A control method and system of a machine room temperature control execution device

Abstract

The application relates to the field of data machine room temperature control, and more particularly to a control method and system of a machine room temperature control execution device. The method is applied to a data machine room, which is a data center or a machine room. The method comprises the following steps: obtaining historical operation data and current operation data in the data machine room, and constructing a time delay response graph based on the historical operation data; obtaining a current state of the data machine room at a current moment according to the current operation data, the time delay response graph and the historical operation data; generating a plurality of candidate control action sequences according to the current state and the time delay response graph; obtaining a target control action sequence from the candidate control action sequences according to the current state, the candidate control action sequences and a value network; and controlling the temperature control execution device to execute according to the target control action sequence, thereby completing the control of the temperature control execution device. The method is used for matching the thermal inertia of the data machine room when controlling the temperature control execution device, i.e. the air conditioner, so that an attributable air conditioner energy-saving temperature control decision can be realized.

Description

Technical Field

[0001] This invention relates to the field of data center temperature control, and more specifically, to a control method and system for a data center temperature control execution device. Background Technology

[0002] In small and medium-sized data center scenarios, multiple temperature control devices such as precision air conditioners, cabinet air conditioners, or split air conditioners are typically configured. The data center's operating environment is maintained through a limited number of temperature measuring points, power data from the temperature control devices, and power environment monitoring equipment. Unlike large data centers, small and medium-sized data centers lack high-precision thermal simulation models that can be directly used for real-time control, and they also lack precise control over airflow and thermal field distribution. On-site control still relies mainly on fixed thresholds, manual patrols, and experience-based start-stop systems. Existing technologies have attempted to introduce intelligent decision-making methods into the field of data center temperature control, but they still face inherent limitations: the thermal process in data centers has significant inertia, and the intensity, onset time, and peak time of the action of the same temperature control actuator are not consistent at different temperature measurement points. The true effect of a single control action often spans multiple sampling periods, while existing solutions still use the decision of a single control as the basic unit, which is difficult to match this physical characteristic. At the same time, non-control factors such as changes in data center load and fluctuations in the external environment can easily overlap with the effects of the temperature control actuator's actions, making it difficult to evaluate the true effect of a single control action. In addition, relying solely on thresholds or penalty mechanisms is insufficient to effectively constrain temperature risks while optimizing energy consumption. Summary of the Invention

[0003] This invention provides a control method and system for a data center temperature control execution device, which is used to match the thermal inertia of the data center when controlling the temperature control execution device, and to achieve attributable energy-saving temperature control decisions.

[0004] According to a first aspect of this application, a control method for a temperature control actuator in a data center is provided. The method is applied to a data center, which includes several devices, each including a load, a temperature control actuator, and temperature monitoring points. The method includes: Acquire historical operating data from the data center, and construct a time-delay response diagram based on the historical operating data. The time-delay response diagram includes data on the temperature impact of the temperature control execution device on the temperature monitoring point. Obtain the current operating data of the data center; Based on the current operating data, the time-delay response diagram, and the historical operating data, obtain the current status of the data center at the current moment; Based on the current state and the time-delay response diagram, several candidate control action sequences are generated; Based on the current state, the candidate control action sequence, and the preset value network, obtain the target control action sequence from the candidate control action sequence; The temperature control device is controlled to perform according to the target control action sequence, thereby completing the control of the temperature control device.

[0005] Understandably, by constructing a time-delay response map of temperature monitoring points for temperature control actuators in the data center, the current state of deep reinforcement learning and the generation of action sequences are structurally coupled around the thermal response patterns in the time-delay response map, achieving precise matching of the thermal inertia of the data center. Compared to existing methods where single-step actions and rewards cannot cover the complete thermal response process, this application uses action sequences as decision units and can design multi-step control strategies based on the delay in the impact of each temperature control actuator on different temperature measurement points, improving the safety and energy-saving effect of data center temperature control.

[0006] Optionally, constructing the time-delay response diagram based on the historical operational data includes: Based on the historical operating data, obtain the state change events of the temperature control actuator, and obtain the occurrence time of the state change events accordingly; In the state change event, if the state change event meets the preset state conditions, then the state change event is taken as the anchor event, and the occurrence time of the state change event is taken as the occurrence time of the anchor event; wherein the preset state conditions are that the equipment state of the data center meets the preset state during a preset first time period before the occurrence time of the state change event, and the equipment state of the data center meets the preset state during a preset second time period after the occurrence time of the state change event. A baseline window is obtained based on a preset first time period before the occurrence of the anchor event, and a response window is obtained based on a preset second time period after the occurrence of the anchor event. Based on a preset time interval, the first actual temperature of the temperature monitoring point in the baseline window is obtained from the historical operating data, and the baseline temperature rise rate of the temperature monitoring point is obtained based on the first actual temperature. Based on the time interval, obtain the heat data generated by the data center load, the outdoor temperature, and the second actual temperature of the temperature monitoring point in the response window from the historical operating data. Based on the baseline temperature rise rate, the heat data generated by the load, the outdoor temperature, the second actual temperature, and the preset exogenous disturbance model, a directed edge is established between the temperature control execution device and the temperature monitoring point. A time-delay response graph is constructed based on the temperature control execution device, the temperature monitoring point, and the directed edge.

[0007] Understandably, by setting a baseline window and a response window for anchor point events, and by using an exogenous disturbance model to eliminate the impact of natural load variations and outdoor environmental fluctuations from actual temperature changes, the independent contribution of each temperature control actuator to each temperature measurement point can be accurately identified. This avoids the reward bias caused by misattributing exogenous disturbances in traditional methods. The constructed time-delay response graph records the time-delay response relationship of the temperature control actuator to each temperature monitoring point in the form of directed edges, providing an accurate thermal response knowledge base for subsequent action sequence compilation, safety verification, and value estimation.

[0008] Optionally, establishing a directed edge between the temperature control execution device and the temperature monitoring point based on the baseline temperature rise rate, the heat data generated by the load, the outdoor temperature, the second actual temperature, and a preset exogenous disturbance model includes: The baseline temperature rise rate, the heat data generated by the load, and the outdoor temperature are input into the exogenous disturbance model for processing to obtain the predicted temperature of the temperature monitoring point when there is no temperature control execution device in the response window. The difference between the second actual temperature and the predicted temperature is used as the total net temperature effect of the temperature control execution device on the temperature monitoring point in the anchor point event. Based on all anchor point events of the temperature actuator on the temperature monitoring point, the total net temperature effect of the temperature actuator on the temperature monitoring point and the corresponding confidence interval are calculated. Based on the total net temperature effect and the confidence interval, obtain the temperature impact data of the temperature control device on the temperature monitoring point; Based on the total net temperature effect and the confidence interval, a directed edge is established from the temperature control execution device to the temperature monitoring point, and the corresponding temperature influence data is used as the attribute of the directed edge.

[0009] Understandably, by predicting the baseline temperature without the influence of temperature-controlled actuators using an exogenous disturbance model, the net total temperature effect of the temperature-controlled actuators is accurately extracted from actual temperature changes. The statistical accumulation of multiple anchor events is then used to estimate the temperature impact data, which is then used as the attribute of the directed edges. Valid directed edges are established based on the net total temperature effect and the corresponding confidence intervals, ensuring the confidence and reliability of each edge in the time-delay response graph. This provides a reliable graph constraint basis for the compilation and verification of subsequent action sequences.

[0010] Optionally, establishing a directed edge from the temperature control actuator to the temperature monitoring point based on the total net temperature effect and the confidence interval includes: Determine whether the total net temperature effect is greater than or equal to a preset total threshold, and whether the lower confidence level is greater than or equal to a preset confidence level. If so, establish a directed edge from the temperature control execution device to the temperature monitoring point; otherwise, do not establish a directed edge from the temperature control execution device to the temperature monitoring point. The lower confidence level is obtained based on the confidence interval.

[0011] Understandably, by simultaneously requiring both the peak value of the temperature effect and the lower boundary of the confidence level to meet the preset thresholds, that is, only valid edges with significant cooling effects and statistical reliability are used to build directed edges in the time-delay response graph, it is ensured that each edge in the time-delay response graph has a clear physical meaning and statistical credibility. This avoids false edges from entering the action compilation stage due to insufficient sample size or random noise, thereby improving the accuracy of subsequent security verification and evaluation.

[0012] Optionally, obtaining the current state of the data center at the current moment based on the current operating data, the time-delay response diagram, and the historical operating data includes: Based on the current operating data, obtain the heat data generated by the load in the data center at the current moment, the current temperature data of several temperature monitoring points, the current operating data of several temperature control devices, and the current environmental disturbance data. The temperature impact data of the temperature control execution device on the temperature monitoring point is obtained based on the time delay response diagram. Based on the historical operation data, obtain the historical execution data of the previous target control action sequence; The current state of the data center is obtained by fusing one or more of the following: the heat data generated by the load at the current moment, the temperature data of the temperature monitoring point at the current moment, the operating data of the temperature control execution device at the current moment, the environmental disturbance data at the current moment, the temperature influence data, and the historical execution data of the previous target control action sequence.

[0013] Understandably, by integrating the current temperature data of the temperature monitoring points, the current operating data of the temperature control actuators, the current heat data generated by the load, the current environmental disturbance data, the temperature impact data, and the historical execution data of the previous target control action sequence into the current state of the data center, the value network can perceive the specific time-delay response relationship of each temperature control actuator to each temperature monitoring point when making decisions. Compared with the ordinary state that only contains sensor readings, the current state obtained by this application can guide the value network to fully consider thermal inertia and time-delay characteristics when evaluating actions, thereby making decisions that match the physical process.

