Laboratory air-conditioning temperature control system and method based on artificial intelligence

Abstract

The invention discloses a laboratory air-conditioning temperature control system and method based on artificial intelligence, and relates to the field of laboratory temperature control. Optimizing the positions of the air supply and exhaust outlets by using a pre-constructed airflow analysis model; according to the optimized positions of the air supply and exhaust outlets, the air conditioning system is used for controlling airflow to flow to the corresponding air supply and exhaust outlets, and directional airflow from the first area to the second area is generated; according to laboratory electrical load data, an optimal charging and discharging strategy of an energy storage device is generated by using a reinforcement learning algorithm, and the reinforcement learning algorithm enables an intelligent agent to learn optimal strategies in different states by constructing a state space including electricity price, electrical load, energy storage electric quantity and photovoltaic power generation and setting a reward function with maximum load stability. Aiming at the low laboratory temperature control precision in the prior art, the temperature control precision is improved by simulating an airflow velocity field and radiation quantity distribution and optimizing the layout of air supply and exhaust outlets by using CFD simulation and a genetic algorithm.

Description

Technical Field

[0001] The present application relates to the field of laboratory temperature control, and in particular to an artificial intelligence-based laboratory air conditioning temperature control system and method. Background Art

[0002] With the rapid development of science and technology, various laboratories have increasingly higher requirements for environmental temperature control. In the fields of precision instruments, chemical experiments, life sciences, etc., temperature stability has become a key factor affecting the accuracy and repeatability of experiments. Taking biological laboratories as an example, even slight fluctuations in temperature may cause significant changes in the growth rate of bacterial colonies, thereby affecting the validity of experimental results. In precision measurement laboratories, thermal deformation of instruments and characteristic drift of electronic devices caused by temperature changes are one of the main sources of measurement errors. Therefore, high-precision, all-weather temperature control has become an indispensable basic condition for modern laboratories.

[0003] However, traditional laboratory air conditioning temperature control systems have many shortcomings in meeting the increasingly demanding temperature control needs. Most of these systems rely on manually set target temperatures and maintain constant temperatures through fixed PID control. However, the thermodynamic characteristics of the laboratory environment are complex, and temperature changes show obvious nonlinearity and uncertainty. It is difficult for traditional control algorithms to accurately model and dynamically respond to such complex characteristics, resulting in frequent fluctuations between the actual temperature and the target value, and it is impossible to stabilize near the set value. At the same time, this passive adjustment is also difficult to provide real-time feedback on sudden changes in the external environment, and the system's rapid response capability is insufficient. The manual tuning of PID parameters also makes the control performance limited by the operator's experience, and the robustness and adaptability are poor.

[0004] In terms of energy consumption management, it is also difficult to achieve fine optimization of traditional air conditioning systems. The frequent replacement of experimental equipment makes the heat load of the laboratory present dynamic characteristics, and the traditional constant temperature control mode cannot adjust the cooling output in time, which inevitably leads to excessive or insufficient cooling. Non-optimal air supply strategies may also lead to local overcooling and overheating, resulting in energy waste. Summary of the invention

[0005] In view of the low laboratory temperature control accuracy existing in the prior art, the present application provides a laboratory air conditioning temperature control system and method based on artificial intelligence, which improves the temperature control accuracy by simulating the air flow velocity field and radiation distribution, and optimizing the supply and exhaust air vent layout using CFD simulation and genetic algorithm.

[0006] The purpose of this application is achieved through the following technical solutions.

[0007] One aspect of the present application provides an artificial intelligence-based laboratory air conditioning temperature control system, including: an acquisition module for collecting laboratory data; a control module for optimizing the positions of supply and exhaust air vents based on the laboratory data using a pre-built airflow analysis model to obtain optimized supply and exhaust air vent installation positions; a ventilation module for generating a directional airflow from a first area to a second area using an air conditioning system based on the optimized supply and exhaust air vent positions; an energy storage module for generating an optimal charging and discharging strategy for an energy storage device based on laboratory power load data using a reinforcement learning algorithm; the reinforcement learning algorithm enables the intelligent agent to learn the optimal strategy under different states by constructing a state space including electricity price, power load, energy storage power and photovoltaic power generation, and setting a reward function for maximizing load stability.

[0008] Another aspect of the present application provides an artificial intelligence-based laboratory air conditioning temperature control method, including: S1, collecting laboratory data, wherein the laboratory data includes the temperature and humidity, wind speed, air volume, indoor radiation and outdoor radiation leakage of the radioactive laboratory; S2, according to the laboratory data, using a pre-constructed airflow analysis model to optimize the supply and exhaust air vent positions to obtain the optimized supply and exhaust air vent positions; S3, according to the optimized supply and exhaust air vent positions, using the air conditioning system to control the airflow to the corresponding supply and exhaust air vents, generating a directional airflow from the first area to the second area; and filtering the air exhausted from the laboratory at the corresponding supply and exhaust air vents; the first area is a clean area, and the second area is a polluted area; S4, according to the laboratory power load data, using a reinforcement learning algorithm to generate the optimal charging and discharging strategy of the energy storage device, the reinforcement learning algorithm constructs a state space including electricity price, power load, energy storage power and photovoltaic power generation, and sets a reward function that maximizes load stability, so that the intelligent agent learns the optimal strategy under different states.

[0009] Furthermore, S2, based on laboratory data, uses a pre-built airflow analysis model to optimize the supply and exhaust air vent positions to obtain optimized supply and exhaust air vent positions, including: S21, based on laboratory data, uses a pre-built airflow analysis model to calculate the airflow velocity, temperature and radiation distribution at different supply and exhaust air vent positions; S22, according to the temperature and humidity threshold and radiation threshold of the first area, as well as the airflow isolation threshold between the second area and the first area, a multi-objective function is set; S23, using the multi-objective function as a fitness function, and using a genetic algorithm to obtain the optimal supply and exhaust air vent positions.

[0010] Further, S21, based on the laboratory data, using the pre-built airflow analysis model, calculate the airflow velocity, temperature and radiation distribution at different supply and exhaust air outlet positions, including: S211, using CFD software to build a three-dimensional model of the laboratory content according to the laboratory's geometric dimensions and boundary conditions; S212, based on the three-dimensional model, using the temperature and humidity, wind speed and air volume in the laboratory data as the inlet boundary conditions, using the indoor radiation as the body heat source term, and using the outdoor radiation leakage as the radiation outlet boundary; S213, based on the three-dimensional model with the inlet and outlet boundaries set, constructing the Navier-S Tokes equations, energy equations and radiation transfer equations; use the standard k-ε turbulence model and discrete coordinate radiation model (DO) to describe the turbulence characteristics and radiation heat transfer characteristics of the flow field respectively; S214, couple the Navier-Stokes equations, energy equations and radiation transfer equations as the airflow analysis model; use the finite volume method to discretely solve the airflow analysis model to obtain the velocity field, temperature field and radiation intensity field inside the laboratory; S215, use the velocity field, temperature field and radiation intensity field to calculate the airflow velocity, temperature and radiation at different air outlet positions;

[0011] Specifically, the airflow in the laboratory is generally in a turbulent state, and the turbulence intensity and scale directly affect the velocity and temperature distribution of the flow field. The standard k-ε model introduces two variables, turbulent kinetic energy k and turbulent dissipation rate ε, to establish additional transport equations, which are coupled with the Navier-Stokes equations and the energy equations to better describe the time-averaged characteristics of turbulent flow. In step S214, the standard k-ε model provides closed turbulent stress terms and turbulent heat flux terms, which enable the Navier-Stokes equations and the energy equations to be solved in a closed manner, and the influence of turbulent pulsation on the flow field is taken into account. Through coupled solution, the velocity field and temperature field distribution considering the turbulence effect can be obtained, providing more accurate flow field information for the subsequent optimization of the supply and exhaust vent positions.

