A solar energy driven cold storage control system

Abstract

The present application belongs to the field of refrigeration technology and new energy application, and particularly relates to a cold storage control system driven by solar energy. The present application adopts a three-level control mode to control the compressor in different time periods. First, factors directly related to the working state of the cold storage compressor are taken as input prediction to obtain a second target frequency as a real-time control instruction to adjust the working state of the compressor. Then, based on a heat load prediction model, the cold quantity is scheduled in advance, and a third target frequency is taken as a real-time control instruction to adjust the working state of the compressor. The control instruction has a higher priority than the control instruction of the second target frequency. Finally, the compressor frequency output by a multi-objective collaborative optimization algorithm is taken as the highest control instruction, and multiple targets such as energy efficiency, temperature accuracy, battery health and compressor life are comprehensively weighed to ensure the overall performance of the system is optimal.

Description

Technical Field

[0001] This invention belongs to the field of refrigeration technology and new energy applications, and in particular relates to a solar-powered cold storage control system. Background Technology

[0002] Traditional cold storage systems rely heavily on the power grid, resulting in high energy consumption and limited applicability in areas without or with insufficient electricity. While solar-powered cold storage systems alleviate this dependence to some extent, their development faces two core challenges: efficiently capturing unstable solar energy and achieving precise and stable temperature control within the storage facility under limited energy conditions. Currently, most solar-powered cold storage systems employ the traditional perturbation-observation method for maximum power point tracking (MPPT) in energy management. This method suffers from issues such as continuous oscillations near the maximum power point and tracking inaccuracies during rapid changes in sunlight, leading to low energy capture efficiency. Regarding temperature control, existing systems often use simple on / off control or fixed-parameter PID regulation, which struggles to effectively handle changes in cold storage load, ambient temperature fluctuations, and thermal disturbances during defrosting. This results in significant temperature fluctuations within the storage facility, affecting the quality of stored goods and causing energy waste. Furthermore, existing systems generally lack the ability to perceive and adapt to user habits and lack data-driven system health status prediction and maintenance reminders, indicating a significant need to improve their overall intelligence level. Therefore, there is an urgent need to develop a cold storage system solution that is highly efficient and adaptable in both energy capture and temperature control, and integrates intelligent management and autonomous decision-making capabilities. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of the prior art and provide a solar-driven cold storage system with high photovoltaic energy capture efficiency, precise temperature control, and intelligent and stable operation, which controls the compressor through a three-level control method in different time periods.

[0004] This invention proposes a solar-powered cold storage control system, the system comprising:

[0005] The control unit is used to control the various modules in the cold storage control system.

[0006] The photovoltaic power optimization module uses deep reinforcement learning to model the photovoltaic system power optimization process as a Markov decision process. Through trial and error learning, it performs maximum power point tracking in complex environments to obtain the first output power.

[0007] The adaptive temperature control module uses a deep neural network (DNN) to predict and obtain the second target frequency of the variable frequency compressor, with T1 as the time period. Target opening degree of electronic expansion valve , the second target frequency Target opening degree of electronic expansion valve This serves as a real-time control command to adjust the compressor's operating status.

[0008] The energy efficiency adaptive optimization module uses a heat load prediction model to calculate the cooling demand within a future time window, with T2 as the time period, and then calculates the frequency adjustment amount of the third compressor based on the cooling demand. Adjust the frequency of the third compressor Used for the second target frequency Compensation is performed to obtain the third target frequency. The third target frequency As a real-time control command to adjust the compressor's operating state, this control command has a higher priority than the second target frequency. Control commands;

[0009] The multi-objective collaborative optimization module, with a time period of T3, combines the first output power and the second target frequency. Third compressor frequency adjustment amount Battery SOC, ambient temperature , storage temperature deviation Compressor running time With current As input, the core of this module lies in defining a comprehensive optimization objective function (i.e., a multi-objective function) and mapping the aforementioned input parameters to the quantized values ​​of each optimization term in this objective function. Optimization using the multi-objective function yields target control commands, including MPPT power output setting, battery charging / discharging strategy, and optimal compressor frequency; the optimal compressor frequency serves as the highest-level control command.