[0014] Optionally, the step of generating a plurality of candidate control action sequences based on the current state and the time-delay response diagram includes: The target temperature monitoring point is determined based on the current state; wherein the first difference between the actual temperature of the target temperature monitoring point and the preset temperature threshold is less than or equal to the preset difference threshold. Based on the current state of the data center, acquire the temperature impact data of each temperature control execution device on the target temperature monitoring point; Based on the first difference and the temperature impact data of each of the temperature control actuators on the target temperature monitoring point, the target temperature control actuator is determined from the temperature control actuators; The target temperature control actuators are sorted based on temperature influence data to obtain a first order; based on the first order, several candidate control action sequences with different control strategies are generated, each candidate control action sequence including at least two consecutive control steps, and each control step including a control operation on the target temperature control actuator and a corresponding preset execution time.

[0015] Understandably, based on the temperature impact data of each temperature control actuator on the target temperature monitoring point in the time-delay response diagram, a candidate control action sequence containing multiple consecutive control steps is generated, realizing dynamic generation of the action space rather than fixed action settings. Because the generation of the candidate control action sequence is constrained by the temperature impact data and guided by the first order, the candidate control action sequence not only ensures the cooling requirement of the target temperature monitoring point but also eliminates a large number of physically invalid combinations. This allows the value network to evaluate only the refined candidate control action sequence, significantly reducing the search complexity of the action space and improving decision-making efficiency and interpretability.

[0016] Optionally, obtaining the target control action sequence from the candidate control action sequence based on the current state, the candidate control action sequence, and a preset value network includes: Based on the candidate control action sequence, obtain one or more safety control action sequences that satisfy preset safety constraints; The current state and the security control action sequence are input into the value network, and value estimation is performed on the security control action sequence to obtain the value estimate corresponding to the security control action sequence. The sequence of safety control actions with the highest estimated value is selected as the target control action sequence.

[0017] Understandably, before a candidate control action sequence enters the value network, it is first checked whether it meets the preset safety constraints, and candidate control action sequences that do not meet the safety constraints are eliminated. Only candidate control action sequences that meet the safety constraints can enter the value network to participate in value ranking. This ensures that high energy consumption benefits cannot be obtained by sacrificing safety, thereby guaranteeing the safety of exploration actions and overcoming the limitations of traditional methods that rely solely on reward penalties to constrain temperature risks.

[0018] Optionally, controlling the temperature control actuator to perform according to the target control action sequence includes: The temperature control execution device is controlled sequentially according to the control steps of the target control action sequence. After each control step is completed, it is checked whether the actual temperature of the temperature monitoring point meets the preset first condition and whether the temperature control execution device meets the preset second condition. If so, the currently executed control step is updated to the preset safety prefix data. Otherwise, cancel the unexecuted control steps in the target control action sequence and restore the control steps corresponding to the security prefix data, and re-execute the step of obtaining the target control action sequence from the candidate control action sequence based on the current state, the candidate control action sequence and the preset value network.

[0019] Understandably, by controlling the temperature control device sequentially according to the target control action sequence, without re-decision at each control step, the frequent policy changes and start-stop oscillations caused by single-step decision-making in traditional reinforcement learning are avoided. This effectively reduces mechanical wear and energy waste caused by frequent switching of the temperature control device. Simultaneously, through a safety prefix saving mechanism and anomaly recovery mechanism, in the event of prediction deviation or equipment malfunction, it can quickly revert to a verified safe control state and reacquire the safe control action sequence, achieving both safety and continuity of the control process under thermal inertia conditions.

[0020] Optionally, after the step of controlling the temperature control actuator according to the target control action sequence to complete the control of the temperature control actuator, the method further includes: Record the actual execution data performed by the temperature control device; Obtain the counterfactual control action sequence of the target control action sequence; Based on the counterfactual control action sequence, counterfactual state reasoning is performed on the data center to obtain counterfactual execution data; Obtain counterfactual attribution benefits based on the actual execution data and the counterfactual execution data; The network parameters of the value network are updated based on the counterfactual attribution benefits.

[0021] Understandably, by constructing a counterfactual control action sequence corresponding to the target control action sequence and comparing it with actual execution data under the same load and environmental conditions, the reward accurately reflects the incremental control effect generated by the target control action sequence, avoiding reward bias caused by misattributing natural load decreases or environmental changes to control actions. Based on counterfactual attribution rewards and value target updates, value estimation more accurately reflects the true technical effect of the target control action sequence, improving the convergence stability of the value network under thermal inertia conditions and the energy-saving effect of the control strategy.

[0022] According to a second aspect of this application, a control system for a temperature control execution device in a data center is provided. The system is applied in a data center, which includes several devices, each including a load, a temperature control execution device, and temperature monitoring points. The system includes: The time-delay response diagram acquisition module is used to acquire historical operating data in the data center and construct a time-delay response diagram based on the historical operating data. The time-delay response diagram includes temperature impact data of temperature control execution equipment on temperature monitoring points. The current operating data acquisition module is used to acquire the current operating data of the data center; The current status acquisition module is used to acquire the current status of the data center at the current moment based on the current operating data, the time-delay response diagram, and the historical operating data. The candidate control action sequence acquisition module is used to generate several candidate control action sequences based on the current state and the time-delay response diagram; The target control action sequence acquisition module is used to acquire the target control action sequence from the candidate control action sequence based on the current state, the candidate control action sequence, and a preset value network. The control module is used to control the temperature control actuator to perform according to the target control action sequence, thereby completing the control of the temperature control actuator.

[0023] According to a third aspect of this application, an electronic device is provided, comprising: Memory, used to store one or more computer programs; A processor, when the one or more computer programs are executed by the processor, implements the control method for a computer room temperature control device as described in the first aspect above.

[0024] According to a fourth aspect of this application, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions, the computer instructions being configured to cause a processor to execute and implement the control method of the computer room temperature control execution device described in the first aspect above.

[0025] Based on any of the above aspects, the control method, system, electronic device, and storage medium for a computer room temperature control execution device provided in this application embodiment can achieve the following technical effects: By constructing a time-delay response diagram of the temperature control execution device to the temperature monitoring point, and using an exogenous disturbance model to accurately eliminate the influence of natural load changes and outdoor environmental fluctuations from actual temperature changes, the acquisition of the current state pair, the target action sequence, and the value estimation are all structurally coupled around the time-delay thermal response law of the time-delay response diagram. This achieves control action sequence control that continuously executes multiple control steps from a single decision. Compared with traditional single-step decision-making methods, the action sequence of this application can match the complete response cycle of the data center thermal process from its onset to its peak and then to its stabilization. This effectively solves the problem that single-step actions and single-step rewards cannot cover the real thermal response process of a single control action, avoids decision lag and ineffective control caused by thermal inertia, and improves the stability and energy-saving effect of data center temperature control.

[0026] Before candidate control action sequences enter the value network, their complete execution trajectories undergo safety verification. Candidate control action sequences that do not meet safety constraints are eliminated, ensuring that the value network only ranks the safety control action sequences throughout the entire process. Simultaneously, during the execution of the target control action sequence, a safety prefix saving and recovery mechanism allows for rapid rollback to a verified safety control state and execution of recovery operations in the event of an anomaly. This ensures that high energy-saving benefits cannot be obtained at the expense of safety, overcoming the limitations of traditional methods where energy-saving benefits numerically offset temperature-limit-exceeding penalties due to relying solely on reward penalties to constrain temperature risks. This effectively improves the safety of the control process.

[0027] By constructing a counterfactual control action sequence of the target control action sequence and comparing it with actual execution data under the same load and environmental conditions, the reward accurately reflects the temperature impact of the target control action sequence, avoiding misattribution of rewards due to natural load decreases or changes in the outdoor environment. Simultaneously, value estimation is performed on the safety control action sequence, adapting to the differences in the impact of different temperature-controlled execution devices within the safety control action sequence, thus improving the convergence stability of the value network under thermal inertia conditions and the energy-saving effect of the control strategy. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1This is a flowchart illustrating a control method for a computer room temperature control execution device provided in this embodiment.

[0030] Figure 2 The flowchart for constructing the time-delay response diagram provided in this embodiment.

[0031] Figure 3 This is a flowchart for establishing directed edges in this embodiment.

[0032] Figure 4 This is a flowchart for obtaining the current state provided in this embodiment.

[0033] Figure 5 This is a flowchart illustrating the generation of candidate control action sequences provided in this embodiment.

[0034] Figure 6 This is a flowchart for obtaining the target control action sequence provided in this embodiment.

[0035] Figure 7 The flowchart for counterfactual reasoning provided in this embodiment.

[0036] Figure 8 This is a schematic diagram of the functional modules of a control system for a computer room temperature control execution device provided in this embodiment.

[0037] Figure 9 This is a schematic diagram of the structure of the electronic device provided in this embodiment. Detailed Implementation

[0038] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this application. To better illustrate the following embodiments, some components in the drawings may be omitted, enlarged, or reduced, and do not represent the actual dimensions of the product; it is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0039] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0040] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0041] Existing single-step decision-making schemes are difficult to match the significant inertia and time delay characteristics of thermal processes in digital data centers, making it impossible to accurately evaluate the true effect of control actions. At the same time, the effects of non-control factors such as load fluctuations and environmental changes are superimposed on the effects of control actions, further exacerbating reward attribution bias. Thresholds or penalty mechanisms alone cannot effectively constrain temperature risks while optimizing energy consumption, making it difficult to achieve an effective balance between safety, stability, and energy efficiency in data center temperature control.

