A machine room air supply control method and system based on multiple parameters

Abstract

The present invention discloses a machine room air supply control method and system based on multiple parameters, which relates to the field of air conditioning energy saving technology. Based on temperature index data, energy consumption index data and PUE index data, the present invention uses a reinforcement learning algorithm to process the control parameters of the machine room, obtains a corresponding control strategy, and can realize the control of each cabinet, and perform targeted temperature control, so as to achieve the purpose of cooling the equipment first and then cooling the space, effectively improve the availability of refrigeration capacity, and reduce the working time and power consumption of the air conditioner, extend the working life of the air conditioner, and reduce resource waste; the system structure of the present invention is simple, only temperature index data, energy consumption index data and PUE index data and control parameters are required, and the reinforcement learning algorithm can be used to solve the problem of poor effect on thermal relay and high temperature of some cabinets, reduce energy waste, and further ensure the working efficiency of the cabinets in each working area.

Description

Technical Field

[0001] The present invention relates to the technical field of air conditioning energy saving, and in particular to a method and system for controlling air supply in a machine room based on multiple parameters. Background Art

[0002] The existing precise air supply system is a way to accurately deliver the cold air of the air conditioner to the designated area through the air duct. Precise air supply is currently a good way to solve the environmental problems of computer rooms with dense equipment, many servers, high heat generation, and air supply dead zones (it is difficult to add air conditioners). It is effective in solving the "heat island" problem in computer rooms, but it is less effective in solving the problems of heat relay and high temperatures in some cabinets.

[0003] Most of the old-fashioned data rooms have the "heat island" phenomenon with high temperatures in some areas due to the air supply mode (direct blowing) and installation location of the air conditioner in the room. The cabinets are arranged in order, and the air inlet of the cabinets in the back row is directly opposite to the heat exhaust vents of the cabinets in the front row. While inhaling the cooling cold air, it also inhaled the hot air discharged by the cabinets in the front row. Each row accumulates to form a heat relay. The traditional room cooling adopts the cooling mode of "cooling the space first, then cooling the equipment". It is impossible to carry out targeted cooling for the high-temperature cabinets that are distributed at multiple points and constantly changing (affected by the uncertain flow of equipment service users, the heat generated by the cabinets in a unit time is not uniform, and it fluctuates). In order to take care of the cooling needs of some local and a small number of high-temperature cabinets, the air conditioner setting temperature has to be greatly lowered to achieve the purpose by greatly reducing the overall space temperature of the room, but it will lead to an increase in the cooling capacity of the air conditioner, a longer working time, an increase in power consumption, and a shortened life. Compared with equipment in non-high-temperature areas, communication equipment that is often in high-temperature areas has a higher failure rate and is more prone to aging. Summary of the invention

[0004] The object of the present invention is to provide a machine room air supply control method and system based on multiple parameters to improve the above technical problems.

[0005] In order to achieve the above-mentioned object of the invention, the embodiment of the present invention provides the following technical solutions:

[0006] A method for controlling air supply in a computer room based on multiple parameters, comprising:

[0007] S1, collect multi-parameter data of each cabinet in a certain computer room;

[0008] S2. Calculate the temperature index data, energy consumption index data and PUE index data corresponding to each cabinet based on the multi-parameter data;

[0009] S3, obtaining the upper and lower air supply control parameters of the automatic compensation precision air supply system corresponding to the computer room;

[0010] S4. Based on the temperature index data, energy consumption index data and PUE index data, the upper and lower air supply control parameters are processed using the reinforcement learning algorithm to obtain the corresponding control strategy;

[0011] S5. Based on the control strategy, update the upper and lower air supply control parameters.

[0012] Furthermore, the multi-parameter data includes the air conditioning energy consumption data of the computer room, the power consumption of IT equipment, the energy consumption of the digital power distribution system, other energy consumption data and the upper and lower temperature data of each cabinet;

[0013] The up and down air supply control parameters include the rotation speed and damper opening corresponding to the up and down air supply fans corresponding to each cabinet.