[0010] Wherein, T1 < T2 < T3. The specific values ​​of the three-level control cycles T1, T2, and T3 can be set according to the system hardware performance and dynamic characteristics. For example, a feasible setting is: T1 = 5 seconds (fast real-time control), T2 = 5 minutes (proactive energy scheduling), and T3 = 30 minutes (global system optimization).

[0011] The photovoltaic power optimization module specifically includes:

[0012] The system initialization submodule is used to initialize the deep reinforcement learning agent, including its policy network, target network, and experience replay buffer; and to load the pre-trained network model weights.

[0013] The state-aware submodule samples the output voltage of the photovoltaic array in each control cycle. Output current And calculate the current power. ; to include the current and N past moments Data as a state vector .

[0014] The action decision submodule inputs the state vector s(k) into the policy network of the DRL agent, and the network outputs an action a(k), which represents the adjustment command for the power converter switching signal.

[0015] The "Execute Actions and Environmental Feedback" submodule performs the actions. This involves changing the system's operating parameters to adjust the power output state of the photovoltaic array; in the next cycle, the new power is sampled. The degree of improvement in power output characteristics is used as an immediate reward. .

[0016] The experience storage and learning submodule will store the current experience tuples. Store the data in the experience replay buffer; periodically sample a batch of experience data randomly from the buffer to update the weights of the policy network, with the goal of maximizing long-term cumulative rewards.

[0017] The output control submodule converts the actions output by the strategy network to obtain the first output power, and generates actual adjustment commands based on the first output power to drive the actuators in the power controller; and waits for the next control cycle to repeat the photovoltaic power optimization process.

[0018] The adaptive temperature control module includes:

[0019] The system preparation submodule loads a pre-trained DNN controller model; this model has undergone extensive offline supervised learning using a large amount of historical running data before deployment.

[0020] The status data acquisition submodule collects multi-dimensional status information with a time period of T1 and constructs an input vector. The input vector includes: the current return air temperature. Target set temperature Temperature error Temperature error change rate Ambient temperature evaporator outlet temperature ,Battery and current time .

[0021] The DNN forward inference submodule feeds the constructed input vector into the DNN controller; the network then performs multiple nonlinear transformations and directly outputs the control vector. The output control vector u(k) includes: the second target frequency of the variable frequency compressor. Target opening degree of electronic expansion valve ;

[0022] The first real-time controller will set the second target frequency. Target opening degree of electronic expansion valve This serves as a real-time control command to adjust the compressor's operating status.

[0023] The energy efficiency adaptive optimization module includes:

[0024] The parameter loading module loads the parameters for the heat load prediction model.

[0025] The data acquisition submodule collects data on ambient temperature, number and duration of door openings, photovoltaic power generation, battery SOC, and internal temperature of the storage facility, using T2 as the time period.

[0026] The heat load prediction submodule uses a heat load prediction model to calculate the cooling demand within a future time window.

[0027] The control quantity prediction submodule determines whether the predicted heat load exceeds a threshold. If a high-temperature period or frequent door opening is predicted, proactive control is executed. When photovoltaic power is sufficient, the compressor is instructed to cool down in advance to build up cooling capacity reserves. The control output is the frequency adjustment amount of the third compressor. .

[0028] The second real-time controller adjusts the frequency of the third compressor. Used for the second target frequency Compensation is performed to obtain the third target frequency. The third target frequency As a real-time control command to adjust the compressor's operating state, this control command has a higher priority than the second target frequency. Control commands.

[0029] In the multi-objective collaborative optimization module, the first output power and the second target frequency are combined. Third compressor frequency adjustment amount The system uses battery SOC, ambient temperature, storage temperature deviation, compressor running time, and current as inputs, and optimizes using a multi-objective function to obtain target control commands, including MPPT power output setting, battery charging and discharging strategy, and optimal compressor frequency; the optimal compressor frequency is used as the highest-level control command.