[0012] In addition to the convective heat transfer of the fluid inside the radioactive laboratory, there is also significant radiation heat transfer. The distribution of radiant heat is directly related to the temperature uniformity and local high temperature areas in the room. The discrete coordinate radiation model can simulate the propagation, absorption, emission and scattering process of radiation energy in the medium by discretizing the entire space into a series of discrete directions and solving the radiation intensity in each direction. In step S212, the indoor radiation source is added to the radiation transfer equation as a body heat source term and the outdoor radiation leakage is added as a boundary condition. In step S214, the radiation transfer equation is coupled with the Navier-Stokes equations and the energy equation, and the solution is unified to obtain the temperature field distribution after comprehensive consideration of convection and radiation heat transfer. In step S215, the radiation intensity field obtained by the solution can be used to calculate the radiation heat flux at different positions and evaluate the impact of radiation heat transfer on the local thermal environment. The standard k-ε turbulence model and the discrete coordinate radiation model improve the accuracy of numerical simulation from the perspectives of flow and heat transfer, respectively, and provide flow field temperature field information with smaller deviation and more realistic for the subsequent optimization of the supply and exhaust vent layout. During the numerical solution, the two models respectively introduce additional transport equations to form a closed set of equations with the basic fluid and heat transfer control equations, and finally obtain high-fidelity simulation results through coupled solution.

[0013] Among them, traditional methods often ignore radiation heat transfer or use simplified linear radiation models, which makes it difficult to accurately predict laboratory environmental parameters with significant radiation effects. On the one hand, this application uses a discrete coordinate (DO) model to describe the spatial angular distribution of radiation intensity, which is suitable for complex radiation environments such as absorption, emission and anisotropic scattering, and improves the accuracy of temperature and radiation prediction. On the other hand, this application makes comprehensive use of a variety of cutting-edge technologies in computational fluid dynamics (CFD), including Navier-Stokes equations, turbulence models, multi-physics field coupling, finite volume discretization, etc., to more completely restore the velocity field, temperature field and radiation intensity field inside the laboratory. In particular, the standard k-ε two-equation turbulence model is used, which can reasonably simulate the flow characteristics of the near-wall area and the fully developed turbulent area, and is suitable for engineering application conditions such as high Reynolds numbers and rotating flows.

[0014] Further, S22, according to the temperature and humidity threshold and the radiation threshold of the first area, and the airflow isolation threshold between the second area and the first area, a multi-objective function is set, including: according to the temperature threshold TL of the first area 1 and TH 1 , Radiation threshold QL 1 and QH 1 , construct the clean area temperature objective function FT 1 (x) and radiation objective function FQ 1(x); the independent variable X of the objective function is the supply and exhaust air vent position vector, each component represents the installation position of a candidate supply and exhaust air vent; according to the airflow isolation threshold CL between the second area and the first area, the airflow isolation objective function FC(x) is constructed; the first area objective function FT 1 (x) and FQ 1 (x), and the second regional objective function FC(x), are combined to form a multi-objective function F(x).

[0015] Further, S23, taking the multi-objective function as the fitness function, and using the genetic algorithm to obtain the optimal supply and exhaust air vent positions, including: S231, randomly generating N initial solutions to form an initial population, each initial solution corresponding to a group of candidate supply and exhaust air vent installation positions; S232, using a hybrid coding strategy to perform chromosome encoding on the initial solution, the chromosome consists of two parts, the first part is the real-coded spatial coordinates of the supply and exhaust air vents, and the second part is the integer-coded supply and exhaust air vent types; the first part and the second part of the chromosome are associated through a mapping relationship; S233, for each individual in the initial population, Utilize the airflow analysis model of S21 to calculate the airflow velocity, temperature and radiation distribution under the supply and exhaust air vents layout scheme corresponding to the individual, and substitute the calculation result into the multi-objective function F(x) constructed in S22 as the fitness value of the current individual; S234, according to the fitness value of the individual, perform the selection, crossover and mutation operations of the genetic algorithm on the current population to generate the next generation population; S235, repeat S233 and S234 until the convergence condition is met, select the individual with the highest fitness value from the population, decode the chromosome of the individual, and obtain the optimal supply and exhaust air vents layout scheme.

[0016] Among them, on the one hand, the optimization of the layout of supply and exhaust air vents is a multi-parameter, strongly coupled, and nonlinear combinatorial optimization problem. The simple use of real number coding can only capture the spatial position information of each air vent, ignoring the matching characteristics of different types of air vents, and the search space is relatively blind. After the application introduces integer coding of air vent types, the coordination mode of the air vent can be directly optimized, which greatly narrows the solution space, reduces invalid searches, and accelerates the convergence of the algorithm.

[0017] On the other hand, the hybrid coding strategy can flexibly handle discrete and continuous variables, which enhances the adaptability of the algorithm. In actual engineering, the layout plan of the supply and exhaust air vents usually needs to consider both continuous variables (such as vent coordinates) and discrete variables (such as vent type and size). Traditional binary coding or floating-point coding is difficult to effectively handle this hybrid optimization problem, while chromosome segments with different coding methods can distinguish the two types of variables, avoid artificial discretization errors, and improve the feasibility and accuracy of the layout plan. In the real number coding segment, each gene bit corresponds to an air vent coordinate, and the mutation operation can be locally fine-tuned within the installation range of the air vent; in the integer coding segment, each gene bit corresponds to an air vent type, and the mutation operation can jump between different types. The two mutation methods cooperate with each other, which can not only perform fine searches in the promising area, but also jump out of the local optimum and converge to the global optimal solution.

[0018] Furthermore, the clean area temperature objective function FT 1 (x) is:

[0019]

[0020] Among them, T min (x) and T max (x) are the minimum and maximum values ​​of the temperature distribution in the first region corresponding to the layout scheme X, which can be calculated by solving the Navier-Stokes equations and energy equation constructed in S213; TL 1 and TH 1 They are respectively the lower temperature threshold and the upper temperature threshold of the first area, which are preset according to the temperature requirements of the clean area.

[0021] Furthermore, the radiation objective function FQ 1 (x) is:

[0022]

[0023] Among them, Q min (x) and Q max (x) are the minimum and maximum values ​​of the radiation distribution in the first area corresponding to the layout scheme X, respectively; they can be calculated by solving the radiation transfer equation constructed in S213; QL 1 and QH 1 They are respectively the lower and upper thresholds of the radiation amount in the first area, and are preset according to the radiation level requirements of the clean area.

[0024] Furthermore, the airflow isolation objective function FC(X) is: Among them, C(X) represents the actual airflow isolation between the second area and the first area corresponding to the layout scheme X, and CL represents the target value of the airflow isolation between the two areas, which is a constant pre-set according to the protection level requirements of the clean area. The physical meaning of FC(X) is the relative deviation between the actual airflow isolation and the target airflow isolation. When C(X) is greater than or equal to CL, FC(X) is less than or equal to 0, indicating that the airflow isolation performance meets the requirements; when C(X) is less than CL, FC(X) is greater than 0, and the smaller C(X), the larger FC(X), indicating that the airflow isolation performance is worse and the deviation from the target value is greater.