[0014] Furthermore, the temperature index data, energy consumption index data and PUE index data corresponding to each cabinet are calculated based on the multi-parameter data, including:

[0015] S2-1, calculating the corresponding reference temperature based on the upper and lower temperature data of each of the cabinets;

[0016] S2-2, calculating the correlation of the reference temperatures corresponding to the cabinets;

[0017] S2-3. Based on the correlation, calculate the temperature index corresponding to each of the cabinets; wherein the temperature index data includes the temperature index corresponding to each of the cabinets;

[0018] S2-4, calculating the energy consumption index data based on the air conditioning energy consumption data;

[0019] S2-5. Calculate the PUE index data based on the air conditioning energy consumption data, the IT equipment power consumption, the digital power distribution system energy consumption, and the other energy consumption data.

[0020] Furthermore, based on the correlation, the temperature index corresponding to each of the cabinets is calculated, and the corresponding formula is:

[0021]

[0022] in, represents the temperature index corresponding to the i-th cabinet, Ti and Ti,j represent the reference temperature corresponding to the i-th cabinet and the reference temperature corresponding to the j-th cabinet adjacent to the i-th cabinet, respectively, Ni represents the total number of adjacent cabinets of the i-th cabinet, and ri,j represents the correlation between the i-th cabinet and its adjacent j-th cabinet.

[0023] Furthermore, the reinforcement learning algorithm adopts the PPO algorithm;

[0024] Based on the temperature index data, energy consumption index data and PUE index data, the upper and lower air supply control parameters are processed by using a reinforcement learning algorithm to obtain corresponding control strategies, including:

[0025] S4-1, using the temperature index data, the energy consumption index data and the PUE index data as a state space S;

[0026] S4-2, using the up and down air supply control parameters as the action space;

[0027] S4-3, initialization strategy;

[0028] S4-4. Select the current state based on the current strategy and obtain the corresponding probability distribution;

[0029] S4-5. Based on the probability distribution, select the current action using the ε-greedy strategy;

[0030] S4-6, execute the current action and obtain the corresponding reward and next state;

[0031] S4-7. Based on the reward, obtain the corresponding state value and expected return;

[0032] S4-8. Calculate a corresponding advantage function based on the expected return and the state value;

[0033] S4-9, calculate the probability ratio;

[0034] S4-10, calculating an optimization objective function based on the probability ratio and the advantage function;

[0035] S4-11. Based on the optimization objective function, update the policy parameters of the current policy to obtain an updated policy;

[0036] S4-12, repeat S4-4 to S4-11 until the preset number of iterations is reached, and the final strategy, that is, the control strategy, is obtained.

[0037] Furthermore, the formula of the reward Rt is:

[0038]

[0039] Among them, ω1, ω2, and ω3 represent weight coefficients, TPUE,t and CPUE,t represent the target PUE value and the actual PUE value corresponding to the current time step t, respectively, TTem,t and CTem,t represent the target temperature index and the actual temperature index corresponding to the current time step t, respectively, and Tec,t and Cec,t represent the target energy consumption index and the actual energy consumption index corresponding to the current time step t, respectively.

[0040] Furthermore, the formula corresponding to the probability ratio is:

[0041]

[0042] Among them, rt represents the probability ratio, Indicates the current strategy The probability of executing the current action at in the current state st, The strategy for the previous state The probability of executing the current action at in the current state st.

[0043] Furthermore, the formula of the optimization objective function L(θt) is:

[0044]

[0045] Where rt represents the probability ratio, min(·) represents the minimum function, clip(·) represents the clip algorithm, ε represents a constant, st represents the current state, at represents the current action, and A(st,at) represents the advantage function corresponding to the current state st executing the current action at. represents the expectation at the current time step t.

[0046] A machine room air supply control system based on multiple parameters includes a machine room air supply data acquisition module, an index calculation module and a machine room air supply control module; wherein:

[0047] The computer room air supply data acquisition module is used to collect multi-parameter data of each cabinet in a computer room in real time and automatically compensate the upper and lower air supply control parameters of the precision air supply system;

[0048] Index calculation module, used to calculate the temperature index data, energy consumption index data and PUE index data corresponding to each cabinet;

[0049] The computer room air supply control module is used to process the upper and lower air supply control parameters based on the temperature index data, energy consumption index data and PUE index data using the reinforcement learning algorithm to obtain the corresponding control strategy.