[0030] This invention employs a three-level control method to control the compressor within different time periods, firstly based on the return air temperature. The system uses several factors directly related to the operating state of the cold storage compressor, such as the target set temperature, temperature error, rate of change of temperature error, ambient temperature, and evaporator outlet temperature, as inputs to predict and obtain a second target frequency, which is then used as a real-time control command to adjust the compressor's operating state. Next, based on a heat load prediction model, the system proactively schedules cooling capacity and uses a third target frequency as a real-time control command to adjust the compressor's operating state; this third frequency has a higher priority than the control command for the second target frequency. Finally, the compressor frequency output by the multi-objective collaborative optimization algorithm is used as the highest control command, comprehensively balancing multiple objectives such as energy efficiency, temperature accuracy, battery health, and compressor lifespan to ensure optimal overall system performance. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 The working principle of a solar-powered cold storage facility;

[0033] Figure 2 This is a schematic diagram of the solar-powered cold storage control system proposed in this invention.

[0034] Figure 3 This is a block diagram illustrating the working principle of the photovoltaic power optimization module proposed in this invention.

[0035] Figure 4 This is a block diagram illustrating the working principle of the adaptive temperature control module proposed in this invention.

[0036] Figure 5 This is a block diagram illustrating the working principle of the energy efficiency adaptive optimization module proposed in this invention. Detailed Implementation

[0037] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.

[0038] like Figure 1As shown, the DC power generated by the photovoltaic power generation unit is first converted to maximum power point tracking and voltage stability through a photovoltaic power optimization control algorithm based on deep reinforcement learning (DRL), and then fed into the system's DC bus. The energy storage unit uses lithium iron phosphate battery packs, which are directly connected to the DC bus, and store or release electrical energy according to the system's energy balance state.

[0039] The DC bus serves as the system's energy hub, distributing electrical energy in a unified manner: on the one hand, it directly supplies power to various DC loads; on the other hand, it transmits electrical energy to the sine wave inverter module integrated inside the control unit, converting it into 220V, 50Hz pure sine wave AC power, specifically for driving a 3-horsepower AC variable frequency compressor.

[0040] The core MCU of the control unit comprehensively collects sensor data such as temperature, voltage, and current, and centrally executes core algorithms such as power optimization control and temperature control. The control unit precisely controls the compressor, electronic expansion valve, fan, and other actuators through the drive circuit.

[0041] like Figure 2 As shown, this invention proposes a solar-powered cold storage control system, the system comprising:

[0042] The control unit is used to control the various modules in the cold storage control system.

[0043] The photovoltaic power optimization module uses deep reinforcement learning to model the photovoltaic system power optimization process as a Markov decision process. Through trial and error learning, it performs maximum power point tracking in complex environments to obtain the first output power.

[0044] The adaptive temperature control module uses a deep neural network (DNN) to predict and obtain the second target frequency of the variable frequency compressor, with T1 as the time period. Target opening degree of electronic expansion valve , the second target frequency Target opening degree of electronic expansion valve This serves as a real-time control command to adjust the compressor's operating status.

[0045] The energy efficiency adaptive optimization module uses a heat load prediction model to calculate the cooling demand within a future time window, with T2 as the time period, and then calculates the frequency adjustment amount of the third compressor based on the cooling demand. Adjust the frequency of the third compressor Used for the second target frequency Compensation is performed to obtain the third target frequency. The third target frequency As a real-time control command to adjust the compressor's operating state, this control command has a higher priority than the second target frequency. Control commands;

[0046] The multi-objective collaborative optimization module, with a time period of T3, combines the first output power and the second target frequency. Third compressor frequency adjustment amount The system uses battery SOC, ambient temperature, storage temperature deviation, compressor running time, and current as inputs, and optimizes using a multi-objective function to obtain target control commands, including MPPT power output setting, battery charging and discharging strategy, and optimal compressor frequency; the optimal compressor frequency is used as the highest-level control command.

[0047] Wherein, T1 < T2 < T3. The specific values ​​of the three-level control cycles T1, T2, and T3 can be set according to the system hardware performance and dynamic characteristics. For example, a feasible setting is: T1 = 5 seconds (fast real-time control), T2 = 5 minutes (proactive energy scheduling), and T3 = 30 minutes (global system optimization).