[0025] Furthermore, the calculation formula of the actual airflow isolation C(X) is: Where n represents the total number of interfaces between the second region and the first region, Q i (X) represents the actual airflow exchange volume on the i-th interface under layout scheme X, Q i,ref It represents the reference airflow exchange volume on the i-th interface, which can be pre-set according to the protection level requirements of the clean area.

[0026] Q i The calculation formula of (X) is: Where m represents the total number of monitoring points on the i-th interface, q ij (X) represents the airflow exchange flux at the jth monitoring point on the i-th interface under layout scheme X.

[0027] q ij The calculation formula of (X) is: ij (X) = v ij (X)×(c 2ij (X)-c 1ij (X)), where v ij (X) represents the normal airflow velocity at the jth monitoring point on the i-th interface under layout scheme X, c 2ij (X) and c 1ij (X) represents the pollutant concentrations in the second area and the first area at the monitoring point respectively.

[0028] The actual clean room interface is often large in area and the airflow distribution is uneven. The formula is calculated by calculating the airflow exchange flux q at m monitoring points on each interface. ij (X) Take the average and get the overall airflow exchange volume Q of the interface i (X), and then evaluate the isolation performance of the entire clean room. This multi-point averaging method can more comprehensively reflect the overall isolation effect of the interface, avoid the evaluation error caused by the measurement deviation of individual points, and improve the reliability of isolation measurement. ij The calculation of (X) takes into account the influence of wind speed and concentration gradient.ij The calculation formula of (X) shows that the airflow exchange between the two areas depends not only on the normal wind speed v on the interface ij (X), and also the difference in pollutant concentration between the two sides (c 2ij (X)-c 1ij The greater the wind speed and the greater the concentration difference, the greater the airflow exchange flux and the worse the corresponding isolation effect. Considering the coupling effect of wind speed and concentration gradient, the impact of layout scheme on isolation effect can be more accurately characterized. 11.10

[0030] S4, based on the laboratory power load data, uses the reinforcement learning algorithm to generate the optimal charging and discharging strategy of the energy storage device. The reinforcement learning algorithm constructs a state space containing electricity price, power load, energy storage capacity and photovoltaic power generation, and sets a reward function that maximizes load stability, so that the intelligent agent learns the optimal strategy under different states, including:

[0031] S41, setting a state space S, wherein S includes a real-time electricity price s 1 , Laboratory power load 2 , energy storage device capacity s 3 and photovoltaic power generation 4 ;

[0032] Set the action space A, where A includes the charging power a of the energy storage device 1 and discharge power a 2 ;

[0033] Set the reward function R, Where P grid The power purchased by the laboratory from the power grid, P bat is the charging and discharging power of the energy storage device, cost is the operating cost, P pv,use is the actual photovoltaic power generation power, P pv,max is the rated power of the photovoltaic power station, λ 1 ,λ 2 ,λ 3 ,λ 4 is the weight coefficient; specifically, |P grid The term | describes the absolute value of the power purchased by the laboratory from the power grid, corresponding to the difference in peak and valley power. By introducing -λ in the reward function 1 ×|P grid |, it can punish high-power electricity purchase behavior, encourage intelligent agents to store more electricity when electricity prices are low, and release more electricity when electricity prices are high, thereby achieving peak-to-valley arbitrage and reducing system operating costs. Traditional energy storage scheduling methods often only consider electricity prices in a single period of time, making it difficult to learn the global optimal peak-to-valley arbitrage strategy. The term represents the actual utilization rate of photovoltaic power generation, reflecting the proportion of photovoltaic energy consumed on-site. It can reward the efficient use of photovoltaic energy, suppress the phenomenon of power curtailment, and increase the proportion of renewable energy supply to load. Traditional scheduling methods often regard photovoltaic output as a disturbance of load, and it is difficult to actively adapt to its intermittent and volatile nature. This application incorporates the photovoltaic consumption rate into the optimization target, promotes the synergy and complementarity of energy storage and photovoltaics, and maximizes the use value of photovoltaic energy.

[0034] S42, initialize the agent's strategy network π and value network V. The strategy network π takes the state S as input and outputs the probability distribution of the corresponding action A. The value network V takes the state S as input and outputs the corresponding state value. S43, according to the current state S t , using the policy network π to generate action A t , that is, the charging and discharging power of the energy storage device; according to action A t and environmental model, update the laboratory status to S t+1 , and calculate the reward value R t ; will (S t ,A t ,R t ,S t+1 ) is stored as a set of training data in the experience replay pool D; the above steps are repeated until the termination condition is met; S44, a batch of training data (S t ,A t ,R t ,S t+1 ), use the temporal difference algorithm to update the parameters of the policy network π and the value network V, so that the actions generated by the policy network π can maximize the long-term cumulative reward; S45, repeat steps S43 and S44, update the policy network π and the value network V, until the convergence conditions are met.

[0035] S46, using the trained strategy network π, according to the real-time state S of the laboratory, generate the optimal charging and discharging strategy A* of the energy storage device, A*=π(S), while satisfying the constraint condition a 1 ≤min(P pv,max ,P bat,max ), where P bat,max is the maximum charging power of the energy storage device; specifically, the real-time power generation power P of the photovoltaic power station pv,max It depends on external conditions such as the current light intensity and temperature, and is characterized by intermittency and fluctuation. When the photovoltaic output is large and the load demand is small, blindly storing all photovoltaic power in the energy storage device may cause the battery to overcharge, resulting in a waste of photovoltaic power. 1 No more than P pv,max, while meeting the charging demand, it can maximize the local consumption of photovoltaic power, avoid unnecessary abandonment of light and power rationing, and improve the utilization efficiency of photovoltaic power generation. bat,max It is determined by the physical and chemical properties of the battery. Charging beyond this power will cause the battery's internal resistance to rise sharply, resulting in dangerous situations such as overheating and swelling. In serious cases, it may cause safety accidents such as combustion and explosion. 1 Restricted to P bat,max It can effectively prevent battery overcharging, extend battery life, and improve system operation safety. Traditional energy storage scheduling algorithms often ignore charging power constraints, which poses a risk of battery overcharging.

[0036] S47, according to the charging power a in strategy A* 1 , control the energy storage device to power a 1 Charging from the grid or photovoltaic power station; according to the discharge power a in strategy A* 2 , control the energy storage device to power a 2 Supply power to the laboratory.

[0037] Compared with the prior art, the advantages of this application are:

[0038] The internal environment of the laboratory is a complex multi-physics coupling system, including fluid flow, heat conduction / convection / radiation, pollutant diffusion and other physical processes. These physical processes affect each other. For example, the flow direction and speed of the airflow will affect the temperature distribution and pollutant concentration distribution. Conversely, the temperature gradient and concentration gradient will drive the airflow. Traditional simplified models often ignore this multi-physics coupling effect, and solve the flow field, temperature field and concentration field independently, which cannot accurately capture the interaction between them. This solution uses CFD simulation technology to introduce the energy equation and radiation transfer equation into the basic equation of fluid mechanics (NS equations) to form a flow-heat-radiation coupling model. By coupling and solving the velocity field, temperature field and radiation intensity field, the mutual influence between multiple physical quantities can be fully considered to obtain more realistic and accurate results. This high-fidelity simulation lays the foundation for the subsequent supply and exhaust air optimization, and can accurately predict the distribution of environmental parameters under different supply and exhaust air layouts, avoiding the errors caused by model simplification. At the same time, the radiation transfer model is introduced to analyze the spatial distribution law of radiation intensity, optimize the supply and exhaust air scheme, and effectively block the migration and diffusion of radioactive aerosols according to the special needs of radioactive laboratories.