[0042] This embodiment provides a technical solution that can solve the above problems. The specific implementation of this application will be described in detail below with reference to the accompanying drawings.

[0043] like Figure 1 As shown, this embodiment provides a control method for a temperature control actuator in a data center. The method is applied to a data center, which includes several devices, including a load, a temperature control actuator, and temperature monitoring points. The method can be further divided into the following steps: S100. Obtain historical operating data in the data center, and construct a time-delay response diagram based on the historical operating data. The time-delay response diagram includes temperature impact data of the temperature control execution device on the temperature monitoring point. It is understood that the data center can include a data center or a general data center. A data center is typically a dedicated facility that centrally houses servers, storage, and network equipment and provides an operating environment. It possesses comprehensive power supply, precision air conditioning, security monitoring, and redundant backup systems, and typically serves large-scale business systems, serving as the core infrastructure for computing power and data storage. A general data center is a small, independent space configured with a small number of servers or network devices. It is typically equipped with several precision air conditioners or cabinet air conditioners and maintains operation through limited temperature measurement points and environmental monitoring. Its thermal field management accuracy and control automation level are lower than those of large data centers. Specifically, this embodiment can be applied to the aforementioned data centers and general data centers, particularly those lacking high-precision thermal simulation models for direct real-time control and precise understanding of airflow and thermal field distribution. This allows data centers or general data centers to achieve safe and energy-efficient temperature control decisions without relying on precise physical modeling, and can be widely applied to various small and medium-sized general data center scenarios.

[0044] Understandably, the data center or ordinary computer room typically includes loads, temperature control devices, and stability monitoring points. The load refers to the information technology equipment operating in the data center or ordinary computer room, such as servers, switches, and storage devices. These devices generate heat during operation and are the primary source of heat in the data center or ordinary computer room, representing the core heat load that the temperature control devices need to handle. The temperature control devices are used to regulate the temperature of the data center or ordinary computer room, such as precision air conditioners, cabinet air conditioners, and split air conditioners. They actively control the temperature of the data center or ordinary computer room environment by starting / stopping, adjusting cooling capacity, or changing air supply parameters. Temperature monitoring points refer to temperature sensors or temperature measuring points deployed in different locations within the data center or ordinary computer room. They are used to collect temperature data from various areas of the data center or ordinary computer room in real time and are the fundamental data source for sensing the thermal state of the data center or ordinary computer room and evaluating the effectiveness of temperature control.

[0045] In this embodiment, the system first retrieves temperature data from the historical database of the data center at different historical times, status data of several temperature monitoring points in the data center at different historical times, load data at different historical times, and outdoor temperature at different historical times. The minimum sampling period for this data can be set according to the on-site communication conditions and control accuracy requirements; for example, it can be set to 1 minute or 5 minutes. Preferably, the status data of the temperature control actuators can be the start / stop status or power data of the temperature control actuators.

[0046] In this embodiment, a Time-lag Response Graph (TRG) is constructed based on historical operating data. The TRG uses temperature control actuators and temperature monitoring points as nodes, and directed edges represent the temperature influence relationship between the temperature control actuators and the temperature monitoring points. Each directed edge records the temperature influence data of the temperature control actuator on the temperature monitoring point. This temperature influence data includes at least one or more of the following: onset delay, peak delay, peak temperature influence, net total temperature effect, effective temperature effect per unit energy consumption, reliable lower boundary value of the peak temperature influence, and graph version information. Understandably, the onset delay refers to the time when the temperature first undergoes a detectable change after the temperature control actuator performs a control action; the peak delay refers to the time when the temperature change reaches its maximum value after the temperature control actuator performs a control action. The onset delay and peak delay together determine the minimum duration of the control action and the timing of benefit assessment; the temperature influence peak reflects the maximum temperature influence value of the temperature control actuator on the temperature monitoring point, used to determine whether the temperature change requirement of the temperature monitoring point is met; the net temperature effect total is the cumulative total value of the temperature effect at each moment within a preset window, used to calculate the effective temperature per unit energy consumption of the temperature control actuator. The effective temperature influence quantity is the ratio of the total net temperature effect to the temperature control actuator within a preset window. Preferably, the effective temperature influence quantity per unit energy consumption of the temperature control actuator is used to obtain the subsequent first sequence, thereby realizing the generation of an energy-saving oriented control action sequence. The reliable lower boundary value of the temperature influence peak value is a conservative estimate, reflecting that, under a preset confidence level, the actual temperature influence peak value is not lower than the reliable lower boundary value of the temperature influence peak value. The graph version identifies the update and iteration of the time-delay response graph, ensuring that when generating the control action sequence in the subsequent generation of the control action sequence, the control action sequence obtained from the generation experience is consistent with the graph version used when generating the action.

[0047] Understandably, the time-delay response diagram constructed based on the historical operating data is used to reflect the time-delay thermal response law of the temperature control execution equipment in the data center to the temperature monitoring points. It is the physical knowledge basis for the subsequent construction of the current state of the data center, the target control action sequence, and the value estimation.

[0048] Preferably, for temperature control actuators of different types and specifications, their temperature impact data can be identified separately through corresponding independent historical data batches to ensure that the time-delay response diagram can accurately reflect the differentiated response characteristics of each temperature control actuator under actual operating conditions. Preferably, the construction of the time-delay response diagram can adopt a combination of offline batch updates and online recursive updates: the initial time-delay response diagram is constructed based on historical data in the offline stage, and the time-delay response diagram is recursively updated based on newly completed effective control events in the online stage to adapt to changes in computer room layout, equipment performance degradation, and airflow organization.

[0049] Specifically, such as Figure 2 As shown, constructing the time-delay response diagram based on the historical operational data may include the following steps: S110. Obtain the state change event of the temperature control execution device based on the historical operating data, and obtain the occurrence time of the state change event accordingly; In this embodiment, the status change events of the temperature control actuator in the historical operation data are first detected, such as the temperature control actuator changing from off to on, from on to off, changes in set temperature or wind speed, etc., and these are identified as preliminary anchor point events. At the same time, the occurrence time of the preliminary anchor point event is recorded, which is the data basis for obtaining the anchor point event.

[0050] S120. In the state change event, if the state change event meets a preset state condition, then the state change event is taken as an anchor event, and the occurrence time of the state change event is taken as the occurrence time of the anchor event; wherein the preset state condition is that the equipment state of the data center meets the preset state during a preset first time period before the occurrence time of the state change event, and the equipment state of the data center meets the preset state during a preset second time period after the occurrence time of the state change event. In this embodiment, anchor events are obtained from state change events based on preset state conditions. The preset state is that none of the temperature control devices detect a corresponding state change event, the heat data generated by the load does not experience a preset degree of abrupt change, and none of the temperature monitoring points malfunction. The first and second time periods can be adapted to actual conditions; for example, the first time period can be set to 15 minutes, and the second time period can be set to 30 minutes. For example, if the state change event occurs at 10:00, and the data center equipment states meet the preset state for the 15 minutes preceding and the 30 minutes following the event, then the state change event is used as the anchor event.

[0051] Preferably, if a state change event cannot meet the preset state conditions, it is marked as a mixed sample and excluded to avoid introducing interference data that affects the identification accuracy of the time delay response diagram.

[0052] S130. Obtain a reference window based on a preset first time period before the occurrence time of the anchor event, and obtain a response window based on a preset second time period after the occurrence time of the anchor event. In this embodiment, two time windows are established centered on the occurrence time of the anchor event: a baseline window and a response window. The baseline window is used to establish a temperature change baseline under conditions where there is no change in the state of the temperature-controlled actuator. Within the baseline window, the state of the temperature-controlled actuator remains unchanged to ensure that temperature changes primarily reflect external disturbances, such as changes in heat generated by the load or changes in outdoor temperature. The response window is used to capture the complete temperature response process after a change in the state of the temperature-controlled actuator. Preferably, the complete response process includes an onset phase, a peak phase, and a stabilization phase. If the data center has a large space and a dense load distribution, resulting in high thermal inertia, the lengths of the first and second time periods can be appropriately extended to ensure complete capture of the temperature response process.

[0053] S140. Based on a preset time interval, obtain the first actual temperature of the temperature monitoring point in the reference window from the historical operating data, and obtain the baseline temperature rise rate of the temperature monitoring point based on the first actual temperature. In this embodiment, the actual temperature values ​​of each temperature monitoring point within a baseline window are extracted from historical operating data at preset time intervals and recorded as the first actual temperature. Preferably, the time interval can be set to 1 minute or 5 minutes. Based on the change of the first actual temperature within the baseline window over time, the baseline temperature rise rate of each temperature monitoring point is calculated. The baseline temperature rise rate reflects the inherent temperature change trend of the temperature monitoring point itself under the influence of factors such as the natural changes in heat generated by the load and fluctuations in the outdoor environment, under the condition of no temperature control execution device state change event.

