extracts heat energy and supplies heat to the greenhouse heating system.

3. A plateau multi-energy complementary agricultural energy management system according to claim 1, characterized in that: The battery energy storage system includes multiple battery modules and a matching battery management system. The battery management system controls the charging paths and power distribution of different battery modules through a central intelligent management system.

4. A plateau multi-energy complementary agricultural energy management system according to claim 1, characterized in that: The phase change material heat storage system is connected to the greenhouse heating system via a heat exchanger and a circulation pump.

5. The plateau multi-energy complementary agricultural energy management system according to claim 1 is characterized in that: The installed capacity of the wind power generation module is 500kW, the installed capacity of the photovoltaic power generation module is 1000 kW, the installed capacity of the geothermal heating module is 1000kWth, the power storage capacity of the battery energy storage system is 1000kWh, and the heat storage capacity of the phase change material thermal energy storage system is 500kWth.

6. An energy management method for a plateau multi-energy complementary agricultural energy management system according to claim 1, characterized in that: The following steps are involved: Step 1: System startup and initialization phase, including the following steps: Step S1, equipment and monitoring platform startup, including: starting the wind turbine, photovoltaic inverter, geothermal well circulation pump, and ground source heat pump unit in sequence; checking whether the battery energy storage system meets the initial safety power requirements, checking the valves, temperature sensors, and insulation components of the phase change material heat storage system; starting the on-site remote terminal unit, data acquisition and monitoring control system or Internet of Things platform to confirm the real-time collection and transmission of key data; Step S2: Reading system configuration and restrictions, including: Capacity limits: wind turbine rated power, maximum grid-connected power of photovoltaic inverters, and maximum heating power of geothermal units; Energy storage upper and lower limits: the minimum and maximum state of charge of the battery, the maximum heat storage capacity and operating temperature range of the phase change heat storage system; Load priority and safety threshold: the load priority is: irrigation system > grape processing equipment > greenhouse heating system, and the safety threshold is the minimum heating temperature of the greenhouse heating system; Protection thresholds, including: battery temperature threshold, geothermal well flow or pressure threshold, and phase change material temperature threshold; Step S3: system self-check and error handling, including: Equipment fault detection, including: fan fault alarm, photovoltaic module communication interruption, geothermal water pump startup failure; Safety protection: If it is detected that the current state of charge of the battery is lower than the safety value and there is a risk of leakage in the phase change heat storage tank, the system will alarm and shut down or switch to safety mode; Initialization log: record the startup time, device initial status, battery charge status, and phase change material temperature information for subsequent reference; Step 2: Data collection and preprocessing, including the following steps: Step S4, collecting real-time operation data, including: power generation data, energy storage data and load data, the power generation data includes: wind power real-time power, photovoltaic power, voltage and current information, geothermal real-time output, the geothermal real-time output is: heating power or flow rate × temperature difference; the energy storage data includes: battery current charge state, battery temperature, phase change heat storage tank internal temperature, heat storage and valve switch state; the load data includes: irrigation pump power, grape processing equipment operating power or state, greenhouse fan, auxiliary heating, lighting electricity and heating demand; Step S5, collecting external environment data, including: meteorological information, historical records and electricity price information, the meteorological information includes: wind speed and direction, irradiance, outdoor temperature, relative humidity; the historical records include: wind power and photovoltaic power generation module output records, load module power and heat consumption records and energy storage state changes of energy storage modules in the previous cycle or longer; the electricity price information includes: current electricity price or peak and valley electricity price period; Step S6, data preprocessing and cleaning, including: filtering outliers, including: instantaneous spikes, offline data; interpolating or filling missing samples; statistically analyzing key quantities needed for subsequent prediction or scheduling decisions, including: mean value, variance, and change trend; Step 3: Short-term forecast and load priority confirmation, including the following steps: Step S7, wind, solar and geothermal forecast, including: based on current and historical meteorological data, using a machine learning model to make a short-term forecast of wind power and photovoltaic power in the next period, the machine learning model is a random forest regression or a multi-layer perceptron neural network; making a basic estimate of geothermal heating capacity: obtaining the upper limit of thermal power based on the product of geothermal well flow and water temperature; Step S8, load demand forecasting, comprising; Irrigation system: predict the pumping load for the next period based on crop planting plans and seasonal water demand; Grape processing equipment: If it is at the peak of brewing or processing, the load is stable or at a peak during a specific period; Greenhouse heating system load: predict the heating demand of the greenhouse heating system based on outdoor temperature and crop growth requirements; Step S9, load priority strategy, including: determining power supply priority, determining heating priority, and if the function is insufficient, limiting or postponing low priority loads, if there is excess energy supply, storing excess electricity and heat in batteries and phase change materials; the power supply priority is: irrigation system>grape processing equipment>greenhouse heating system; the heating priority is: greenhouse heating system heating required amount>excess heat entering