[0039] The optimization of the supply and exhaust air layout is a complex combinatorial optimization problem with multiple parameters, multiple constraints and multiple objectives. The enumeration method and gradient descent method commonly used in the past have the following problems: the enumeration method has a huge amount of calculation and it is difficult to enumerate all possible layout schemes; the gradient descent method requires the objective function to be differentiable and is prone to fall into local optimality. Genetic algorithm is a heuristic search algorithm based on biological evolution theory. It simulates the natural selection process of "survival of the fittest" to find the optimal solution in a complex solution space. It uses encoding, selection, crossover, mutation and other operations to iteratively evolve individuals with the highest fitness, which can not only jump out of the local optimality, but also converge efficiently within limited computing resources. This solution innovatively combines CFD simulation results with genetic algorithms. According to the CFD calculation results of each candidate supply and exhaust air layout, a fitness function is constructed and used as the environmental fitness evaluation index of the genetic algorithm. Through iterative evolution, the best layout scheme with excellent performance in multiple dimensions such as cleanliness, temperature and radiation is selected. At the same time, when constructing the fitness function, multiple control objectives such as temperature and humidity, cleanliness and radiation level of clean and contaminated areas are comprehensively considered, forming a "multi-objective optimization" problem. By weighing the Pareto optimal solution set, we seek a balance between various objectives and obtain a comprehensive optimal layout that takes into account all indicators. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The present application will be further described in the form of exemplary embodiments, which will be described in detail by the accompanying drawings. These embodiments are not restrictive, and in these embodiments, the same number represents the same structure, wherein:

[0041] Figure 1 is an exemplary flow chart of an artificial intelligence-based laboratory air conditioning temperature control method according to some embodiments of the present application;

[0042] Figure 2 is an exemplary flow chart of another method for controlling temperature of laboratory air conditioning based on artificial intelligence according to some embodiments of the present application;

[0043] Figure 3 This is an exemplary flow chart for determining the optimal supply and exhaust air vent positions according to some embodiments of the present application. DETAILED DESCRIPTION

[0044] The method and system provided in the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0045] like Figure 1As shown, laboratory data is collected, and the laboratory data includes the temperature and humidity, wind speed, air volume, indoor radiation and outdoor radiation leakage of the radioactive laboratory; according to the laboratory data, the pre-built airflow analysis model is used to optimize the supply and exhaust air vents to obtain the optimized supply and exhaust air vents; according to the optimized supply and exhaust air vents, the air conditioning system is used to control the airflow to the corresponding supply and exhaust air vents to generate a directional airflow from the first area to the second area; and the air exhausted from the laboratory is filtered at the corresponding supply and exhaust air vents; the first area is a clean area, and the second area is a polluted area; according to the laboratory power load data, the reinforcement learning algorithm is used to generate the optimal charging and discharging strategy of the energy storage device, and the reinforcement learning algorithm constructs a state space including electricity price, power load, energy storage power and photovoltaic power generation, and sets a reward function that maximizes load stability, so that the intelligent agent learns the optimal strategy under different states.

[0046] like Figure 2 As shown, the laboratory air conditioning temperature control system based on artificial intelligence provided by this application is applied to the biosafety laboratory (BSL-2+) and radioactive laboratory of a central building and the research building of the Institute of Environment. First, for the airtightness and directional flow field control of the P2 biosafety laboratory, high-performance airtight materials and construction processes, such as anti-penetration sealants, electromagnetic shielding silica gel, etc., are used in the enclosure structures such as doors, windows, walls, and floor slabs of the laboratory to prevent gas leakage from the source and ensure the high airtightness of the laboratory. In the design and construction of the ventilation system, according to the method of this application, the layout of the air supply and exhaust vents is optimized. Through CFD simulation analysis and genetic algorithm optimization, the optimal air supply vent scheme with uniform local wind speed and reasonable airflow organization is obtained, and the construction and installation are carried out strictly in accordance with the optimization results. The directional flow field control strategy from the clean area to the contaminated area is enabled. A high-density hundred-level air supply vent is arranged in the clean area, and a negative pressure exhaust vent is arranged in the contaminated area to form a directional airflow in the room, and the airflow is guaranteed not to be reversed by controlling the room pressure difference. A high-efficiency filter (HEPA) is installed in the exhaust branch to fully filter and purify the exhaust gas, remove microbial contamination, achieve clean exhaust, and avoid pollution to the external environment.

[0047] Secondly, in several laboratories, such as the Co60γ-ray radiotherapy standard laboratory, afterloading therapy and brachytherapy dose standard laboratory, high-precision temperature and humidity sensors, wind speed and volume sensors and radiation monitors are deployed to collect environmental parameters and radiation data in real time, providing data support for temperature control optimization. Through data analysis and radiation field simulation, the distribution characteristics of radiation heat load in different laboratories are accurately mastered, and the thermal and humid environment of the room is optimized and controlled based on this, overcoming the problem of large thermal disturbance in the radiation environment. Intelligent control modules are built into environmental air-conditioning units and various terminal devices. Through the reinforcement learning algorithm, the operation strategy of air-conditioning is optimized to meet the temperature and humidity control requirements of 18-26℃ and 30%-70%, while achieving energy saving and efficiency improvement. In response to the high requirements of system joint debugging, an intelligent group control coordination strategy has been developed. The environmental air-conditioning unit adopts a main-standby redundant design. When the main unit fails, the standby unit can quickly cut in within 1 minute to restore the temperature and humidity conditions of the room, and ensure the stability of the pressure gradient of the adjacent room, avoiding the risk of negative pressure room becoming positive pressure room. A centralized monitoring platform based on digital twins has been built to uniformly manage 1,039 air-conditioning terminals and 3,886 sensors, control valves and other points, realizing visual holographic perception and macro-control of the experimental environment, and providing laboratory managers with an intuitive and friendly human-computer interaction interface.

[0048] S1, collect laboratory data, arrange temperature and humidity sensors in different functional areas of the laboratory, and select high-precision Pt100 platinum resistance temperature sensors and capacitive humidity sensors. Install wind speed sensors and air volume measuring devices at the air supply and exhaust vents of the laboratory. The wind speed sensor uses a thermistor anemometer. For radioactive laboratories such as the Co60 gamma-ray radiotherapy standard laboratory, gamma radiation dose rate meters and neutron dose equivalent rate meters are arranged at different locations inside. The gamma radiation dose rate meter uses a high-voltage ionization chamber with a range of 0.1μSv / h to 10Sv / h, and the neutron dose equivalent rate meter uses a 3He proportional counter tube with a range of 0.1μSv / h to 10mSv / h.

[0049] like Figure 3 As shown, S2, based on the laboratory data, the supply and exhaust air vents are optimized using a pre-built airflow analysis model to obtain the optimized supply and exhaust air vents positions, including: S21, using the computational fluid dynamics (CFD) method to perform numerical simulation analysis on the airflow, temperature and radiation distribution inside the laboratory to provide a quantitative basis for the optimization of the supply and exhaust air vents. The specific implementation is as follows:

[0050] S211, based on the floor plan and cross-section of the laboratory, construct a three-dimensional geometric model of the laboratory in CFD pre-processing software (such as ICEM CFD, Gambit, etc.), and simplify it appropriately to remove detailed structures that have little effect on the airflow and improve the efficiency of meshing. Mesh the simplified geometric model, and encrypt the mesh in the near-wall area (such as near the supply and exhaust vents) to ensure accurate capture of the boundary layer flow. The entire calculation domain uses an unstructured tetrahedral mesh, and the number of cells is controlled within 5 million to balance the calculation accuracy and efficiency. Check the mesh quality, such as mesh size, aspect ratio, distortion and other indicators to ensure that the mesh meets the requirements of the CFD solver.