[0054] For example, if the baseline window is 15 minutes, the first actual temperature measured at the beginning of the first time period is 26.3°C, and the first actual temperature measured at the end of the first time period is 26.5°C, then the baseline temperature rise rate is (26.5-26.3) / 15 ≈ 0.013°C / minute. This baseline temperature rise rate is used to construct the exogenous disturbance model in subsequent steps to accurately separate the net temperature effect of the temperature control actuator from the actual temperature changes.

[0055] S150. Based on the time interval, obtain the heat data generated by the data center load, the outdoor temperature, and the second actual temperature of the temperature monitoring point in the response window from the historical operating data. In this embodiment, at the same time interval as in step S140, multi-dimensional data within the response window is extracted from historical operating data, including: heat data generated by the data center load, and the second actual temperature of each temperature monitoring point in the response window. The heat data generated by the load can be the power consumption of the load equipment, reflecting the heat load of the data center. This data will serve as input to the exogenous disturbance model in step S160, used to predict the temperature change trajectory of the temperature monitoring points under conditions of no temperature control execution equipment state change events.

[0056] Preferably, in addition to the heat data generated by the data center load and the outdoor temperature, optional data such as the fan speed of the temperature control actuators, compressor status, alarm information, communication quality of the data center, access control, or personnel activity can also be collected to further improve the prediction accuracy of the exogenous disturbance model. For example, if frequent personnel entering and exiting the data center leads to the loss of cooling capacity, this information can be included as a disturbance factor in the exogenous disturbance model.

[0057] S160. Based on the baseline temperature rise rate, the heat data generated by the load, the outdoor temperature, the second actual temperature, and the preset exogenous disturbance model, establish a directed edge between the temperature control execution device and the temperature monitoring point, and construct a time-delay response graph based on the temperature control execution device, the temperature monitoring point, and the directed edge.

[0058] In this embodiment, the baseline temperature rise rate, the heat data generated by the load, the outdoor temperature, and the second actual temperature are input into a preset exogenous perturbation model to predict the directed edges of the temperature control execution device to the temperature monitoring point at each acquisition time in the response window, thus serving as the data basis for obtaining the time-delay response map. The exogenous perturbation model can be a linear regression model, a Lightweight Gradient Boosting Machine (LightGBM) model, an eXtreme Gradient Boosting (XGBoost) model, a state-space model, a Gated Recurrent Unit (GRU) model, a Long Short-Term Memory (LSTM) network model, a Temporal Convolutional Network (TCN) model, a Transformer model, a digital twin model, or an integration of the above models.

[0059] In this embodiment, a time-delay response graph is constructed using temperature control actuators and temperature monitoring points as nodes and all directed edges as edges. This time-delay response graph records the time-delay response relationship of each temperature actuator in the data center to each temperature measurement point, including attributes such as effective delay, peak delay, peak temperature effect, net total temperature effect, effective temperature effect per unit energy consumption, reliable lower boundary value of the peak temperature effect, and graph version.

[0060] Specifically, such as Figure 3 As shown, establishing a directed edge between the temperature control execution device and the temperature monitoring point based on the baseline temperature rise rate, the heat data generated by the load, the outdoor temperature, the second actual temperature, and a preset exogenous disturbance model may include the following steps: S161. Input the baseline temperature rise rate, the heat data generated by the load and the outdoor temperature into the exogenous disturbance model for processing, and obtain the predicted temperature of the temperature monitoring point when there is no temperature control execution device in the response window. In this embodiment, the baseline temperature rise rate, the heat data generated by the load within the response window, and the outdoor temperature are used as input features and fed into a preset exogenous disturbance model. The exogenous disturbance model outputs the predicted temperature of each temperature monitoring point at each acquisition time within the response window. This predicted temperature represents the temperature change trajectory of the temperature monitoring point under the condition that the temperature control execution device does not undergo a state change and maintains the initial control state unchanged.

[0061] For example, if a temperature control device is activated at 10:00, the exogenous disturbance model uses the heat data generated by the load, the outdoor temperature, and the baseline temperature rise rate from 10:00 to 10:30 to predict the temperature change at the temperature monitoring point during the same period if no temperature control device is activated and the original control state is maintained. This predicted temperature reflects the temperature evolution trend of the load under the individual effects of natural operating changes and outdoor environmental fluctuations.

[0062] S162. The difference between the second actual temperature and the predicted temperature is taken as the total net temperature effect of the temperature control execution device on the temperature monitoring point in the anchor event. In this embodiment, the second actual temperature of the temperature monitoring point within the response window is subtracted from the predicted temperature at each acquisition time, and the difference is calculated. This difference is used as the total net temperature effect of the temperature control execution device on the temperature monitoring point in the anchor point event. The total net temperature effect reflects the temperature change contributed solely by the change in the state of the temperature control execution device, excluding the influence of exogenous disturbances such as natural changes in heat generated by the load and fluctuations in the outdoor environment.

[0063] For example, if at the data acquisition time of 10:06, the actual temperature at the temperature monitoring point is 26.4°C, while the predicted temperature by the exogenous disturbance model is 26.6°C, then the difference is -0.2°C, indicating that the temperature execution device generated a net temperature effect of 0.2°C on the temperature monitoring point 6 minutes after it was turned on. Similarly, at the data acquisition time of 10:21, the actual temperature is 25.7°C, and the predicted temperature is 26.9°C, with a difference of -1.2°C, representing the net temperature effect reached by the temperature execution device 21 minutes after it was turned on.

[0064] S163. Based on the total net temperature effect of the temperature control execution device on the temperature monitoring point among all anchor point events of the temperature execution device on the temperature monitoring point, calculate the total net temperature effect of the temperature control execution device on the temperature monitoring point and the corresponding confidence interval. In this embodiment, the total net temperature effect of all anchor events of the same temperature control actuator on the same temperature monitoring point is collected, and the mean of all the total net temperature effects is calculated as the total net temperature effect. The standard deviation of all the total net temperature effects is calculated. Preferably, the sample size of all anchor events of the same temperature control actuator on the same temperature monitoring point can be obtained, and the degrees of freedom are obtained based on the sample size. The critical value of all anchor events of the same temperature control actuator on the same temperature monitoring point is obtained based on the sample size and a preset distribution table. Preferably, the distribution table can be preset based on the sample size and can be a t-distribution table or a z-distribution table. The error of all anchor events of the same temperature control actuator on the same temperature monitoring point is obtained based on the standard deviation and the sample size. The upper and lower boundaries of the confidence interval are obtained based on the total net temperature effect, the critical value, and the error, thereby constituting the confidence space.

[0065] For example, consider all anchor point events of the first temperature control actuator on the first temperature monitoring point. At the 21st minute of the data acquisition, the system collected the total net temperature effect of the first temperature control actuator on the first temperature monitoring point from 15 anchor point events, which are: -1.0, -1.2, -0.9, -1.3, -1.1, -1.4, -1.2, -1.0, -1.3, -1.1, -0.8, -1.2, -1.0, -1.1, and -1.4°C. The mean of these 15 data points is calculated to be -1.2°C, and this mean is used as the total net temperature effect at this acquisition time, with a standard deviation of 0.3°C. Based on the confidence interval estimation requirements, the sample size at this acquisition time is 15. The degrees of freedom df = 15 are obtained based on the sample size of 15. 1 = 14. Based on the preset 95% confidence level and 14 degrees of freedom, referring to the t-distribution table, the critical value is approximately 2.145. Subsequently, based on the standard deviation of 0.3 and the sample size of 15, the standard error is calculated as 0.3 / √15 ≈ 0.077. Finally, the net temperature effect of -1.2 is added to and subtracted from the product of the critical value 2.145 and the standard error 0.077, respectively, to obtain the lower boundary of the confidence interval as -1.2. 2.145 × 0.077 ≈ -1.365°C, with an upper boundary of -1.2 + 2.145 × 0.077 ≈ -1.035°C, thus forming the confidence interval for this data acquisition moment. This confidence interval indicates that, at a 95% confidence level, the actual average cooling effect of the first temperature control actuator on the first temperature monitoring point 21 minutes after activation falls between -1.365°C and -1.035°C. Through this method, even with a limited sample size, the reliability of the temperature effect estimate can be quantified based on the confidence interval, and the lower boundary of the confidence level used for establishing directed edges can be further extracted as a conservative estimate of the temperature effect in subsequent safety verification.

[0066] S164. Based on the total net temperature effect and the confidence interval, obtain the temperature impact data of the temperature control execution device on the temperature monitoring point; In this embodiment, based on the total net temperature effect and confidence interval, the average temperature influence curve of the temperature control actuator on the temperature monitoring point on the time axis is obtained, thereby estimating the temperature influence data of the temperature control actuator on the temperature monitoring point. The temperature influence data reflects the complete distribution characteristics of the temperature influence of the temperature control actuator on the temperature monitoring point, including but not limited to onset delay, peak delay and temperature influence peak value, total net temperature effect, effective temperature effect per unit energy consumption, reliable lower boundary value of temperature influence peak value and one or more of the following graph versions.

[0067] For example, by cumulatively estimating 15 anchor point events of the first temperature control actuator, the peak temperature effect of the first temperature control actuator on the temperature monitoring point is -1.4°C, the preset 95% confidence interval is [-1.7°C, -1.1°C], the onset delay is 6 minutes, the peak delay is 21 minutes, the cumulative temperature effect is -18.2°C·min, and the effective temperature effect per unit energy consumption is -0.35°C / kW.