phase change heat storage; Step 4: Rolling scheduling and decision making, including the following steps, which are performed periodically: Step S10, constructing the state required for scheduling, including: sorting out the wind power generation, photovoltaic power generation, geothermal heat, remaining capacity of the energy storage system and load demand at the current time and in the forecast period; Step S11, execution of the scheduling algorithm, including: inputting the current state into the scheduling algorithm model; giving the next stage of power allocation or energy storage charging and discharging instructions according to the set load priority and equipment constraints; Step S12, power distribution and energy storage operation, including: Wind and solar power generation: Calculate the total power demand of the irrigation system, grape processing equipment and greenhouse heating system; when the power generation of wind and solar power generation meets the power demand of the load module, the load module power demand is met first, and the remaining power is used to charge the battery energy storage system; when the battery energy storage system is fully charged, it is connected to the grid or limited; Geothermal heating: Prioritize heat supply to the greenhouse heating system, and the excess heat enters the phase change material thermal energy storage system. If the phase change material thermal energy storage system is full, geothermal extraction is reduced or the excess heat is discharged; Energy storage dispatch, including: battery energy storage system: discharge support when load peaks or wind and solar power is insufficient; phase change material thermal storage energy storage system: release heat to supplement when greenhouse heating system load increases and geothermal supply is insufficient; Step S13, outputting a scheduling instruction, includes: Wind power and photovoltaic control: Set the output power limit or start-stop strategy of the grid-connected inverter to give priority to high-priority loads; Geothermal control: Control the heating power of geothermal wells or the circulating water flow to provide basic heating for the greenhouse heating system; Battery energy storage: set the charge and discharge power or working mode; Phase change heat storage: Set the valve opening or heat exchanger flow to determine whether the phase change material heat storage system absorbs or releases heat; Step S14, execution recording and updating, including: writing the scheduling results into the log, including the power or heat obtained by each load, the state change of the energy storage device and whether there is surplus grid connection or waste; preparing the data input and state update of the next cycle; Step 5: Real-time monitoring and security control, including the following steps: Step S15, real-time monitoring: compare the dispatching instruction with the actual power, battery SOC and phase change material temperature, and issue a warning if the deviation is too large or an abnormality occurs. The abnormality includes: rapid power loss of the energy storage module, abnormal temperature of the geothermal heating module; monitor whether the key load actually obtains the required energy, that is, whether the irrigation system pump operates normally and whether the power consumption of the grape processing equipment meets the requirements; Step S16, fault detection and protection, includes: Battery safety: If the battery's current state of charge exceeds the threshold or the temperature is too high, charging will automatically stop and an alarm will sound; Phase change heat storage safety: if the internal temperature or pressure exceeds the standard, the pressure will be automatically released or the heat source input will be stopped; Energy supply module: If the wind turbine speed or photovoltaic inverter temperature is abnormal, the load will be automatically reduced or the system will be shut down for protection; Load module: If the load is short-circuited or over-current occurs, the corresponding circuit will be cut off and an alarm will be sounded; Step S17, emergency dispatch switching: If a major fault is found, including: battery failure and inability to store energy and geothermal well failure and inability to provide heat, quickly switch to the emergency strategy: When the power supply is insufficient, temporarily sacrifice low priority loads and retain the irrigation system load; When the heat is insufficient, the phase change material heat storage system is activated to ensure that the greenhouse temperature is not lower than the minimum heating temperature; Step 6: Data recording and performance analysis, including the following steps: Step S18, data recording, including: The power generation, load, energy storage charge and discharge, system dispatch instructions, fault and alarm information of each cycle are written into the database in a unified manner; At the same time, weather and market electricity price information are saved to provide reference for subsequent energy optimization, performance evaluation and model training; The performance indicators of the performance evaluation include: Renewable utilization rate: measures the proportion of wind, solar and geothermal energy that is actually utilized, stored and connected to the grid; Load satisfaction: the ratio of the demands of each priority load being met during the scheduling period; Energy storage efficiency: battery charging and discharging loss, phase change heat storage and heat absorption and release loss; Economic indicators: whether the amount of electricity purchased from outside has been reduced or the operating cost has been lowered; Environmental indicators: greenhouse gas emission reduction and clean energy use ratio; Step S19, fault statistics and maintenance, includes: Analyze the cause and frequency of failures and assess whether inspection or maintenance is required; Count the lifespan impacts of key equipment and develop maintenance plans. The lifespan impacts include: battery charge and discharge cycles and fan operating hours; Step 7: Rolling iteration and optimization, including the following steps: Step S20, periodic scheduling: repeat the above steps 2-6 in each period and make dynamic adjustments; Step S21, long-term experience accumulation: each scheduling result and reward feedback are included in the experience replay pool, and the scheduling algorithm model is continuously trained and iteratively optimized; Step S22, adjusting objectives and strategies: adjusting load priority, energy allocation strategy or multi-objective weights according to seasonal changes, market electricity prices, and crop growth stage factors.