[0050] The beneficial effects of the present invention are:

[0051] Based on temperature index data, energy consumption index data and PUE index data, this method uses reinforcement learning algorithm to process the control parameters of the computer room and obtains the corresponding control strategy. It can control each cabinet and perform targeted temperature control, so as to achieve the purpose of cooling the equipment first and then the space, effectively improve the availability of cooling capacity, reduce the working time and power consumption of the air conditioner, extend the working life of the air conditioner, and reduce resource waste.

[0052] The system has a simple structure and only requires temperature index data, energy consumption index data, PUE index data and control parameters. It can use the reinforcement learning algorithm to solve the problems of poor effect on thermal relay and high temperature of some cabinets, reduce energy waste, and further ensure the working efficiency of cabinets in each working area. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0054] Figure 1 A flow chart of a method in an embodiment of the present invention;

[0055] Figure 2 This is a structural diagram of the automatic compensation precise air supply system in an embodiment of the present invention;

[0056] Figure 3 1 is a system structure diagram in an embodiment of the present invention. DETAILED DESCRIPTION

[0057] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present invention.

[0058] See also Figure 1 The present embodiment provides a method for controlling air supply in a computer room based on multiple parameters, which is characterized by comprising:

[0059] S1, collect multi-parameter data of each cabinet in a certain computer room;

[0060] The multi-parameter data includes the air conditioning energy consumption data of the computer room, the power consumption of IT equipment, the energy consumption of the digital power distribution system, other energy consumption data and the upper and lower temperature data of each cabinet. The upper and lower temperature data are collected through the temperature sensor of the corresponding cabinet; the air conditioning energy consumption data, IT equipment power consumption, digital power distribution system energy consumption, and other energy consumption data of the computer room are obtained through power measurement equipment, energy management software, etc.

[0061] S2. Based on the multi-parameter data, calculate the temperature index data, energy consumption index data and PUE index data corresponding to each cabinet.

[0062] The temperature index data, energy consumption index data and PUE index data corresponding to each cabinet are calculated based on the multi-parameter data, including:

[0063] S2-1. Based on the upper and lower temperature data of each cabinet, calculate the corresponding reference temperature; the reference temperature data is the average of each upper and lower temperature data, that is, T = (Tup + Tdown) 2. T represents the reference temperature, Tup represents the upper temperature data, and Tdown represents the lower temperature data.

[0064] S2-2, calculating the correlation of the reference temperatures corresponding to the cabinets, wherein the correlation is a Pearson correlation coefficient or other correlation coefficients.

[0065] S2-3. Based on the correlation, calculate the temperature index corresponding to each of the cabinets; wherein the temperature index data includes the temperature index corresponding to each of the cabinets.

[0066] Based on the correlation, the temperature index corresponding to each of the cabinets is calculated, and the corresponding formula is:

[0067]

[0068] in, represents the temperature index corresponding to the i-th cabinet, Ti and Ti,j represent the reference temperature corresponding to the i-th cabinet and the reference temperature corresponding to the j-th cabinet adjacent to the i-th cabinet, respectively, Ni represents the total number of adjacent cabinets of the i-th cabinet, and ri,j represents the correlation between the i-th cabinet and its adjacent j-th cabinet.

[0069] S2-4. Calculate the energy consumption index data based on the air-conditioning energy consumption data. Calculating the energy consumption index data is a prior art and will not be described in detail.

[0070] S2-5. Calculate the PUE index data based on the air conditioning energy consumption data, the IT equipment power consumption, the digital power distribution system energy consumption, and the other energy consumption data. PUE is the power utilization efficiency of the computer room, and its value is the ratio of the total power consumption of the data center to the power consumption of IT equipment. The total power consumption of the data center includes the air conditioning energy consumption data, the digital power distribution system energy consumption, and other energy consumption data.

[0071] S3. Obtain the up and down air supply control parameters of the automatic compensation precision air supply system corresponding to the computer room; the up and down air supply control parameters include the rotation speed and damper opening corresponding to the up and down air supply fans corresponding to each cabinet.