[0068] S165. Based on the total net temperature effect and the confidence interval, establish a directed edge from the temperature control execution device to the temperature monitoring point, and use the corresponding temperature influence data as the attribute of the directed edge.

[0069] Specifically, establishing a directed edge from the temperature control actuator to the temperature monitoring point based on the total net temperature effect and the confidence interval includes: Determine whether the total net temperature effect is greater than or equal to a preset total threshold, and whether the lower confidence level is greater than or equal to a preset confidence level. If so, establish a directed edge from the temperature control execution device to the temperature monitoring point; otherwise, do not establish a directed edge from the temperature control execution device to the temperature monitoring point. The lower confidence level is obtained based on the confidence interval.

[0070] In this embodiment, the establishment of directed edges is determined using a dual-condition approach. First, the total net temperature effect must be greater than or equal to a preset total threshold to ensure that the temperature control actuator has a real and sufficiently significant temperature effect on the temperature monitoring point. Second, the lower confidence level must also be greater than or equal to a preset confidence level threshold to ensure that the temperature influence relationship is statistically reliable and that false edges will not enter the time-delay response graph due to insufficient sample size or random noise. A directed edge from the temperature control actuator to the temperature monitoring point is established only when both conditions are met simultaneously. If the total net temperature effect is greater than or equal to the preset total threshold, and the lower confidence level is greater than or equal to the preset confidence level threshold, then a directed edge from the temperature control actuator to the temperature monitoring point is established. If the total net temperature effect is less than the preset total threshold, and the lower confidence level is less than the preset confidence level threshold, then a directed edge from the temperature control actuator to the temperature monitoring point is not established.

[0071] S200: Obtain the current operating data of the data center; In this embodiment, the current operating data of the data center is collected in real time. This current operating data includes at least: the current temperature value of each temperature monitoring point, the current start / stop status of each temperature control device, the heat data generated by the load in the data center, the current outdoor temperature, and the collection timestamp. Optionally, data such as wind speed, compressor status of the temperature control devices, alarm information of the data center, communication quality, access control, or personnel activity can also be collected.

[0072] S300. Based on the current operating data, the time-delay response diagram, and the historical operating data, obtain the current status of the data center at the current moment; Specifically, such as Figure 4 As shown, obtaining the current status of the data center at the current moment based on the current operating data, the time-delay response diagram, and the historical operating data may include the following steps: S310. Based on the current operating data, obtain the heat data generated by the load in the data center at the current moment, the current temperature data of several temperature monitoring points, the current operating data of several temperature control execution devices, and the current environmental disturbance data; In this embodiment, the heat data generated by the load at the current moment is obtained from the current operating data, that is, the real-time power consumption of the load equipment, reflecting the current heat load level of the data center; the current temperature data of several temperature monitoring points, which can be the real-time temperature value of each temperature monitoring point; the current operating data of several temperature control execution devices, that is, the start / stop status, power, set temperature value, operation / stop time, availability and execution feedback of each temperature control execution device; and the current environmental disturbance data, which can be the outdoor temperature.

[0073] For example, taking a data center with 4 temperature control actuators and 2 temperature monitoring points as an example, the current load generates 35kW of heat at the current moment; the current temperature at the first temperature monitoring point is 26.5°C, with a temperature rise rate of 0.05°C / min; the current temperature at the second temperature monitoring point is 24.8°C; the first temperature control actuator has been running for 2 hours, the second temperature control actuator has been shut down for 30 minutes, AC4 has been running for 2 hours; the outdoor temperature is 32°C, and the current load generates 32kW of heat. Preferably, the heat data for a preset time in the future can also be predicted based on the current load heat data and a preset prediction model. For example, the predicted heat data generated by the load within 30 minutes is stable at 32-36kW.

[0074] S320. Obtain the temperature impact data of the temperature control execution device on the temperature monitoring point based on the time delay response diagram; In this embodiment, based on the current temperature control actuators and temperature monitoring points, the temperature impact data of each temperature control actuator on the temperature monitoring point is retrieved and extracted from the time-delay response diagram. For example, for the first temperature monitoring point, the following is extracted from the time-delay response diagram: the total net temperature effect of the first temperature control actuator on the first temperature monitoring point is -1.4°C, with an onset delay of 6 minutes, a peak delay of 21 minutes, and an effective temperature effect per unit energy consumption of -0.35°C / kW; the total net temperature effect of the second temperature control actuator on the first temperature monitoring point is -0.9°C, with an onset delay of 8 minutes, a peak delay of 18 minutes, and an effective temperature effect per unit energy consumption of -0.28°C / kW.

[0075] Preferably, the temperature impact data can be implemented using methods such as edge attribute statistical pooling, attention aggregation, graph neural networks, or fixed-length summaries formed based on temperature monitoring points. The temperature impact data converts the discrete graph structure information (nodes, edges, and edge attributes) in the time-delay response graph into a fixed-length vector representation acceptable to the value network.

[0076] S330. Obtain the historical execution data of the previous target control action sequence based on the historical operation data; In this embodiment, historical execution data of the previous target control action sequence is extracted from historical operation data. The historical execution data of the previous target control action sequence includes one or more of the following: historical termination type, historical safety prefix length, historical graph version used, historical target control action number, historical control step index, and historical duration of the target control action sequence. For example, the termination type includes normal stabilization, new exogenous disturbance, reaching the maximum duration, safe abort, data center anomaly, communication anomaly, or manual takeover, etc. The safety prefix length is the number of control steps successfully executed when the target control action sequence is safely aborted.

[0077] For example, if the previous target control action sequence terminates prematurely after running 3 control substeps due to a new exogenous disturbance, the historical execution data of the previous target control action sequence will be a historical termination type of "new exogenous disturbance" with a historical safety prefix length of 3. Upon receiving this information, the value network will know that the previous target control action sequence has been interrupted and that a reassessment of environmental stability may be necessary.

[0078] If there is no previous target control action sequence, all historical execution data of the previous target control action sequence are set to default values, such as setting the historical termination type to the initial value, the safety prefix length to 0, and the graph version to the default version.

[0079] S340. The current state of the data center is obtained by fusing one or more of the following: the heat data generated by the load at the current moment, the temperature data of the temperature monitoring point at the current moment, the operating data of the temperature control execution device at the current moment, the environmental disturbance data at the current moment, the temperature influence data, and the historical execution data of the previous target control action sequence.

[0080] In this embodiment, preferably, the heat data generated by the load at the current moment, the temperature data of the temperature monitoring point at the current moment, the operating data of the temperature control execution device at the current moment, the environmental disturbance data at the current moment, the temperature influence data, and the historical execution data of the previous target control action sequence are fused to obtain the current state of the data center at the current moment. The fusion can be achieved by splicing the various components into a unified state vector.

[0081] Specifically, the current state s of the data center k Represented as: s k = Concat(x k , Enc(G k J k hot ), c k ) in, This represents the concatenation function, x k This represents the heat data generated by the load at the current moment, the temperature data of the temperature monitoring point at the current moment, the operating data of the temperature control actuator at the current moment, and the environmental disturbance data at the current moment; Enc(G k J k hot ) represents the temperature impact data of the temperature monitoring points in the time-delay response diagram; c k This represents the historical execution data of the previous target control action sequence.

[0082] S400. Generate several candidate control action sequences based on the current state and the time-delay response diagram; In this embodiment, several candidate control action sequences are dynamically compiled and generated based on the current state and the time-delay response diagram. Unlike traditional methods that use a fixed global action table, this embodiment recompiles dynamic candidate control action sequences based on the current state and the time-delay response diagram. Each candidate control action sequence contains two or more consecutive control steps, preferably including at least two of the following stages: a cooling phase, an observation phase, a release phase, and a stable holding phase.

[0083] Specifically, such as Figure 5 As shown, generating a number of candidate control action sequences based on the current state and the time-delay response diagram may include the following steps: S410. Determine the target temperature monitoring point based on the current state; wherein the first difference between the actual temperature of the target temperature monitoring point and the preset temperature threshold is less than or equal to the preset difference threshold. In this embodiment, based on the current state, a temperature monitoring point whose actual temperature is close to a preset temperature monitoring point is identified as a target temperature monitoring point. Specifically, the difference between the actual temperature of each temperature monitoring point and a preset temperature threshold is calculated. If the difference is less than or equal to the preset difference threshold, that is, the actual temperature has approached or reached a warning value, then the temperature monitoring point is determined as the target temperature monitoring point.

[0084] S420. Based on the current state of the data center, obtain the temperature impact data of each temperature control execution device on the target temperature monitoring point; In this embodiment, the temperature impact data of each temperature control actuator on the target temperature monitoring point is extracted. For example, for the target temperature monitoring point, the following data are obtained: the total net temperature effect of the first temperature control actuator on the target temperature monitoring point is -1.4°C, and the effective temperature effect per unit energy consumption is -0.35°C / kW; the total net temperature effect of the second temperature control actuator on the target temperature monitoring point is -0.9°C, and the effective temperature effect per unit energy consumption is -0.28°C / kW; the total net temperature effect of the third temperature control actuator on the target temperature monitoring point is -0.4°C, and the effective temperature effect per unit energy consumption is -0.12°C / kW.