[0048] Specifically, such as Figure 3 As shown, the photovoltaic power optimization module specifically includes:

[0049] The system initialization submodule is used to initialize the deep reinforcement learning agent, including its policy network, target network, and experience replay buffer; and to load the pre-trained network model weights.

[0050] The state-aware submodule samples the output voltage of the photovoltaic array in each control cycle. Output current And calculate the current power. ; to include the current and N past moments Data as a state vector .

[0051] The action decision submodule inputs the state vector s(k) into the policy network of the DRL agent, and the network outputs an action a(k), which represents the adjustment command for the power converter switching signal.

[0052] The "Execute Actions and Environmental Feedback" submodule performs the actions. This involves changing the system's operating parameters to adjust the power output state of the photovoltaic array; in the next cycle, the new power is sampled. The degree of improvement in power output characteristics is used as an immediate reward. .

[0053] The experience storage and learning submodule will store the current experience tuples. Store the data in the experience replay buffer; periodically sample a batch of experience data randomly from the buffer to update the weights of the policy network, with the goal of maximizing long-term cumulative rewards.

[0054] The output control submodule converts the actions output by the strategy network to obtain the first output power, and generates actual adjustment commands based on the first output power to drive the actuators in the power controller; and waits for the next control cycle to repeat the photovoltaic power optimization process.

[0055] This deep reinforcement learning power optimization algorithm does not rely on a precise mathematical model of the photovoltaic array and can effectively cope with complex operating conditions such as rapid fluctuations in solar radiation and changes in output characteristics caused by cloud cover. Through autonomous learning, it can avoid lingering at suboptimal operating points, directly and quickly find the optimal power output state of the system, and autonomously adjust control parameters according to environmental changes to ensure that the system always operates in a superior state.

[0056] like Figure 4 As shown, the adaptive temperature control module includes:

[0057] The system preparation submodule loads a pre-trained DNN controller model; this model has undergone extensive offline supervised learning using a large amount of historical running data before deployment.

[0058] The status data acquisition submodule collects multi-dimensional status information with a time period of T1 and constructs an input vector. The input vector includes: the current return air temperature. Target set temperature Temperature error Temperature error change rate Ambient temperature Evaporator outlet temperature, battery SOC, and current time.

[0059] The deep neural network (DNN) controller in the adaptive temperature control module of this invention is a fully connected feedforward neural network specifically designed for cold storage systems. Its input layer receives an 8-dimensional state vector. The output layer directly regresses the 2D control vector ( and The main body of the model consists of four hidden layers (with 256, 128, 64, and 32 neurons respectively). Each layer is followed by batch normalization, ReLU activation function, and Dropout layers (with dropout rates of 0.3, 0.3, 0.2, and 0.1 respectively) to optimize training stability and generalization ability. The model uses the mean squared error loss function, employs the Adam optimizer, and applies a preheating and cosine annealing learning rate scheduling strategy for offline training. Ultimately, it learns the precise nonlinear mapping relationship from the state of the complex system to the optimal control command. ,in By using fixed parameters obtained from network training, end-to-end real-time precise control can be achieved.

[0060] The DNN forward inference submodule feeds the constructed input vector into the DNN controller; the internal processing of the network can be represented as follows: , Represents the parameters obtained from training A defined nonlinear function for a deep neural network. This function has learned the optimal control strategy for the system through offline training. The network undergoes multiple nonlinear transformations and directly outputs the control vector. The output control vector u(k) includes: the second target frequency of the variable frequency compressor. Target opening degree of electronic expansion valve ;

[0061] The first real-time controller will set the second target frequency. Target opening degree of electronic expansion valve This serves as a real-time control command to adjust the compressor's operating status.

[0062] This invention uses a data-driven deep neural network to replace the traditional fuzzy PID controller, directly achieving end-to-end precise control of the refrigeration unit. By learning the high-dimensional nonlinear mapping relationship of the DNN system, it directly achieves end-to-end precise control of the cold storage compressor frequency and the opening of the electronic expansion valve. It can implicitly learn and compensate for the defrosting process and thermal disturbances during door opening, ensuring stable storage temperature, and no longer relies on manually set PID parameters and fuzzy rules.