[0072] like Figure 2 As shown, the automatic compensation precise air supply system is a cabinet-type variable automatic compensation precise air supply system, which includes an air supply subsystem, a collection and monitoring module, an automatic compensation air supply module and a control air supply fan module; wherein the air supply subsystem is divided into an upper air supply module and a lower air supply module according to the air supply mode of the original computer room air conditioner; the collection and monitoring module is respectively connected to the upper air supply module, the lower air supply module, the automatic compensation air supply module and the control air supply fan module. The upper air supply module and the lower air supply module include an upper air supply fan and a lower air supply fan, respectively.

[0073] S4. Based on the temperature index data, energy consumption index data and PUE index data, the upper and lower air supply control parameters are processed using the reinforcement learning algorithm to obtain the corresponding control strategy;

[0074] The reinforcement learning algorithm adopts the PPO algorithm; the PPO algorithm has the advantages of easy implementation, high efficiency, stability and balanced exploration. It can ensure the security of strategy updates without complex calculations, reduce the calculation speed, and provide timely feedback on strategies. It approximates the KL divergence constraint through the clipping mechanism, has lower computational complexity, is easy to implement and parallelize, and performs well in many practical tasks. Therefore, the PPO algorithm is selected to process the upper and lower air supply control parameters to obtain the corresponding control strategy.

[0075] Based on the temperature index data, energy consumption index data and PUE index data, the upper and lower air supply control parameters are processed by using a reinforcement learning algorithm to obtain corresponding control strategies, including:

[0076] S4-1, the temperature index data, the energy consumption index data and the PUE index data are used as the state space S; wherein S = [s0, s1, ..., st, ..., sm], m represents the total number of states, s0, s1, ..., st, ..., sm represent the states, represents the temperature index data under the transition state st, Pa,t represents the energy consumption index data under the transition state st, and PPUE,t represents the PUE index data under the transition state st.

[0077] S4-2. Use the upper and lower air supply control parameters as the action space a=[a0,a1,…,at,…,an]; at=[vup,t,vdown,t,wup,t,wdown,t]; wherein n represents the total number of actions, a=[a0,a1,…,at,…,an] represents the action, vup,t represents the corresponding rotation speed of the upper fan under the action at, vdown,t represents the corresponding rotation speed of the lower air supply fan under the action at, wup,t represents the corresponding damper opening of the upper air supply fan under the action at, and wdown,t represents the corresponding damper opening of the lower air supply fan under the action at.

[0078] S4-3. Initialization strategy

[0079] S4-4. Based on the current strategy Select the current state st and get the corresponding probability distribution.

[0080] S4-5. Based on the probability distribution, the current action at is selected using the ε-greedy strategy.

[0081] S4-6. Execute the current action at and obtain the corresponding reward Rt and the next state st+1.

[0082] When the action is executed, if the temperature index drops, positive feedback is obtained, otherwise negative feedback is obtained; if the energy consumption index data drops, positive feedback is obtained, otherwise negative feedback is obtained; if the PUE index data drops, positive feedback is obtained, otherwise negative feedback is obtained. Furthermore, the formula for the reward Rt is:

[0083]

[0084] Among them, ω1, ω2, and ω3 represent weight coefficients, TPUE,t and CPUE,t represent the target PUE value and the actual PUE value corresponding to the current state t, respectively, TTem,t and CTem,t represent the target temperature index and the actual temperature index corresponding to the current state t, respectively, Tec,t and Cec,t represent the target energy consumption index and the actual energy consumption index corresponding to the current state t, respectively.

[0085] S4-7. Based on the reward Rt, obtain the corresponding state value V(st) and expected return Q(st,at).

[0086] The formula of the state value V(st) is:

[0087]

[0088] in, Indicates the current strategy The expected function under , γ represents the discount factor, and V(st+1) represents the state value of state st+1.

[0089] The formula for the expected return Q(st,at) is:

[0090]

[0091] Among them, at+1 represents the action, st+1 represents the state, Q(st+1,at+1) represents the expected return of executing action at+1 in state st+1, and Rt+1 represents the reward obtained by executing action at+1 in state st+1. Represents the expected function.

[0092] S4-8. Based on the expected return Q(st,at) and the state value V(st), calculate the corresponding advantage function A(st,at).