[0085] S430. Based on the first difference and the temperature impact data of each of the temperature control actuators on the target temperature monitoring point, determine the target temperature control actuator from the temperature control actuators; In this embodiment, the minimum effective temperature coverage required for the target temperature monitoring point is obtained based on the first difference. This minimum effective temperature coverage reflects the heat intervention required for the target temperature monitoring point to return to a safe state. Combining the temperature impact data of each temperature control actuator on the target temperature monitoring point, a target temperature control actuator that must operate is determined from all temperature control actuators. Specifically, if the total net temperature effect of the first temperature control actuator on the target temperature monitoring point exceeds the minimum effective temperature coverage required for the target temperature monitoring point, then the first temperature control actuator is determined to be a target temperature control actuator that must operate.

[0086] For example, if the minimum effective temperature effect required for the target temperature monitoring point is a temperature drop of at least 0.5°C, the net temperature effect of the first temperature control actuator is -1.4°C, and the net temperature effect of the second temperature control actuator is -0.9°C. That is, both the first and second temperature control actuators meet this requirement and are identified as the target temperature control actuators. The net temperature effect of the third temperature control actuator is only 0.4°C, which is lower than 0.5°C. It cannot be used as the target temperature control actuator alone, but it can be used as an auxiliary device on the basis that the target temperature control actuator is already in operation.

[0087] S440. Sort the target temperature control execution device according to the temperature influence data to obtain a first order; generate a number of candidate control action sequences with different control strategies based on the first order, each candidate control action sequence including at least two consecutive control steps, and each control step including a control operation on the target temperature control execution device and a corresponding preset execution time.

[0088] In this embodiment, based on the effective temperature effect per unit energy consumption of each target temperature actuator in the temperature influence data, the target temperature control actuators are sorted to obtain a first order representing the release priority. The release order is determined according to the principle of releasing the most efficient devices last and the least efficient devices first. For example, the first target temperature actuator with an effective temperature effect per unit energy consumption of -0.35°C / kW is the most efficient device and should be released last; the second target temperature actuator with an effective temperature effect per unit energy consumption of -0.28°C / kW is a medium-efficiency device and should be released next; the third target temperature actuator with an effective temperature effect per unit energy consumption of -0.12°C / kW is an inefficient device and should be released first.

[0089] Then, based on the first sequence, a number of candidate control action sequences corresponding to different release strategies are generated using a preset algorithm. Each candidate control action sequence contains two or more consecutive control steps, each control step includes a control operation on the target temperature control device and a corresponding preset execution time, wherein the control operation may include turning the target temperature control device on or off, and the execution time may be 5 minutes for each control step.

[0090] For example, the following candidate control action sequences can be generated: Candidate Sequence A: The first control step is to first turn on the third target temperature execution device and the second target temperature execution device as a 0-10 minute cooling stage; the second control step is to release and stop the third target temperature execution device after 10 minutes, while keeping the second target temperature execution device running, as a 10-15 minute release stage; the third control step is to keep the second target temperature execution device running until the temperature stabilizes, as a 15-25 minute stabilization stage. Candidate Sequence B: The fourth control step is to first turn on the third target temperature execution device and the second target temperature execution device as a 0-10 minute cooling stage; the fifth control step is to release and stop the second target temperature execution device after 10 minutes, while keeping the third target temperature execution device running, as a 10-15 minute release stage; the sixth control step is to keep the third target temperature execution device running until the temperature stabilizes, as a 15-25 minute stabilization stage. Candidate Sequence C: Seventh control step: Turn on the second target temperature actuator as a 0-10 minute cold preservation stage. The eighth control step is to keep the second target temperature actuator running until the temperature stabilizes as a 10-15 minute stable holding stage.

[0091] Preferably, when generating candidate macro actions, at least one of the following can be used to generate candidate control action sequences: genetic algorithm, beam search algorithm, heuristic search algorithm, or mixed integer programming algorithm.

[0092] S500: Based on the current state, the candidate control action sequence, and the preset value network, obtain the target control action sequence from the candidate control action sequence; In this embodiment, the current state and candidate control action sequences are input into a preset value network, which estimates the value of the candidate control action sequences. The value network can employ a Double DQN architecture, comprising an online network and a target network. The online network selects the target control action sequence from the candidate control action sequences at the current time, and the target network evaluates the subsequent value of the next selected target control action sequence. Specifically, based on the value estimates of each candidate control action sequence, the candidate control action sequence with the highest value estimate is determined as the target control action sequence. During the training phase, the ε-greedy algorithm, the Boltzmann exploration algorithm, or the noise parameter exploration algorithm can be used to select the target control action sequence.

[0093] Specifically, such as Figure 6 As shown, obtaining the target control action sequence from the candidate control action sequence based on the current state, the candidate control action sequence, and a preset value network may include the following steps: S510. Based on the candidate control action sequence, obtain one or more safety control action sequences that satisfy preset safety constraints; In this embodiment, the complete execution trajectory of each candidate control action sequence is verified for temperature coverage, temperature range prediction for each control step, and operational constraints of the temperature control execution device are verified. Specifically, the first constraint is: in, Indicates the first A candidate sequence of control actions, Indicates the first The first step in the candidate control action sequence to start the target temperature control actuator. One control step; For the first The temperature control actuator is for the first The temperature monitoring point at the first The total net temperature effect of each control step For the first The minimum effective temperature coverage required for each temperature monitoring point.

[0094] Second constraint: in, Indicates the first In the candidate control action sequence, the first... During the first control step, the... The upper bound of the predicted temperature for the first temperature monitoring point, preferably, the upper bound of the predicted temperature is based on the first temperature monitoring point. The actual temperature of each temperature monitoring point, the heat generated by the current load, and the preset temperature prediction model are obtained. Indicates the first Temperature threshold at each temperature monitoring point Indicates the first Each temperature monitoring point has a preset safety margin.

[0095] After satisfying the first and second constraints mentioned above, only candidate control action sequences that also satisfy the complete trajectory constraint can be selected as safe control action sequences. Candidate control action sequences that do not satisfy the above constraints will not be included in the subsequent value ranking. For example, if candidate sequence B exceeds the safety margin in the upper bound of the predicted temperature in the second control step, then candidate sequence B is eliminated by the safety mask; if candidate sequence A satisfies the above constraints throughout the entire process, then it enters the set of safe control action sequences.

[0096] S520. Input the current state and the security control action sequence into the value network, perform value estimation on the security control action sequence, and obtain the value estimate corresponding to the security control action sequence; In this embodiment, the current state and the sequence of safety control actions are input into the value network. The value network outputs a corresponding value estimate for the sequence of safety control actions.

[0097] S530. Obtain the safety control action sequence with the highest value estimate as the target control action sequence.

[0098] S600: Control the temperature control actuator to execute according to the target control action sequence, and complete the control of the temperature control actuator.

[0099] Specifically, controlling the temperature control actuator to perform actions according to the target control sequence includes: The temperature control execution device is controlled sequentially according to the control steps of the target control action sequence. After each control step is completed, it is checked whether the actual temperature of the temperature monitoring point meets the preset first condition and whether the temperature control execution device meets the preset second condition. If so, the currently executed control step is updated to the preset safety prefix data. Otherwise, cancel the unexecuted control steps in the target control action sequence and restore the control steps corresponding to the security prefix data, and re-execute the step of obtaining the target control action sequence from the candidate control action sequence based on the current state, the candidate control action sequence and the preset value network.

[0100] In this embodiment, control steps are executed sequentially according to the target control action sequence. After each control step is completed, it is checked whether the actual temperature of the temperature monitoring point meets a preset first condition and whether the temperature control execution device meets a preset second condition. Preferably, the first condition is whether the actual temperature of the temperature monitoring point is within the prediction range and whether the effective temperature coverage meets the required minimum effective temperature coverage. The prediction range is obtained in a similar way to the above-mentioned upper bound of the prediction temperature, based on the actual temperature of the temperature monitoring point, the heat data generated by the current load, and a preset temperature prediction model. Second, it is checked whether the actual state of the temperature control execution device is consistent with the control command and whether the communication status of the temperature control execution device is normal. If both checks pass, the currently executed control step sequence is saved as a safety prefix.

[0101] If the check fails, the unexecuted control steps in the target control action sequence are canceled, and the device state corresponding to the safety prefix is ​​restored. Risk temperature monitoring points are obtained based on the first condition, and available temperature control devices whose net temperature effect total satisfies the first condition are selected from the time-delay response diagram based on the risk temperature monitoring points to perform directional recovery. If the directional recovery still does not satisfy the first and second conditions, step S500 is re-executed to obtain a new target control action sequence.

[0102] Specifically, such as Figure 7 As shown, after the step of controlling the temperature control actuator according to the target control action sequence to complete the control of the temperature control actuator, the method further includes the following steps: S710. Record the actual execution data executed by the temperature control actuator; In this embodiment, the actual execution data of this target control action sequence is recorded. The actual execution data includes: the actual energy consumption data of each temperature control execution device, the actual temperature value of each temperature monitoring point at each control step, the actual total duration from the start control step to the end control step, the actual termination type, and the safety prefix record during the execution process.

[0103] For example, if the temperature stabilizes and terminates normally after the target control action sequence has been executed for 25 minutes, the actual energy consumption data is recorded as the total power consumption data of the first temperature control execution device and the second temperature control execution device within 25 minutes, the actual temperature values ​​of the first temperature monitoring point and the second temperature monitoring point in each control step within 25 minutes are recorded, the actual total duration is recorded as 25 minutes, and the actual termination type is normal stabilization.