[0051] S212, based on the air supply speed and air volume in the laboratory data, combined with the area of ​​the air supply outlet, calculate the average wind speed of each air supply outlet as the velocity inlet boundary; at the same time, the air supply temperature and humidity are used as the inlet temperature and water vapor mass fraction boundaries. The exhaust outlet is set as the pressure outlet boundary, and the pressure is set to standard atmospheric pressure. The solid boundaries such as walls, floors, and ceilings are set as adiabatic no-slip walls, and the mahjong radiation outlet boundary (outdoor radiation leakage) is set as a radiation flux boundary. The indoor radiation is converted into radiation heat flux density and added to the corresponding calculation unit as the source term of the energy equation.

[0052] S213, based on the three-dimensional model with the inlet and outlet boundaries set, first, establish the continuity equation, which describes the law of conservation of fluid mass, that is, the fluid mass will not be created or disappeared out of thin air during the transport process in space. Where ρ is the fluid density, t is the time, U is the velocity vector, Represents the mass flux of the fluid; using the finite volume method, the grid cells are integrated to obtain Gauss's theorem is used to transform the volume integral into the surface integral, which is then discretized using the difference format to obtain an algebraic equation.

[0053] The momentum equation (NS equation) describes the motion law of fluid under the action of pressure gradient, viscous stress, gravity, etc. It is the embodiment of Newton's second law in fluid. Where p is the pressure, μ is the dynamic viscosity, S is the source term (such as gravity), represents the convection term; represents the viscous stress term; the finite volume method is also used, the second-order upwind format is used for the nonlinear convection term, and the central difference format is used for the diffusion term, which is then coupled with the time term to form a set of algebraic equations.

[0054] The energy equation describes the temperature change law of the fluid under heat exchange such as convection, heat conduction, and radiation, and reflects the law of conservation of heat. Where T is temperature, k is thermal conductivity, q is radiation heat flux, Φ is viscous dissipation term, represents the heat conduction term, Represents the radiation term; the finite volume method is used for discretization, the convection term adopts the second-order upwind method, the diffusion term adopts the central difference method, and the radiation term is solved by the DO model, as detailed below.

[0055] On the other hand, most of the airflow in the laboratory is in a turbulent state, showing strong pulsation characteristics. Therefore, this application simulates the influence of turbulent pulsation on the time-averaged flow field by solving the transport equations of turbulent kinetic energy k and turbulent dissipation rate ε, closes the Reynolds stress term, and makes the momentum equation solvable. Specifically, the turbulent kinetic energy equation is:

[0056] Turbulence dissipation rate equation:

[0057] in, is the turbulent viscosity, G k is the turbulent kinetic energy generation term, σ k ,σ ε ,C 1ε ,C 2ε ,C μ is the model constant. More specifically, by modeling two key statistics of turbulent pulsation (turbulent kinetic energy and dissipation rate), the turbulent pulsation intensity, turbulent mixing degree and other characteristics are well predicted, which can meet the requirements of engineering practice for turbulence simulation accuracy and computational efficiency. In the simulation of laboratory airflow organization, turbulent pulsation has an important influence on local heat transfer and pollutant transport. The standard k-ε model can capture these turbulent effects more accurately, providing a reliable flow basis for the prediction of laboratory environmental parameter distribution.

[0058] On the one hand, for radioactive laboratories, radiation heat transfer is an important factor affecting the indoor temperature distribution. The DO model solves the radiation transfer equation (RTE) to simulate the absorption, emission and scattering process of radiation energy in the medium, obtains the radiation heat flux distribution, and realizes the coupling of radiation and convection heat transfer. Radiation transfer equation: Among them, I is the radiation intensity, r is the position vector, s is the direction vector, β is the medium extinction coefficient, and S is the radiation source term, including the medium emission term and the scattering term. RTE is discretized into a series of directional differential equations in space, angle and wavelength, and the finite volume method is used to solve it to obtain the radiation intensity distribution in each discrete direction, and then the total radiation heat flux is obtained by angle integration. The DO model can accurately predict the impact of the radiation heat flux distribution on the indoor temperature field by finely characterizing the radiation transfer process. Compared with simplified radiation models (such as P1 and Rosseland), the DO model takes into account the anisotropy of radiation, is suitable for optically thick and thin media, and has better adaptability to complex boundary conditions and scattering effects. In the simulation of radioactive laboratory environment, the introduction of the DO model can quantitatively evaluate the impact of radioactive substances on indoor radiation levels and accurately capture the coupling effect of radiation and convection, thereby providing a reliable basis for laboratory radiation protection and temperature control scheme optimization.

[0059] S214, the control equations are spatially discretized using the finite volume method to obtain a set of algebraic equations. The SIMPLE algorithm is used to achieve the coupling of the velocity field and the pressure field, and the energy equation and the DO equation are coupled with the momentum equation for iterative solution. The calculation convergence criterion is that the residual of each physical quantity is less than 10-6 and the parameter change rate of the monitoring point is less than 0.1%. After obtaining the converged solution, the velocity vector, temperature scalar and radiation intensity vector data of each grid unit inside the laboratory are extracted.

[0060] S215, import the grid data calculated by S214 into the CFD post-processing software. Define the supply and exhaust air vents measuring points in different areas of the laboratory, and extract the velocity, temperature and radiation intensity values ​​of the grid units where they are located. Perform area-weighted averaging on the values ​​of multiple grid units of the same supply and exhaust air vents to obtain the average velocity, temperature and radiation parameters of the supply and exhaust air vents. Repeat the above process for different supply and exhaust air vent layout schemes to obtain the airflow parameter distribution of each supply and exhaust air vent under different schemes. In this application, radiation transfer is introduced into conventional flow field-temperature field analysis, which can accurately evaluate the impact of radioactive substances on the distribution of indoor environmental parameters, which has important guiding significance for the optimization of supply and exhaust air in radioactive laboratories.

[0061] S22, according to the temperature and humidity threshold and the radiation threshold of the first area, and the airflow isolation threshold between the second area and the first area, setting a multi-objective function; including:

[0062] Clean area temperature objective function FT 1 (x), describes the temperature distribution of the clean area under the candidate layout scheme X to meet the regional temperature requirements [TL 1 ,TH 1 ] degree. When the temperature distribution falls completely within the target range, FT 1(x) is 0, indicating the optimal temperature performance; when the temperature distribution exceeds the target range, FT 1 (x) is greater than 0, and the greater the deviation, the FT 1 The larger the (x), the worse the temperature performance. The CFD method is used to solve the temperature distribution corresponding to the layout scheme X and extract the maximum temperature T in the clean area. max (x) and minimum temperature T min (x), and we can calculate FT by substituting it into the above formula. 1 (x).