[0093] The formula of the advantage function A(st,at) is:

[0094] A(st,at)=Q(st,at)-V(st).

[0095] S4-9. Calculate the probability ratio.

[0096] The formula corresponding to the probability ratio is:

[0097]

[0098] Among them, rt represents the probability ratio, Indicates the current strategy The probability of executing the current action at in the current state st, The strategy for the previous state The probability of executing the current action at in the current state st.

[0099] S4-10. Calculate an optimization objective function based on the probability ratio and the advantage function At(st,at).

[0100] The formula of the optimization objective function L(θt) is:

[0101]

[0102] Where rt represents the probability ratio, min(·) represents the minimum function, clip(·) represents the clip function, ε represents a constant, st represents the current state, at represents the current action, and A(st,at) represents the advantage function corresponding to the current state st executing the current action at. Represents the expectation of the current time step t. The clip function limits the range of variation of the probability ratio rt to prevent the update step from being too large.

[0103] S4-11, based on the optimization objective function, update the current strategy The policy parameter θt is used to obtain the updated policy

[0104] S4-12, repeat S4-4 to S4-11 until the preset number of iterations is reached, and the final strategy, that is, the control strategy, is obtained.

[0105] S5. Based on the control strategy, the gradient ascent method is used to update the upper and lower air supply control parameters.

[0106] Based on temperature index data, energy consumption index data and PUE index data, this method uses reinforcement learning algorithm to process the control parameters of the computer room and obtains the corresponding control strategy. It can control each cabinet and perform targeted temperature control, so as to achieve the purpose of cooling the equipment first and then the space, effectively improve the availability of cooling capacity, reduce the working time and power consumption of the air conditioner, extend the working life of the air conditioner, and reduce resource waste.

[0107] like Figure 3 As shown, a multi-parameter-based machine room air supply control system includes a machine room air supply data acquisition module, an index calculation module and a machine room air supply control module; wherein:

[0108] The computer room air supply data acquisition module is used to collect multi-parameter data of each cabinet in a computer room in real time and automatically compensate the upper and lower air supply control parameters of the precision air supply system;

[0109] Index calculation module, used to calculate the temperature index data, energy consumption index data and PUE index data corresponding to each cabinet;

[0110] The computer room air supply control module is used to process the upper and lower air supply control parameters based on the temperature index data, energy consumption index data and PUE index data using the reinforcement learning algorithm to obtain the corresponding control strategy.

[0111] The system has a simple structure and only requires temperature index data, energy consumption index data, PUE index data and control parameters. It can use the reinforcement learning algorithm to solve the problems of poor effect on thermal relay and high temperature of some cabinets, reduce energy waste, and further ensure the working efficiency of cabinets in each working area.

[0112] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A multi-parameter based air supply control method for a computer room, characterized in that: include: S1, collect multi-parameter data of each cabinet in a certain computer room; S2. Calculate the temperature index data, energy consumption index data and PUE index data corresponding to each cabinet based on the multi-parameter data; S3, obtaining the upper and lower air supply control parameters of the automatic compensation precision air supply system corresponding to the computer room; S4. Based on the temperature index data, energy consumption index data and PUE index data, the upper and lower air supply control parameters are processed using the reinforcement learning algorithm to obtain the corresponding control strategy; S5. Based on the control strategy, update the upper and lower air supply control parameters.

2. The method for controlling air supply in a computer room based on multiple parameters according to claim 1, characterized in that: The multi-parameter data includes the air conditioning energy consumption data of the computer room, the power consumption of IT equipment, the energy consumption of the digital power distribution system, other energy consumption data and the upper and lower temperature data of each cabinet; The up and down air supply control parameters include the rotation speed and damper opening corresponding to the up and down air supply fans corresponding to each cabinet.