[0104] S720. Obtain the counterfactual control action sequence of the target control action sequence; In this embodiment, a counterfactual control action sequence corresponding to the target control action sequence is obtained, that is, a reference action sequence that maintains the state of all devices in the data center unchanged when the target control action sequence begins execution. A counterfactual control action sequence refers to a sequence in which the target control action sequence is not executed for the same actual total duration as the target control action sequence, but rather the device states and settings in the data center remain unchanged when the target control action sequence begins execution.

[0105] For example, if the first temperature actuator is running and the second, third, and fourth temperature actuators are stopped when the target control action sequence begins execution, and the target control action sequence turns on the second and third temperature actuators, then the counterfactual control action sequence is that the first temperature actuator is running and the second, third, and fourth temperature actuators are stopped.

[0106] S730. Based on the counterfactual control action sequence, perform counterfactual state reasoning on the data center to obtain counterfactual execution data; In this embodiment, the aforementioned exogenous disturbance prediction model is used to perform counterfactual state reasoning on the counterfactual control action sequence under the conditions of heat data generated by the same load and outdoor environmental data as the target control action sequence. Specifically, starting from the initial state of the data center equipment, and using heat data generated by the same load and outdoor environmental data as the input, the exogenous disturbance model is driven to predict the predicted temperature of each temperature monitoring point and the predicted energy consumption data of the temperature control execution device under the condition that the initial state of the data center equipment remains unchanged.

[0107] For example, if the first temperature actuator is running while the second, third, and fourth temperature actuators are shut down, under the same load conditions of 35kW of heat and 32°C outdoor temperature, the exogenous disturbance model predicts that the temperature at the first temperature monitoring point will naturally rise from 26.5°C to 28.0°C. However, after the actual execution of the target control action sequence, the actual temperature at the first temperature monitoring point drops to 25.6°C. By comparing the two, the actual temperature effect of the target control action sequence can be accurately attributed to the actual temperature effect.

[0108] S740. Obtain counterfactual attribution benefits based on the actual execution data and the counterfactual execution data; In this embodiment, counterfactual periodic attribution benefits are calculated based on a comparison between actual execution data and counterfactual execution data. The counterfactual attribution benefits... It can be represented as: in, This represents the predicted energy consumption data of temperature-controlled actuators in the counterfactual execution data. This refers to the actual energy consumption data of the temperature control actuator in the actual execution data; This represents the excess temperature between the predicted temperature and the corresponding temperature threshold at the temperature monitoring point in the counterfactual execution data. The actual temperature of the temperature monitoring point in the actual execution data exceeds the corresponding temperature threshold. This represents the number of times the temperature control device status was switched in the actual execution data. To predict costs; The cost of safe abort and recovery is defined as follows: the predicted cost is obtained based on temperature prediction cost, and the safe abort and recovery cost is calculated based on the actual execution data if it includes a safe abort and recovery process. These represent the corresponding preset weights.

[0109] S750. Update the network parameters of the value network based on the counterfactual attribution benefits.

[0110] In this embodiment, a target control action sequence is used as a period to serve as an empirical sample for updating the value network. It is understood that the duration of the period is determined by events such as temperature stabilization, exogenous disturbances, maximum execution time, or safety abort; therefore, different empirical samples have different actual time spans. When the online network selects a target control action sequence in the next safety control action sequence, the target network updates the target control action sequence, i.e., the [number missing]th [number missing]. Network parameters for evaluating a sequence of target control actions : in, The termination type of the target control action sequence; This represents the preset time discount factor. This represents the natural exponential function. The periodic duration of the target control action sequence; Represents the target network function. Indicates the first The state of a target control action sequence Indicates the first A set of safety control action sequences This represents the online network function. For target control action sequences that terminate normally or with new disturbances that can be controlled, their subsequent value is retained; for target control action sequences that terminate due to safety aborts, equipment failures, communication anomalies, or manual takeover, Set to 0 or use a specific safety target value.

[0111] Understandably, compared to ordinary fixed step size, the above network parameters are simultaneously affected by the set of security control action sequences, cycle duration, and termination reason. This can reduce the value deviation caused by different thermal response cycles being mistakenly regarded as single-step transfers of equal duration, making the value estimation more accurately match the actual physical time scale of the data center thermal process, thereby improving the convergence stability of the value network under thermal inertia conditions and the energy-saving effect of the control strategy.

[0112] like Figure 8 As shown in the illustration, this application also provides a control system for a data center temperature control execution device. Optionally, the system is applied to a data center, which includes several devices, each including a load, a temperature control execution device, and temperature monitoring points. The system includes: The system includes a time-delay response diagram acquisition module 811, a current running data acquisition module 812, a current state acquisition module 813, a candidate control action sequence acquisition module 814, a target control action sequence acquisition module 815, and a control module 816, among which: The time-delay response diagram acquisition module 811 is used to acquire historical operating data in the data center and construct a time-delay response diagram based on the historical operating data. The time-delay response diagram includes temperature impact data of temperature control execution equipment on temperature monitoring points. In this embodiment, the time-delay response graph acquisition module 811 can be used to perform... Figure 1 For a detailed description of the time-delay response diagram acquisition module 811, please refer to the description of step S100 shown.

[0113] Specifically, the time-delay response diagram acquisition module 811 is also used to perform Figure 2 Steps S110-S160 shown and Figure 3 The specific implementation of steps S161-S165 shown can be found in the description of steps S110-S160 and S161-S165.

[0114] The current operating data acquisition module 812 is used to acquire the current operating data of the data center; In this embodiment, the current running data acquisition module 812 can be used to execute... Figure 1 For a detailed description of the currently running data acquisition module 812, please refer to the description of step S200 shown.

[0115] The current status acquisition module 813 is used to acquire the current status of the data center at the current moment based on the current operating data, the time delay response diagram and the historical operating data; In this embodiment, the current state acquisition module 813 can be used to perform... Figure 1For a detailed description of the current state acquisition module 813, please refer to the description of step S300 shown.

[0116] Specifically, the current state acquisition module 813 is also used to perform Figure 4 For specific implementation of steps S310-S340 shown, please refer to the description of steps S310-S340.

[0117] The candidate control action sequence acquisition module 814 is used to generate several candidate control action sequences based on the current state and the time-delay response diagram; In this embodiment, the candidate control action sequence acquisition module 814 can be used to execute... Figure 1 For a detailed description of the candidate control action sequence acquisition module 814, see step S400 shown below. For a detailed description of step S400, please refer to the description of step S400.

[0118] Specifically, the candidate control action sequence acquisition module 814 is also used to execute Figure 5 For a detailed implementation of steps S410-S440, please refer to the description of steps S410-S440.

[0119] The target control action sequence acquisition module 815 is used to acquire the target control action sequence from the candidate control action sequence based on the current state, the candidate control action sequence, and a preset value network. In this embodiment, the target control action sequence acquisition module 815 can be used to execute... Figure 1 For a detailed description of the target control action sequence acquisition module 815, see step S500 shown. For a detailed description of step S500, please refer to the description of step S500.

[0120] Specifically, the target control action sequence acquisition module 815 is also used to execute Figure 6 The specific implementation of steps S510-S530 shown can be found in the description of steps S510-S530.

[0121] The control module 816 is used to control the temperature control actuator to perform according to the target control action sequence, thereby completing the control of the temperature control actuator.

[0122] In this embodiment, the control module 816 can be used to execute... Figure 1 For a detailed description of the control module 816, please refer to the description of step S600 shown.

[0123] Specifically, the control module 816 is also used to perform Figure 7The specific implementation of steps S710-S750 shown can be found in the description of steps S710-S750.

[0124] This application also provides an electronic device, the structure of which is as follows: Figure 9 As shown, the electronic device includes a memory 911, a processor 912, a communication module 913, and an input / output interface 914, etc. Optionally, the memory 911, the processor 912, the communication module 913, and the input / output interface 914 can be connected and communicate with each other through a bus 915.

[0125] The memory 911 is used to store one or more computer programs and to transfer the code of the computer programs to the processor 912; when the one or more computer programs are executed by the processor 912, a control method for a computer room temperature control execution device according to an embodiment of this application is implemented.

[0126] Optionally, the electronic device can be connected to a network via the communication module 913 to communicate with other devices, such as terminals or servers, to achieve data interaction. The electronic device can be various forms of digital computers, exemplarily such as desktop computers, servers, workbenches, mainframes, or other types of computers. The electronic device can also be various forms of mobile terminals, exemplarily such as smartphones, tablets, wearable devices (such as helmets, glasses, watches, etc.), and other similar mobile terminals.

[0127] Optionally, the electronic device can connect to required input / output devices, such as a keyboard or display device, via the input / output interface 914. The electronic device itself may have a display device, and other display devices can also be connected externally via the input / output interface 914. Optionally, a storage device, such as a hard disk, can also be connected via the input / output interface 914 to store data from the electronic device, read data from the storage device, or store data from the storage device in the memory 911. It is understood that the input / output interface 914 can be a wired interface or a wireless interface. Depending on the actual application scenario, the device connected to the input / output interface 914 can be a component of the electronic device or an external device connected to the electronic device when needed.

[0128] Optionally, the memory 911 may be a volatile memory and / or a non-volatile memory. The volatile memory may be a random access memory, etc., and the non-volatile memory may be a read-only memory, a programmable read-only memory, an erasable programmable read-only memory, an electrically erasable programmable read-only memory, or a flash memory, etc.