[0063] Clean area radiation objective function FQ 1 (x), describes the radiation distribution of the clean area under the candidate layout scheme X to meet the regional radiation level requirements [QL 1 ,QH 1 ] level. When the radiation dose falls completely within the target range, FQ 1 (x) is 0, indicating the optimal radiation protection performance; when the radiation dose exceeds the target range, FQ 1 (x) is greater than 0, and the greater the deviation, the FQ 1 The larger the (x), the worse the radiation protection performance. The DO radiation model is used to solve the radiation intensity field corresponding to the layout scheme X and extract the maximum radiation Q in the clean area. max (x) and minimum radiation Q min (x), and FQ can be calculated by substituting it into the above formula 1 (x).

[0064] The airflow isolation objective function FC(X) describes the degree to which the actual airflow isolation C(X) between the contaminated area and the clean area deviates from the target isolation CL under the candidate layout scheme X. When C(X) is greater than or equal to CL, FC(X) is less than or equal to 0, indicating that the airflow isolation performance meets the requirements; when C(X) is less than CL, FC(X) is greater than 0, and the smaller C(X), the larger FC(X), indicating that the airflow isolation performance is worse. in,

[0065] Q i (X) is the actual airflow exchange amount on the i-th interface, Q i,ref It is used as a reference for air flow exchange and can be set according to the protection level requirements. Among them, q ij (X) is the airflow exchange flux at the jth monitoring point of the i-th interface. ij (X) = v ij (X)×(c 2ij (X)-c 1ij (X)), where v ij (X) is the normal velocity at the monitoring point, c2ij (X) and c 1ij (X) are the pollutant concentrations in the polluted area and the clean area, respectively, which can be solved by the Species transport equation.

[0066] Construct a multi-objective function F(X), F(X) = {FT 1 (X),FQ 1 (X), FC(X)}, the temperature performance, radiation protection performance and airflow isolation performance of the clean area are taken as three parallel optimization objectives to form a multi-objective function vector. In this embodiment, F(X) = w 1 ×FT 1 (X)+w 2 ×FQ 1 (X)+w 3 ×FC(X), where w 1 ,w 2 ,w 3 is the weight coefficient of each sub-goal.

[0067] S23, taking the multi-objective function as the fitness function, and using the genetic algorithm to obtain the optimal air supply and exhaust vent positions. Including:

[0068] S231, determine the population size N, which is generally an integer between 20 and 100, taking into account both search efficiency and global optimization capability. Randomly generate N initial solutions, each of which corresponds to a set of supply and exhaust vent installation locations. Latin hypercube sampling and other methods can be used to generate representative uniformly distributed initial solutions to accelerate algorithm convergence. The N initial solutions form the initial population Population(0) as the starting point of the genetic algorithm.

[0069] S232, adopts a hybrid coding strategy, combining the advantages of real number coding and integer coding. The chromosome is divided into two parts: the first part is the real number coded spatial coordinates (x, y, z) of the air supply and exhaust outlets, and each air supply and exhaust outlet corresponds to three consecutive gene positions; the second part is the integer coded air supply and exhaust outlet type (such as air supply outlet, exhaust outlet, equalizing outlet, etc.), and each air supply and exhaust outlet corresponds to a discrete gene position. The two parts of the chromosome are associated through a one-to-one mapping of gene positions. For example, the 1st to 3rd bits of the chromosome in the first part represent the spatial coordinates of the first air supply and exhaust outlet, the 1st bit of the chromosome in the second part represents the type of the first air supply and exhaust outlet, and so on. Complete the chromosome encoding of all individuals in the initial population and obtain the initial search space of the genetic algorithm.

[0070] S233, for each individual in the initial population, extract its chromosome information and decode it to obtain the corresponding supply and exhaust vent layout plan. Input the layout plan into the airflow analysis model constructed in S21, and use the CFD method to calculate its corresponding airflow velocity field, temperature field and radiation intensity field. Substitute the CFD calculation results into the multi-objective function F(X) constructed in S22 to calculate the value of each sub-objective function FT 1 (x), FQ 1 (x) and FC(X), and then perform weighted summation according to the set weight coefficient to obtain the fitness value Fitness(X) of the current individual. Complete the fitness evaluation of the entire population and obtain the fitness distribution of the population.

[0071] S234, using the roulette wheel selection method, the probability of being selected is determined according to the fitness value of the individual. The higher the fitness value, the greater the probability of being selected. Keep the population size unchanged during the selection process. Use a combination of arithmetic crossover and uniform crossover. For the real-coded spatial coordinates, arithmetic crossover is used to generate a linear combination between two crossover points; for the integer-coded air outlet types, uniform crossover is used to randomly determine which parent individual each gene position is taken from. The crossover probability is between 0.6 and 0.9. Use uniform mutation. For the real-coded spatial coordinates, random perturbations are added to the mutation points, and the perturbation amount follows a normal distribution; for the integer-coded air outlet types, the gene values ​​at the mutation points are replaced with other types with a certain probability. The mutation probability is between 0.01 and 0.1. After completing the genetic operation, the newly generated offspring individuals are merged with the parent individuals to form a new population Population(t+1), and a new round of iteration begins.

[0072] S235, repeat S233 and S234, set the maximum evolutionary number T max (e.g. 100 to 500), if the current evolutionary generation t reaches T max , the algorithm is terminated and the individual with the highest fitness value in the current population is output as the optimal solution. Set the minimum fitness change threshold ε (such as 0.0001). If the average fitness change of several consecutive generations of populations is less than ε, the algorithm is considered to have converged and the iteration is terminated. Decode the optimal individual to obtain the specific parameters of the optimal supply and exhaust air outlet layout plan, including the spatial position coordinates and type of each supply and exhaust air outlet.

[0073] Genetic algorithm is an efficient method for solving the optimization problem of air outlet layout. Compared with the traditional enumeration method and random search method, genetic algorithm can quickly locate the global optimal area in the complex solution space by simulating the biological evolution mechanism, using group search and intelligent operation, and can effectively handle large-scale, multi-constrained, nonlinear combination optimization problems.

[0074] S3, control the air outlets. According to the position coordinates and types of the air outlets optimized in S23, several air outlets are arranged in the clean area. The variable air volume control technology of the air conditioning system is used to adjust the air volume and wind direction angle of each air outlet so that the air flow is directed to the contaminated area. A high-efficiency air filter (HEPA) is used to filter the fresh air sent into the clean area, and the filtration efficiency reaches 99.97% (for particles of 0.3μm) to ensure the level of the clean area. The air volume ratio of each air outlet is adjusted to form a uniform positive pressure environment in the clean area, and the pressure difference is not less than 5Pa (corresponding to a Class 100 clean area).

[0075] Control the exhaust vents. According to the exhaust vent location coordinates and types obtained by S23 optimization, several exhaust vents are arranged in the contaminated area. The variable air volume control technology of the air conditioning system is used to adjust the air volume of each exhaust vent to form a negative pressure environment in the contaminated area with a pressure difference of not less than 5Pa. The dirty air in the contaminated area is discharged to the outside through the exhaust duct, and the exhaust air is filtered by an activated carbon adsorption device to remove pollutants such as radioactive aerosols and organic waste gas. The total exhaust volume should be greater than the total supply volume to maintain negative pressure in the contaminated area and prevent the spread of pollutants.

[0076] Reasonably arrange the air flow distribution devices inside the clean area and the contaminated area, such as high-efficiency air supply ports and air suction hoods, to form directional airflow. Set up a downward-sending and upward-returning airflow organization in the clean area, and use laminar air supply and floor air supply systems to form a stable unidirectional flow in the working area. Set up an upward-sending and downward-returning airflow organization in the contaminated area, and use the top air supply and ground channel exhaust system to form turbulence around the pollution source to enhance the dilution and capture of pollutants. Install high-efficiency filters on the air supply and exhaust ducts to purify the supply and exhaust air.