7. The energy management method of a plateau multi-energy complementary agricultural energy management system according to claim 6 is characterized in that: The machine learning model in step S7 is a random forest regression, which includes the following steps: Step 1: Data preparation and preprocessing, including the following steps: Step S1, collecting and integrating data, including: Collect historical meteorological data, including: wind speed v, wind direction , irradiance G, ambient temperature T, and the real power / thermal power value y of the corresponding period; Form the initial training set: ， ， in, represents the input feature vector of the i-th sample, Indicates wind speed, Indicates wind direction. represents irradiance, Indicates the ambient temperature, represents the target output of the i-th sample, N represents the number of samples, Represents the transpose operation; Step S2, data cleaning and feature engineering: remove missing values ​​or use interpolation or mean method to fill in; remove obvious outliers; Step S3, divide the training, validation, and test sets: split the data in chronological order or randomly; Step S4: Output a clean and representative training set: ; Step 2: Random forest training and Bootstrap sampling, including the following steps: Step S5: Set the random forest hyperparameters, including: the number of trees K, the maximum depth of each tree max_depth, the minimum number of leaf node samples min_samples_leaf, and the number of features randomly extracted when splitting ; Step S6, Bootstrap sampling, including: Randomly extract N samples with replacement from the original training set to obtain a sub-training set , corresponding to the kth tree; Repeat the above operation K times to form ; The samples not selected for sampling are OBB samples, which are used for internal evaluation; Output K sub-training sets containing repeated samples ; Step 3: Single regression tree training: For the kth sub-training set , train a regression decision tree , including the following steps: Step S7, initialization: let the current node be the root node, including All samples in; Step S8: Random selection of feature subsets: At each split, randomly select from all d features Features participate in splitting candidates; Step S9, splitting criterion: For regression tree, use the variance minimization criterion to select the split point and The fth eigencomponent of , so that the splitting gain is maximized, that is: ， in, Indicates the sample set contained in the current node, the number of samples is n, and They represent the sample sets of the left and right child nodes after splitting, and the number of samples is and , the larger the split gain Δ, the more effective the split is in reducing the variance and improving the prediction ability of the tree; Step S10, recursive growth, includes: If the splitting gain Δ is greater than the threshold and the maximum depth has not been reached, the current node is split into left and right child nodes, and the above process is repeated on the child nodes; When the number of node samples is less than the minimum number of leaf node samples or the maximum depth is reached, growth stops and the node is set as a leaf node. The output of the leaf node is the mean of the target value of the node samples. Step S11: Output a single regression tree , the expression is: ， in, Indicates that the kth tree is input The output, represents the input feature vector, represents the number of leaf nodes of the kth tree, represents the feature space sub-region corresponding to the jth leaf node in the kth tree, Represents the constant prediction value corresponding to the leaf node, that is, the mean or weighted mean of the target values ​​of the training samples in the region; Step 4: Combine and generate random forests: complete K regression trees After training, the random forest model is obtained: ， The model predicts the input sample as the mean of all tree outputs: ， in, Represents the input feature vector The predicted value of, K represents the number of regression trees in the forest, Indicates that the kth tree is input Output: Step 5: Model prediction and evaluation, including the following steps: Step S12: Model prediction: for new samples Input to each tree one by one ,get , the final output of random forest is: ， in, represents the input sample to be predicted, Represents the random forest model for the input The final prediction output, K represents the number of regression trees in the forest, represents the output of the kth regression tree; Step S13: Evaluation indicators include mean square error and mean absolute error. The expression of the mean square error is: ， Among them, MSE represents the mean square error. The lower the MSE, the better the model prediction effect. represents the model prediction value of the i-th test sample, represents the true value of the i-th test sample, Indicates the sample number of the test sample; The expression of the mean absolute error is: ， Among them, MAE represents the mean absolute error. The smaller the MAE, the closer the model prediction value is to the true value; Step S14, adjusting the participants for iterative optimization, including: According to the validation set or OBB error, adjust the number of trees K, the maximum depth of each tree max_depth, and the minimum number of leaf node samples min_samples_leaf; If overfitting occurs, reduce the maximum depth or increase the minimum number of leaf node samples. If the model accuracy is insufficient, increase K or perform more feature engineering.