[0063] like Figure 5 As shown, the energy efficiency adaptive optimization module includes:

[0064] The parameter loading module loads the parameters for the heat load prediction model.

[0065] The data acquisition submodule collects data on ambient temperature, number and duration of door openings, photovoltaic power generation, battery SOC, and internal temperature of the storage facility, using T2 as the time period.

[0066] The heat load prediction submodule uses a heat load prediction model to calculate the cooling demand within a future time window. The heat load prediction model is as follows:

[0067]

[0068] in, Predict heat load, For the ambient temperature, Warehouse door opening frequency It refers to photovoltaic power generation capacity. The battery is in its state of charge. , , , The coefficients of the model, This is the error term. The model coefficients... , , , The following data was obtained through system identification methods: During the system trial operation phase, a large number of historical data samples were collected. , , , The least squares method is used to perform multiple linear regression fitting to obtain the coefficient values ​​that minimize the prediction error. These coefficients can also be periodically calibrated and updated online based on system operating data after deployment.

[0069] The control quantity prediction submodule determines whether the predicted heat load exceeds a threshold. If a high-temperature period or frequent door opening is predicted, proactive control is executed. When photovoltaic power is sufficient, the compressor is instructed to cool down in advance to build up cooling capacity reserves. The control output is the frequency adjustment amount of the third compressor. The calculation formula is:

[0070]

[0071] in, This is the gain coefficient. This is the heat load threshold.

[0072] The second real-time controller adjusts the frequency of the third compressor. Used for the second target frequency Compensation is performed to obtain the third target frequency. The third target frequency This control command, used to adjust the compressor's operating state in real time, has a higher priority than the second target frequency. Control commands.

[0073] In the multi-objective collaborative optimization module, the first output power and the second target frequency are combined. Third compressor frequency adjustment amount Using battery SOC, ambient temperature, storage temperature deviation, compressor running time, and current as inputs, a multi-objective function is used for optimization to obtain target control commands, including MPPT power output setting, battery charging and discharging strategy, and optimal compressor frequency; the optimal compressor frequency is used as the highest-level control command; the multi-objective function... for

[0074] in, Energy consumption cost is related to photovoltaic power output and grid replenishment costs. For the temperature deviation of the storage room, the absolute value of the difference between the input return air temperature and the set temperature is directly taken. Battery loss rate is estimated based on the number of SOC cycles and charge / discharge rate; Compressor fatigue is calculated based on the number of start-stop cycles and load rate. , , , For weights.

[0075] Specifically, in the multi-objective collaborative optimization module, the mapping relationship is used to combine the first output power and the second target frequency. Third compressor frequency adjustment amount Battery SOC, ambient temperature, storage temperature deviation, and compressor running time and current data are mapped to energy consumption costs. , storage temperature deviation Battery loss rate and compressor fatigue The mapping relationship is a mapping model commonly used in this field, such as regression analysis, machine learning, and linear normalization.

[0076] As can be seen from the above formula, the multi-objective function Energy consumption cost , storage temperature deviation Battery loss rate and compressor fatigue These four parameters determine the value of the MPPT power, battery charging / discharging, and compressor frequency. Changing any one of them results in a multi-objective function. The value of will then change.

[0077] The optimization is performed using a multi-objective function, that is, to find a set of control commands (including the MPPT power output setpoint). Battery charging and discharging power commands and ultimately the optimal compressor frequency. This minimizes the objective function value. The optimal compressor frequency is... It will serve as the highest-level control command, directly overriding or integrating the frequency commands given by lower-level modules.

[0078] The weights in the multi-objective function ( , , , The setting of [specific parameters] reflects the emphasis placed on different performance indicators at different operating stages or according to user preferences. For example, when photovoltaic energy is sufficient, [specific parameters] can be appropriately reduced. (Energy cost weighting), improve (Temperature accuracy weighting); may be improved at night or when battery power is low. and (Battery health weights). These weights can be preset in multiple sets according to typical scenarios during system deployment, or authorized users can adjust them within a certain range.