[0129] Optionally, the computer program stored in the memory 911 can be divided into one or more modules, which are stored in the memory 911 and executed by the processor 912 to perform the method provided in this embodiment. The one or more modules can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the electronic device.

[0130] Optionally, the processor 912 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the processor 912 include, but are not limited to, a central processing unit, a graphics processing unit, a digital signal processor, various special-purpose artificial intelligence computing chips, various processors running machine learning model algorithms, and can also be any suitable controller, microcontroller, processor, etc. The processor 912 executes the various methods and processes of this embodiment, exemplarily, such as the control method of a computer room temperature control execution device according to an embodiment of this application.

[0131] Optionally, the bus 915 may include a path for transmitting information. Depending on its function, the bus 915 may be classified as an address bus, a data bus, a control bus, etc.

[0132] In an optional implementation, this application embodiment also provides a computer storage medium storing a computer program thereon. When the computer program is executed by a computer, it enables the computer to perform the methods described in the above-described method embodiments. Part or all of the computer program can be loaded and / or installed on the memory 911 of an electronic device. When the computer program is executed by the processor 912, one or more steps of a control method for a computer room temperature control device according to an embodiment of this application can be performed.

[0133] Optionally, the computer-readable storage medium may be a random access memory, a read-only memory, a programmable read-only memory, an erasable programmable read-only memory, an electrically erasable programmable read-only memory, etc.

[0134] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the technical solution of the present invention, and are not intended to limit the specific implementation of the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the claims of the present invention should be included within the protection scope of the claims of the present invention.

Claims

1. A control method for a computer room temperature control actuator, characterized in that, The method is applied to a data center, which includes several devices, including loads, temperature control actuators, and temperature monitoring points. The method includes: Acquire historical operating data from the data center, and construct a time-delay response diagram based on the historical operating data. The time-delay response diagram includes data on the temperature impact of the temperature control execution device on the temperature monitoring point. Obtain the current operating data of the data center; Based on the current operating data, the time-delay response diagram, and the historical operating data, obtain the current status of the data center at the current moment; Based on the current state and the time-delay response diagram, several candidate control action sequences are generated; Based on the current state, the candidate control action sequence, and the preset value network, obtain the target control action sequence from the candidate control action sequence; The temperature control device is controlled to perform according to the target control action sequence, thereby completing the control of the temperature control device.

2. The method according to claim 1, characterized in that, The construction of the time-delay response diagram based on the historical operational data includes: Based on the historical operating data, obtain the state change events of the temperature control actuator, and obtain the occurrence time of the state change events accordingly; In the state change event, if the state change event meets the preset state conditions, then the state change event is taken as the anchor event, and the occurrence time of the state change event is taken as the occurrence time of the anchor event; wherein the preset state conditions are that the equipment state of the data center meets the preset state during a preset first time period before the occurrence time of the state change event, and the equipment state of the data center meets the preset state during a preset second time period after the occurrence time of the state change event. A baseline window is obtained based on a preset first time period before the occurrence of the anchor event, and a response window is obtained based on a preset second time period after the occurrence of the anchor event. Based on a preset time interval, the first actual temperature of the temperature monitoring point in the baseline window is obtained from the historical operating data, and the baseline temperature rise rate of the temperature monitoring point is obtained based on the first actual temperature. Based on the time interval, obtain the heat data generated by the data center load, the outdoor temperature, and the second actual temperature of the temperature monitoring point in the response window from the historical operating data. Based on the baseline temperature rise rate, the heat data generated by the load, the outdoor temperature, the second actual temperature, and the preset exogenous disturbance model, a directed edge is established between the temperature control execution device and the temperature monitoring point. A time-delay response graph is constructed based on the temperature control execution device, the temperature monitoring point, and the directed edge.

3. The method according to claim 2, characterized in that, The process of establishing a directed edge between the temperature control device and the temperature monitoring point based on the baseline temperature rise rate, the heat data generated by the load, the outdoor temperature, the second actual temperature, and a preset exogenous disturbance model includes: The baseline temperature rise rate, the heat data generated by the load, and the outdoor temperature are input into the exogenous disturbance model for processing to obtain the predicted temperature of the temperature monitoring point when there is no temperature control execution device in the response window. The difference between the second actual temperature and the predicted temperature is taken as the total net temperature effect of the temperature control execution device on the temperature monitoring point in the anchor point event. Based on all anchor point events of the temperature actuator on the temperature monitoring point, the total net temperature effect of the temperature actuator on the temperature monitoring point and the corresponding confidence interval are calculated. Based on the total net temperature effect and the confidence interval, obtain the temperature impact data of the temperature control device on the temperature monitoring point; Based on the total net temperature effect and the confidence interval, a directed edge is established from the temperature control execution device to the temperature monitoring point, and the corresponding temperature influence data is used as the attribute of the directed edge.

4. The method according to claim 3, characterized in that, The step of establishing a directed edge from the temperature control actuator to the temperature monitoring point based on the total net temperature effect and the confidence interval includes: Determine whether the total net temperature effect is greater than or equal to a preset total threshold, and whether the lower confidence level is greater than or equal to a preset confidence level. If so, establish a directed edge from the temperature control execution device to the temperature monitoring point; otherwise, do not establish a directed edge from the temperature control execution device to the temperature monitoring point. The lower confidence level is obtained based on the confidence interval.

5. The method according to claim 1, characterized in that, The step of obtaining the current status of the data center at the current moment based on the current operating data, the time-delay response diagram, and the historical operating data includes: Based on the current operating data, obtain the heat data generated by the load in the data center at the current moment, the current temperature data of several temperature monitoring points, the current operating data of several temperature control devices, and the current environmental disturbance data. The temperature impact data of the temperature control execution device on the temperature monitoring point is obtained based on the time delay response diagram. Based on the historical operation data, obtain the historical execution data of the previous target control action sequence; The current state of the data center is obtained by fusing one or more of the following: the heat data generated by the load at the current moment, the temperature data of the temperature monitoring point at the current moment, the operating data of the temperature control execution device at the current moment, the environmental disturbance data at the current moment, the temperature influence data, and the historical execution data of the previous target control action sequence.

6. The method according to claim 1, characterized in that, The step of generating several candidate control action sequences based on the current state and the time-delay response diagram includes: The target temperature monitoring point is determined based on the current state; wherein the first difference between the actual temperature of the target temperature monitoring point and the preset temperature threshold is less than or equal to the preset difference threshold. Based on the current state of the data center, acquire the temperature impact data of each temperature control execution device on the target temperature monitoring point; Based on the first difference and the temperature impact data of each of the temperature control actuators on the target temperature monitoring point, the target temperature control actuator is determined from the temperature control actuators; The target temperature control actuators are sorted based on temperature influence data to obtain a first order; based on the first order, several candidate control action sequences with different control strategies are generated, each candidate control action sequence including at least two consecutive control steps, and each control step including a control operation on the target temperature control actuator and a corresponding preset execution time.

7. The method according to claim 1, characterized in that, The step of obtaining the target control action sequence from the candidate control action sequence based on the current state, the candidate control action sequence, and a preset value network includes: Based on the candidate control action sequence, obtain one or more safety control action sequences that satisfy preset safety constraints; The current state and the security control action sequence are input into the value network, and value estimation is performed on the security control action sequence to obtain the value estimate corresponding to the security control action sequence. The sequence of safety control actions with the highest estimated value is selected as the target control action sequence.

8. The method according to claim 1, characterized in that, Controlling the temperature control actuator to perform according to the target control action sequence includes: The temperature control execution device is controlled sequentially according to the control steps of the target control action sequence. After each control step is completed, it is checked whether the actual temperature of the temperature monitoring point meets the preset first condition and whether the temperature control execution device meets the preset second condition. If so, the currently executed control step is updated to the preset safety prefix data. Otherwise, cancel the unexecuted control steps in the target control action sequence and restore the control steps corresponding to the security prefix data, and re-execute the step of obtaining the target control action sequence from the candidate control action sequence based on the current state, the candidate control action sequence and the preset value network.

9. The method according to any one of claims 1-8, characterized in that, After the step of controlling the temperature control actuator according to the target control action sequence to complete the control of the temperature control actuator, the method further includes: Record the actual execution data performed by the temperature control device; Obtain the counterfactual control action sequence of the target control action sequence; Based on the counterfactual control action sequence, counterfactual state reasoning is performed on the data center to obtain counterfactual execution data; Obtain counterfactual attribution benefits based on the actual execution data and the counterfactual execution data; The network parameters of the value network are updated based on the counterfactual attribution benefits.

10. A control system for a computer room temperature control actuator, characterized in that, The system is applied in a data center, which includes several devices, including loads, temperature control actuators, and temperature monitoring points. The system includes: The time-delay response diagram acquisition module is used to acquire historical operating data in the data center and construct a time-delay response diagram based on the historical operating data. The time-delay response diagram includes temperature impact data of temperature control execution equipment on temperature monitoring points. The current operating data acquisition module is used to acquire the current operating data of the data center; The current status acquisition module is used to acquire the current status of the data center at the current moment based on the current operating data, the time-delay response diagram, and the historical operating data. The candidate control action sequence acquisition module is used to generate several candidate control action sequences based on the current state and the time-delay response diagram; The target control action sequence acquisition module is used to acquire the target control action sequence from the candidate control action sequence based on the current state, the candidate control action sequence, and a preset value network. The control module is used to control the temperature control actuator to perform according to the target control action sequence, thereby completing the control of the temperature control actuator.

Images