3. The method for controlling air supply in a computer room based on multiple parameters according to claim 2, characterized in that: The temperature index data, energy consumption index data and PUE index data corresponding to each cabinet are calculated based on the multi-parameter data, including: S2-1, calculating the corresponding reference temperature based on the upper and lower temperature data of each of the cabinets; S2-2, calculating the correlation of the reference temperatures corresponding to the cabinets; S2-3. Based on the correlation, calculate the temperature index corresponding to each of the cabinets; wherein the temperature index data includes the temperature index corresponding to each of the cabinets; S2-4, calculating the energy consumption index data based on the air conditioning energy consumption data; S2-5. Calculate the PUE index data based on the air conditioning energy consumption data, the IT equipment power consumption, the digital power distribution system energy consumption, and the other energy consumption data.

4. The method for controlling air supply in a computer room based on multiple parameters according to claim 3, characterized in that: Based on the correlation, the temperature index corresponding to each of the cabinets is calculated, and the corresponding formula is: in, represents the temperature index corresponding to the i-th cabinet, Ti and Ti,j represent the reference temperature corresponding to the i-th cabinet and the reference temperature corresponding to the j-th cabinet adjacent to the i-th cabinet, respectively, Ni represents the total number of adjacent cabinets of the i-th cabinet, and ri,j represents the correlation between the i-th cabinet and its j-th adjacent cabinet.

5. The method for controlling air supply in a computer room based on multiple parameters according to claim 3, characterized in that: The reinforcement learning algorithm adopts the PPO algorithm; Based on the temperature index data, energy consumption index data and PUE index data, the upper and lower air supply control parameters are processed by using a reinforcement learning algorithm to obtain corresponding control strategies, including: S4-1, using the temperature index data, the energy consumption index data and the PUE index data as a state space S; S4-2, using the up and down air supply control parameters as the action space; S4-3, initialization strategy; S4-4. Select the current state based on the current strategy and obtain the corresponding probability distribution; S4-5. Based on the probability distribution, select the current action using the ε-greedy strategy; S4-6, execute the current action and obtain the corresponding reward and next state; S4-7. Based on the reward, obtain the corresponding state value and expected return; S4-8. Calculate a corresponding advantage function based on the expected return and the state value; S4-9, calculate the probability ratio; S4-10, calculating an optimization objective function based on the probability ratio and the advantage function; S4-11. Based on the optimization objective function, update the policy parameters of the current policy to obtain an updated policy; S4-12, repeat S4-4 to S4-11 until the preset number of iterations is reached, and the final strategy, that is, the control strategy, is obtained.

6. The method for controlling air supply in a computer room based on multiple parameters according to claim 5, characterized in that: The formula for the reward Rt is: Among them, ω1, ω2, and ω3 represent weight coefficients, TPUE,t and CPUE,t represent the target PUE value and the actual PUE value corresponding to the current time step t, respectively, TTem,t and CTem,t represent the target temperature index and the actual temperature index corresponding to the current time step t, respectively, and Tec,t and Cec,t represent the target energy consumption index and the actual energy consumption index corresponding to the current time step t, respectively.

7. The method for controlling air supply in a computer room based on multiple parameters according to claim 5, characterized in that: The formula corresponding to the probability ratio is: Among them, rt represents the probability ratio, Indicates the current strategy The probability of executing the current action at in the current state st, The strategy for the previous state The probability of executing the current action at in the current state st.

8. The method for controlling air supply in a computer room based on multiple parameters according to claim 5, characterized in that: The formula of the optimization objective function L(θt) is: Where rt represents the probability ratio, min(·) represents the minimum function, clip(·) represents the clip algorithm, ε represents a constant, st represents the current state, at represents the current action, and A(st,at) represents the advantage function corresponding to the current state st executing the current action at. represents the expectation at the current time step t.

9. A machine room air supply control system based on multiple parameters, used to implement a machine room air supply control method based on multiple parameters as claimed in any one of claims 1 to 8, characterized in that: It includes a computer room air supply data acquisition module, an index calculation module and a computer room air supply control module; wherein: The computer room air supply data acquisition module is used to collect multi-parameter data of each cabinet in a computer room in real time and automatically compensate the upper and lower air supply control parameters of the precision air supply system; Index calculation module, used to calculate the temperature index data, energy consumption index data and PUE index data corresponding to each cabinet; The computer room air supply control module is used to process the upper and lower air supply control parameters based on the temperature index data, energy consumption index data and PUE index data using the reinforcement learning algorithm to obtain the corresponding control strategy.

Images