[0079] An example range for setting initial weights is: :[0.3, 0.6], [0.2, 0.4], :[0.1,0.3], [0.1, 0.3], and satisfying + + + = 1. The system pre-stores multiple weight preset tables for different typical scenarios (such as solar power sufficient daytime mode, nighttime energy storage power supply mode, and low battery protection mode); during operation, the control unit automatically matches or interpolates to calculate the current optimal weight combination based on the current first output power (solar power), battery SOC, and time information. For example, when solar power is sufficient, the system can reduce the energy consumption cost weight. Reduce the temperature accuracy weight to 0.3. Up to 0.4; when the battery SOC is below 20%, the battery health weighting is significantly improved. The weights are adjusted to 0.3, and other weights are reduced. Dynamic weight adjustment strategies can also be achieved through offline simulation optimization or online reinforcement learning.

[0080] To more clearly illustrate the implementation of this invention, a specific application example is provided below: A cold storage room with a capacity of 20 cubic meters and a set temperature of -2℃. The system control cycle is set as follows: T1 = 5 seconds, T2 = 5 minutes, T3 = 30 minutes. The photovoltaic power optimization module: Based on DRL, the intelligent agent samples the voltage and current of the photovoltaic array in real time, decides and adjusts the duty cycle of the DC-DC converter, ensuring the photovoltaic system always operates near its maximum power point. For example, when sunlight intensifies at 10:00 AM, the module quickly adjusts, increasing the first output power from 2.5kW to 3.8kW. The adaptive temperature control module (cycle T1 = 5 seconds): Collects status information every 5 seconds: current return air temperature -1.8℃, target temperature -2℃, temperature error -0.2℃, error change rate 0.01℃ / s, ambient temperature 32℃, evaporator outlet temperature -10℃, battery SOC 85%, and current time 10:00:00. This data forms the input vector X(k), which is then input into a pre-trained DNN model. The DNN model calculates and outputs a control vector u(k), which may include, for example, the second target frequency. =45 Hz, target opening of electronic expansion valve =60%. The first real-time controller immediately sets the compressor frequency to 45Hz and the electronic expansion valve opening to 60%. Energy efficiency adaptive optimization module (cycle T2 = 5 minutes): Every 5 minutes, this module collects the ambient temperature over a period of time, the warehouse door opening record (e.g., predicting two warehouse entry operations within the next 30 minutes, each lasting 1 minute), the current photovoltaic power generation (3.8kW), and the battery SOC (85%). This data is input into the heat load prediction model. Assuming the model coefficients have been identified and determined through historical data: The predicted heat load for the next 30 minutes is calculated. The power output is 15 kW, which is higher than the preset threshold. =12 kW. Calculate the frequency adjustment amount of the third compressor according to the formula. Assuming the gain coefficient k = 0.5, then =0.5*(15-12)=1.5 Hz. The second real-time controller will =1.5 Hz compensation to the current second target frequency of 45 Hz, to obtain the third target frequency. =46.5 Hz, and use this as an instruction to adjust the compressor, increasing the cooling capacity in advance to cope with the upcoming heat load. Multi-objective collaborative optimization module (cycle T3=30 minutes): Every 30 minutes, this module summarizes the inputs: first output power (average 3.5kW), second target frequency (average 44Hz), third compressor frequency adjustment amount (average 0.8Hz over the past 30 minutes), battery SOC (80%), ambient temperature (average 30℃), storage temperature deviation (average 0.15℃), compressor running time (cumulative 8 hours), and current (average 9A). Based on the mapping relationship, the terms of the objective function J are calculated: (Mainly depends on photovoltaic power, assumed to be 0.5) (0.15) (Based on an estimated change in SOC of 0.1) (Based on a load factor estimate of 0.2), all the above values ​​are normalized values. The system selects a weight combination from a preset table based on the current scenario (sufficient solar daytime mode), for example: =0.35, =0.35, =0.15, =0.15. The calculated J = 0.35*0.5 + 0.35*0.15 + 0.15*0.1 + 0.15*0.2 = 0.2975. The algorithm iteratively searches under constraints to find a combination of control commands that minimizes J. Finally, it outputs a set of optimized target control commands, such as: fine-tuning the MPPT power output setpoint, charging the battery with a smaller current, and optimizing the compressor frequency. =45.2 Hz. This is the optimal frequency. As the highest-level instruction, it guides or overrides the frequency control of lower-level modules in the next T3 cycle.