8. The energy management method of a plateau multi-energy complementary agricultural energy management system according to claim 6 is characterized in that: The scheduling algorithm in step S11 includes the following steps: Step 1, initialization, includes the following steps: Step S1, parameter setting: setting discount factor , set the initial weights of multiple objectives: load satisfaction weight , waste penalty weight , cost penalty weight , security breach penalty weight , set the weight of the irrigation system in the reward function calculation , set the weight of grape processing equipment in the reward function calculation , set the weight of the greenhouse heating system in the reward function calculation ; Step S2: Network initialization: Initialize the online Q network and the target Q network ,in, represents the action-value function, Indicates the system status. represents the executable action, θ represents the trainable parameters of the online Q network, and θ — represents the trainable parameters of the target Q network; Step 2: Prediction and status update, including the following steps: Step S3: Obtain the real-time power of wind power generation in the next period , real-time power of photovoltaic power generation , Geothermal real-time thermal power 、The current state of charge of the battery energy storage system , Current heat storage capacity of phase change material thermal energy storage system , Power supply requirements for irrigation systems , Power supply requirements for grape processing equipment , Power demand for greenhouse heating systems , heating demand of greenhouse heating system and historical scheduling strategies and performance indicators ; Step S4: Form a state vector , the state vector The expression is: ， Step 3: Define the action space. The expression of the action space is: ， in, represents the decision vector at time t, i.e., the action, represents the output power of wind power, Represents the output power of photovoltaic, Indicates the charge / discharge power of the battery energy storage system in the current cycle, Represents the heat absorption / release power of the phase change material thermal energy storage system, Indicates the priority of power supply when power is insufficient; Step 4: Multi-objective weighted reward function, the expression of the multi-objective weighted reward function is: ， ， in, represents the immediate reward obtained at time t+1, represents the weighted sum of load satisfaction, It indicates the amount of available wind and solar power and geothermal heat that is not utilized and wasted. Represents the economic cost of purchasing electricity or fossil fuels. Indicates safety or capacity violations, including penalties caused by battery overcharge and discharge, phase change heat storage exceeding the upper limit, and equipment overload. represents the load satisfaction weight, represents the waste penalty weight, represents the cost penalty weight, represents the security breach penalty weight, It represents the electricity demand satisfaction rate of the irrigation system in this cycle, It indicates the electricity demand satisfaction rate of grape processing equipment in this cycle, It indicates the electricity demand satisfaction rate of the greenhouse heating system in this cycle, represents the weight of the irrigation system in the reward function calculation, represents the weight of grape processing equipment in the reward function calculation, represents the weight of the greenhouse heating system in the reward function calculation, ; Step 5: RL core loop, including the following steps: Step S5, state reading and action selection: At time t, according to the current state ,pass Greedy or noisy exploration strategy, from Q network Select the optimal or near-optimal action ,Right now: ， in, represents the optimal action, represents the action space, that is, the set of all possible actions, represents the action-value function, represents the environmental state at time t, represents the executable action, θ represents the trainable parameters of the Q network, represents the exploration rate, Indicates in status Next, the action that maximizes the Q value; Step S6, execution scheduling and feedback, including; System Execution , update wind, solar and geothermal power distribution, battery charge state, phase change material heat storage and load module energy distribution; At the next scheduling time t+1, the actual operation data is measured, and the reward is calculated and output , and the new state ; Step S7, storing experience and neural network training, includes: Will Deposit into the experience replay pool; Every few steps, randomly draw small batches from the replay pool Perform training to minimize the DQN objective function, namely: ， ， in, represents the loss function, B represents the sample size of the mini-batch, represents the target Q value of the i-th sample, Represents the online Q network for the sample The predicted Q value of represents the immediate reward of the i-th sample in the environment, γ represents the discount factor, Indicates the next moment state , among all possible actions Find the Q value corresponding to the action that maximizes the Q value, θ represents the trainable parameters of the online Q network, θ — represents the trainable parameters of the target Q network, represents the environmental state of the i-th sample, Indicates in status The actions performed below, Indicates the state of the environment at the next moment. Indicates in status All possible candidate actions; Step S8: Regularly update the target network , that is: a complete copy of the current online network parameter θ is given to the target network parameter θ — ; Step 6: After the current round of training is completed, enter the next round of training and repeat the above steps 1-5 until the set number of training rounds or a satisfactory operation strategy is reached.

Images