[0077] Install high-efficiency filters on the air supply and exhaust ducts to purify the supply and exhaust air. The air supply branch in the clean area uses HEPA high-efficiency filters to meet the cleanliness level requirements. The exhaust branch in the polluted area uses a combination of chemical filters (activated carbon) and HEPA filters to remove chemical pollutants first and then remove particulate matter. Regularly replace and leak-check filters, and establish a filter ledger management system to ensure filtration efficiency.

[0078] S4, this step uses the reinforcement learning algorithm to autonomously explore the optimal control strategy of the energy storage device through interactive learning between the intelligent agent and the environment, realize peak shaving and valley filling of laboratory electricity load and local consumption of photovoltaic power, and improve the economy and stability of system operation.

[0079] S41, set the state space S, s 1 Real-time electricity price, reflecting the supply and demand balance of the power grid, can be obtained from the power trading system, in units of RMB / kWh. 2The laboratory power load reflects the power demand of the laboratory and can be obtained from the power meter or energy consumption monitoring system in kW. 3 The power of the energy storage device reflects the remaining capacity of the energy storage device and can be obtained from the battery management system (BMS) in kWh. 4 Photovoltaic power generation reflects the power generation capacity of a photovoltaic power station and can be collected from an inverter or combiner box in kW.

[0080] Set the action space A, a 1 The charging power of the energy storage device indicates the power absorbed by the energy storage device from the grid or photovoltaic power station. The unit is kW, and a positive value indicates charging. 2 The energy storage device discharge power represents the power supplied by the energy storage device to the laboratory. The unit is kW, and a positive value indicates discharge. Constraints: a 1 ≤min(P pv,max ,P bat,max ), where P pv,max is the rated power of the photovoltaic power station, P bat,max is the maximum charging power of the energy storage device; 0≤a 2 ≤min(P l ,P bat,max ), where P l It is the electrical load of the laboratory.

[0081] Set the reward function R, The laboratory purchases electricity from the power grid, P grid =P l -P pv -P bat , P pv Represents photovoltaic power generation; P bat Indicates the discharge power of the energy storage device; the unit is kW, a positive value indicates purchasing electricity, and a negative value indicates surplus power is connected to the grid. bat =a 1 -a 2 , in kW, positive values ​​indicate charging, negative values ​​indicate discharging. cost: operating cost, including electricity, equipment depreciation, and operation and maintenance fees, etc., calculated according to the cost model, in yuan. pv,use The actual photovoltaic power generation power, P pv,use =min(P pv ,P l +a 1 ), unit is kW. 1 ,λ 2 ,λ 3 ,λ 4 The weight coefficient reflects the importance of different optimization objectives and can be adjusted according to actual needs.

[0082] S42, initialize the policy network π, π: S->A, which represents the mapping from the state space S to the action space A. Use a deep neural network (DNN) to construct π, the number of input layer nodes is the state dimension |S|, and the number of output layer nodes is the action dimension |A|. Randomly initialize the network parameters of π, such as Xavier initialization.

[0083] Initialize the value network V, V: S->R, which represents the mapping from the state space S to the real number space R, and is used to estimate the long-term value of the state. Use DNN to construct V, with the number of input layer nodes being |S| and the number of output layer nodes being 1. Randomly initialize the network parameters of V.

[0084] S43, according to the current state S t =(s 1t ,s 2t ,s 3t ,s 4t ), S t Input into the policy network π to generate action A t =(a 1t ,a 2t ). Apply the action At to the environment, that is, control the energy storage device to generate power a 1t Charging, with power a 2t Discharge. Environment according to action A t and the system model, updated to the next state S t+1 , and returns the reward R t . Environmental status update formula: S t+1 =(s 1(t+1) ,s 2(t+1) ,s 3(t+1) ,s 4(t+1) );Reward calculation formula: (S t ,A t ,R t ,S t+1 ) as a set of training data and stored in the experience replay pool D. Repeat steps 1-4 until the termination condition is met (such as reaching the maximum number of interaction steps).

[0085] S44, randomly sample a batch of training data (S, A, R, S') from the experience replay pool D. Use the temporal difference (TD) algorithm to update the parameters θ of the value network V v :Calculate the TD target: y = R + γV(S'), where γ is the discount factor. Calculate the TD error: δ = yV(S) = (R + γV(S')) - V(S). Update θ using gradient descent v : Where α is the learning rate.

[0086] Using the policy gradient (PG) algorithm, update the parameters θ of the policy network π π : Calculate the action advantage function: A(S,A)=Q(S,A)-V(S), where Q(S,A) is the state-action value function, which can be obtained by Monte Carlo estimation or TD estimation. S45, repeat S43 and S44, continuously update the policy network π and the value network V, so that π gradually converges to the optimal policy π*. Convergence judgment condition: the improvement of the policy performance is less than the preset threshold. Reach the maximum number of training rounds and meet other predefined convergence criteria.

[0087] S46, using the trained strategy network π*, according to the real-time state S of the laboratory, generate the optimal charging and discharging strategy A* for the energy storage device. Strategy generation formula: A*=π(S), that is, input the state S into π, and obtain the optimal action A*=(a 1 *,a 2 *). Constraint processing: Set the charging power of A to a 1 *Limited to a 1 *∈[0,min(P pv,max ,P bat,max )] interval, the discharge power a 2 *Limited to a 2 *∈[0,min(P l ,P bat,max )] interval.

[0088] S47, charge and discharge control of the energy storage device, according to the charging power a1 in the optimal strategy A, control the energy storage device to charge at a constant power a 1 *Charge from the grid or photovoltaic power station. 1 *>P pv When a is used, photovoltaic power is used for charging first, and the shortfall is charged by purchasing power from the grid. 1 *≤P pv When the power is only charged by photovoltaic power, the remaining power is fully connected to the grid. 2 , control the energy storage device to a constant power a 2 *Supply power to the laboratory. When P l >P pv +a 2 *, the energy storage device discharges and the photovoltaic power generation jointly supplies the laboratory load, and the shortfall is purchased from the grid. l ≤P pv +a 2 *, the energy storage device discharges and photovoltaic power generation is prioritized to supply the laboratory load, and the remaining power is fully connected to the grid. The charge and discharge control cycle is synchronized with the optimal strategy generation cycle, such as 15 minutes, 1 hour, etc.

Claims

1. A laboratory air conditioning temperature control system based on artificial intelligence, characterized in that: include: Collection module, collects laboratory data; The control module optimizes the positions of the supply and exhaust air vents based on the laboratory data using a pre-built airflow analysis model to obtain the optimized installation positions of the supply and exhaust air vents; A ventilation module, based on the optimized supply and exhaust air outlet positions, generates a directional airflow from the first area to the second area using an air conditioning system; The energy storage module uses a reinforcement learning algorithm to generate the optimal charging and discharging strategy for the energy storage device based on the laboratory power load data; The reinforcement learning algorithm enables the agent to learn the optimal strategy under different states by constructing a state space including electricity price, electricity load, energy storage capacity and photovoltaic power generation, and setting a reward function to maximize load stability.