[0081] For example, for combination 1, the MPPT is 3.5kW, the battery is charged at a current of 0.5A, and the compressor frequency is 46.5 Hz. At this time, the calculated value of J is 0.301.

[0082] For combination 2, the MPPT is 3.5kW, the battery is charged at a current of 0.5A, and the compressor frequency is 45.2Hz. At this time, the calculated value of J is 0.2975.

[0083] For combination 3, the MPPT is 3.8kW, the battery is not charged, and the compressor frequency is 45.2 Hz. At this time, the calculated value of J is 0.295.

[0084] Therefore, take When the value is 0.2975, the corresponding compressor frequency is 45.2Hz, MPPT is 3.5kW, and the target control command is to charge the battery with a current of 0.5A.

[0085] The adaptive temperature control algorithm based on deep neural networks (DNNs) directly outputs the second target frequency of the variable frequency compressor. As a low-level real-time controller, it directly calculates the compressor frequency using a deep neural network model based on state variables such as current temperature error and ambient temperature, enabling rapid response to temperature changes and maintaining stable storage temperature. This algorithm operates in a relatively fast control cycle and focuses on high-precision temperature control. It can provide the basic control input for the compressor frequency, serving as the core of temperature control.

[0086] The compressor frequency adjustment amount is output based on the energy efficiency adaptive optimization algorithm. As a forward-looking controller, based on a heat load prediction model, it adjusts the compressor frequency in advance when photovoltaic power is sufficient to build up cooling capacity reserves and mitigate heat load shocks. This algorithm operates in a relatively slow control cycle, focusing on energy efficiency optimization. It can compensate for or adjust the target frequency output by the DNN, serving as a feedforward control variable for optimizing energy efficiency.

[0087] The multi-objective collaborative optimization algorithm directly outputs the compressor frequency as one of the control commands. Acting as a high-level coordinator, it comprehensively balances multiple objectives such as energy efficiency, temperature accuracy, battery health, and compressor lifespan, dynamically adjusting system operating parameters. This algorithm operates in a slower control cycle, optimizing system performance from a global perspective.

[0088] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0089] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0090] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0091] In the embodiments provided by this invention, it should be understood that the disclosed apparatus / terminal devices and methods can be implemented in other ways. For example, the apparatus / terminal device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0092] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0093] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0094] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A solar-powered cold storage control system, characterized in that, The system includes: The control unit is used to control the various modules in the cold storage control system. The photovoltaic power optimization module uses deep reinforcement learning to model the photovoltaic system power optimization process as a Markov decision process. Through trial and error learning, it performs maximum power point tracking in complex environments to obtain the first output power. The adaptive temperature control module uses a deep neural network (DNN) to predict and obtain the second target frequency of the variable frequency compressor, with T1 as the time period. Target opening degree of electronic expansion valve , the second target frequency Target opening degree of electronic expansion valve As a real-time control command, it adjusts the compressor's operating status; The energy efficiency adaptive optimization module uses a heat load prediction model to calculate the cooling demand within a future time window, with T2 as the time period, and then calculates the frequency adjustment amount of the third compressor based on the cooling demand. Adjust the frequency of the third compressor Used for the second target frequency Compensation is performed to obtain the third target frequency. The third target frequency As a real-time control command to adjust the compressor's operating state, this control command has a higher priority than the second target frequency. Control commands; The multi-objective collaborative optimization module, with a time period of T3, combines the first output power and the second target frequency. Third compressor frequency adjustment amount The system uses battery SOC, ambient temperature, storage temperature deviation, compressor running time and current as inputs, and optimizes using a multi-objective function to obtain target control commands, including MPPT power output setting, battery charging and discharging strategy, and optimal compressor frequency; the optimal compressor frequency is used as the highest level control command. Where T1 < T2 < T3.

2. The solar-powered cold storage control system according to claim 1, wherein the adaptive temperature control module comprises: The system preparation submodule loads the pre-trained DNN controller model. The state data acquisition submodule collects multi-dimensional state information with a time period of T1 and constructs an input vector; The DNN forward inference submodule feeds the constructed input vector into the DNN controller; the network then performs multiple nonlinear transformations and directly outputs the control vector. ; The output control vector u(k) includes: the second target frequency of the variable frequency compressor. Target opening degree of electronic expansion valve ; The first real-time controller will set the second target frequency. Target opening degree of electronic expansion valve This serves as a real-time control command to adjust the compressor's operating status.

3. The solar-powered cold storage control system according to claim 2, wherein the input vector includes: Current return air temperature Target set temperature Temperature error Temperature error change rate Ambient temperature Evaporator outlet temperature, battery SOC, and current time.

4. The solar-powered cold storage control system according to claim 1, wherein the energy efficiency adaptive optimization module comprises: The parameter loading module loads the parameters for the heat load prediction model. The data acquisition submodule collects data on ambient temperature, number and duration of door openings, photovoltaic power generation, battery SOC, and internal temperature of the warehouse, using T2 as the time period. The heat load prediction submodule uses a heat load prediction model to calculate the cooling demand within a future time window. The control quantity prediction submodule determines whether the predicted heat load exceeds the threshold; if a high-temperature period or frequent door opening is predicted, it executes forward-looking control; when photovoltaic power is sufficient, it instructs the compressor to cool down in advance to build up cooling capacity reserves; the control output is the frequency adjustment amount of the third compressor. .

5. The solar-powered cold storage control system according to claim 4, wherein the energy efficiency adaptive optimization module further comprises: The second real-time controller adjusts the frequency of the third compressor. Used for the second target frequency Compensation is performed to obtain the third target frequency. The third target frequency As a real-time control command to adjust the compressor's operating state, this control command has a higher priority than the second target frequency. Control commands.

6. The solar-powered cold storage control system according to claim 4, wherein the heat load prediction model is: in, Predict heat load, For the ambient temperature, Warehouse door opening frequency It refers to photovoltaic power generation capacity. The battery is in its state of charge. , , , The coefficients of the model, This is the error term.

7. The solar-powered cold storage control system according to claim 4, wherein the frequency adjustment amount of the third compressor... The calculation formula is: in, This is the gain coefficient. This is the heat load threshold.

8. The solar-powered cold storage control system according to claim 1, wherein the multi-objective function is: in, Energy consumption cost is related to photovoltaic power output and grid replenishment costs. This is due to temperature deviation in the storage area; Battery loss rate is estimated based on the number of SOC cycles and charge / discharge rate; Compressor fatigue is calculated based on the number of start-stop cycles and load rate. , , , For weights.

9. The solar-powered cold storage control system according to claim 1, wherein the photovoltaic power optimization module specifically includes: The system initialization submodule is used to initialize the deep reinforcement learning agent, including its policy network, target network, and experience replay buffer; and to load the pre-trained network model weights. The state-aware submodule samples the output voltage of the photovoltaic array in each control cycle. Output current And calculate the current power. ; to include the current and N past moments Data as a state vector ; The action decision submodule inputs the state vector s(k) into the policy network of the DRL agent, and the network outputs an action a(k), which represents the adjustment command for the switching signal of the power converter. The "Execute Actions and Environmental Feedback" submodule performs the actions. This involves changing the system's operating parameters to adjust the power output state of the photovoltaic array; in the next cycle, the new power is sampled. The degree of improvement in power output characteristics is used as an immediate reward. ; The experience storage and learning submodule will store the current experience tuples. Store the data in the experience replay buffer; periodically sample a batch of experience data randomly from the buffer to update the weights of the policy network, with the goal of maximizing long-term cumulative rewards. The output control submodule converts the action output by the strategy network to obtain the first output power, and generates actual adjustment commands based on the first output power to drive the actuators in the power controller; and waits for the next control cycle to repeat the photovoltaic power optimization process.

10. A cold storage facility, the cold storage facility being implemented based on the cold storage control system according to any one of claims 1-9.

Images