2. A laboratory air conditioning temperature control method based on artificial intelligence, characterized in that: include: S1, collecting laboratory data, wherein the laboratory data includes laboratory temperature and humidity, wind speed, wind volume, indoor radiation level of the radioactive laboratory, and outdoor radiation leakage; S2, based on the laboratory data, the supply and exhaust air vents are optimized using the pre-built air flow analysis model to obtain the optimized supply and exhaust air vents positions; S3, according to the optimized supply and exhaust air outlet positions, the air conditioning system is used to control the air flow to the corresponding supply and exhaust air outlets, generating a directional airflow from the first area to the second area; and the air exhausted from the laboratory is filtered at the corresponding supply and exhaust air outlets; the first area is a clean area, and the second area is a contaminated area; S4, based on the laboratory electricity load data, uses the reinforcement learning algorithm to generate the optimal charging and discharging strategy of the energy storage device. The reinforcement learning algorithm constructs a state space containing electricity price, electricity load, energy storage capacity and photovoltaic power generation, and sets a reward function that maximizes load stability, so that the intelligent agent can learn the optimal strategy under different states.

3. The laboratory air conditioning temperature control method based on artificial intelligence according to claim 2 is characterized in that: S2, based on laboratory data, uses the pre-built airflow analysis model to optimize the supply and exhaust air vent positions, and obtains the optimized supply and exhaust air vent positions, including: S21, based on laboratory data, using the pre-built airflow analysis model, calculate the airflow velocity, temperature and radiation distribution at different air supply and exhaust vent locations; S22, setting a multi-objective function according to the temperature and humidity threshold and the radiation threshold of the first area, and the airflow isolation threshold between the second area and the first area; S23, taking the multi-objective function as the fitness function, and using a genetic algorithm to obtain the optimal supply and exhaust air vent positions.

4. The laboratory air conditioning temperature control method based on artificial intelligence according to claim 3 is characterized in that: S21, based on laboratory data, uses a pre-built airflow analysis model to calculate the airflow velocity, temperature, and radiation distribution at different air outlet locations, including: S211, construct a three-dimensional model of the laboratory using CFD software based on the laboratory's geometric dimensions and boundary conditions; S212, using the temperature, humidity, wind speed and air volume in the laboratory data as the inlet boundary conditions, the indoor radiation as the body heat source term, and the outdoor radiation leakage as the radiation outlet boundary, to set up a three-dimensional model; S213, based on the three-dimensional model with inlet and outlet boundaries set, the Navier-Stokes equations, energy equations and radiation transfer equations are constructed; the standard k-ε turbulence model and the discrete coordinate radiation model DO are used to describe the turbulence characteristics and radiation heat transfer characteristics of the flow field respectively; S214, the Navier-Stokes equations, energy equations and radiation transfer equations are coupled to serve as the airflow analysis model; the airflow analysis model is discretely solved using the finite volume method to obtain the velocity field, temperature field and radiation intensity field inside the laboratory; S215, using the velocity field, temperature field and radiation intensity field, calculate the air flow velocity, temperature and radiation at different air supply and exhaust vent positions.

5. The laboratory air conditioning temperature control method based on artificial intelligence according to claim 4 is characterized in that: S22, according to the temperature and humidity threshold and the radiation threshold of the first area, and the airflow isolation threshold between the second area and the first area, a multi-objective function is set, including: According to the temperature thresholds TL1 and TH1 and the radiation thresholds QL1 and QH1 of the first area, the clean area temperature objective function FT1(x) and the radiation objective function FQ1(x) are constructed; the independent variable X of the objective function is the supply and exhaust air vent position vector, and each component represents the installation position of a candidate supply and exhaust air vent; According to the airflow isolation threshold CL between the second area and the first area, construct an airflow isolation objective function FC(x); The first region objective functions FT1(x) and FQ1(x), and the second region objective function FC(x) are combined to form a multi-objective function F(x).

6. The method for controlling temperature of laboratory air conditioning based on artificial intelligence according to claim 5, characterized in that: S23, taking the multi-objective function as the fitness function, and using the genetic algorithm to obtain the optimal supply and exhaust air outlet positions, including: S231, randomly generate N initial solutions to form an initial population, each initial solution corresponding to a set of candidate supply and exhaust vent installation positions; S232, a hybrid coding strategy is used to encode the initial solution by chromosome, wherein the chromosome consists of two parts, the first part is the space coordinates of the air outlet vents encoded by real numbers, and the second part is the type of the air outlet vents encoded by integers; the first part and the second part of the chromosome are associated through a mapping relationship; S233, for each individual in the initial population, the airflow analysis model of S21 is used to calculate the airflow velocity, temperature and radiation distribution under the supply and exhaust air outlet layout scheme corresponding to the individual, and the calculation result is substituted into the multi-objective function F(x) constructed in S22 as the fitness value of the current individual; S234, performing selection, crossover and mutation operations of the genetic algorithm on the current population according to the fitness value of the individual to generate the next generation population; S235, repeatedly executing S233 and S234 until the convergence condition is met, selecting the individual with the highest fitness value from the population, decoding the chromosome of the individual, and obtaining the optimal supply and exhaust air vent layout solution.

7. The laboratory air conditioning temperature control method based on artificial intelligence according to claim 5 is characterized in that: The clean area temperature objective function FT1(x) is: Among them, T min (x) and T max (x) are the minimum and maximum values ​​of the temperature distribution in the first area corresponding to the layout scheme X, which can be calculated by solving the Navier-Stokes equations and energy equation constructed in S213; TL1 and TH1 are the lower and upper temperature thresholds of the first area, which are pre-set according to the temperature requirements of the clean area.

8. The laboratory air conditioning temperature control method based on artificial intelligence according to claim 5 is characterized in that: The radiation objective function FQ1(x) is: Among them, Q min (x) and Q max (x) are the minimum and maximum values ​​of the radiation distribution in the first area corresponding to the layout scheme X; QL1 and QH1 are the lower and upper thresholds of the radiation in the first area, which are pre-set according to the radiation level requirements of the clean area.

9. The laboratory air conditioning temperature control method based on artificial intelligence according to claim 5 is characterized in that: The airflow isolation objective function FC(X) is: Wherein, C(X) represents the actual airflow isolation between the second area and the first area corresponding to the layout scheme X, and CL represents the target value of the airflow isolation between the two areas, which is a constant pre-set according to the protection level requirements of the clean area; The physical meaning of FC(X) is the relative deviation between the actual airflow isolation and the target airflow isolation. When C(X) is greater than or equal to CL, FC(X) is less than or equal to 0, indicating that the airflow isolation performance meets the requirements. When C(X) is less than CL, FC(X) is greater than 0, and the smaller C(X), the larger FC(X), indicating that the airflow isolation performance is worse and the deviation from the target value is greater.

10. The method for controlling temperature of laboratory air conditioning based on artificial intelligence according to claim 9, characterized in that: The calculation formula for the actual airflow isolation C(X) is: Where n represents the total number of interfaces between the second region and the first region, Q i (X) represents the actual airflow exchange volume on the i-th interface under layout scheme X, Q i,ref It represents the reference airflow exchange volume on the i-th interface, which can be preset according to the protection level requirements of the clean area; Q i The calculation formula of (X) is: Where m represents the total number of monitoring points on the i-th interface, q ij (X) represents the airflow exchange flux at the jth monitoring point on the i-th interface under layout scheme X; q ij The calculation formula of (X) is: q ij (X)=v ij (X)×(c 2ij (X)-c 1ij (X)) Among them, v ij (X) represents the normal airflow velocity at the jth monitoring point on the i-th interface under layout scheme X, c 2ij (X) and c 1ij (X) represents the pollutant concentrations in the second area and the first area at the corresponding monitoring points, respectively.

Images