Water-cooled refrigeration station cooling system optimization control method based on deep reinforcement learning

Abstract

The application relates to the technical field of water-cooled data processing, in particular to a water-cooled refrigeration station cooling system optimization control method based on deep reinforcement learning, which comprises the following steps: after the cooling system sends a control instruction, according to the change trend and change amplitude of the environmental temperature, the synchronous execution degree of the cooling system to the control instruction is analyzed to obtain an instruction response strength value; according to the environmental temperature stability of a local time period before the control instruction is sent and the numerical difference between the instruction response strength values of all control processes, an anti-interference strength index is obtained; the convergence factor of the optimization algorithm in the current control process is adjusted according to the anti-interference strength index; and the adjusted optimization algorithm is used to optimize the control of the cooling system. The application can improve the precision of the cooling equipment control parameters calculated by using the optimization algorithm and realize the optimization control of the cooling system.

Description

Technical Field

[0001] This application relates to the field of water-cooled data processing technology, specifically to an optimization control method for water-cooled refrigeration station cooling systems based on deep reinforcement learning. Background Technology

[0002] Water-cooled refrigeration systems are core energy-consuming units used in large public buildings, commercial complexes, and industrial facilities. They consist of a closed-loop cooling water system formed by the coordinated operation of chillers, cooling towers, pumps, and piping networks. The system utilizes chillers to generate low-temperature chilled water, which is then pumped to air conditioning terminals to absorb indoor heat loads. Waste heat generated during chiller operation is discharged through the cooling water system to the cooling tower, thus achieving precise temperature control over large-scale spaces.

[0003] When optimizing the control of a water-cooled refrigeration station, a large amount of circulating water is used as the heat transfer medium in the system. Due to the water flow through the pipeline and the heat exchange process of the equipment, the changes in key state parameters such as water flow velocity and water temperature obtained by monitoring are lagging. This results in a significant lag and slow dynamic stabilization process in the response of performance indicators such as total system energy consumption obtained based on monitoring data, i.e., there is significant thermal inertia and time delay.

[0004] Among existing optimization algorithms, the Grey Wolf Algorithm, as a heuristic search algorithm, is often used for the optimization and control of water-cooled chiller systems due to its advantages such as few control parameters and ease of implementation. However, when directly using the original Grey Wolf Algorithm for online optimization, due to the aforementioned thermal inertia and time delay, the system energy consumption feedback value obtained by the algorithm in a short period of time is not the true value of stable operation under that set of parameters, but rather includes instantaneous errors in the dynamic process. This inaccurate fitness assessment will seriously mislead the wolf pack's search direction, making it difficult for the algorithm to converge to the true global optimum. Consequently, the calculation accuracy of the control parameters is low, failing to meet the system's control accuracy requirements. Summary of the Invention

[0005] To address the aforementioned technical problems, this application provides an optimized control method for a water-cooled refrigeration station cooling system based on deep reinforcement learning, thereby resolving the existing issues.

[0006] The optimization control method for the cooling system of a water-cooled refrigeration station based on deep reinforcement learning in this application adopts the following technical solution:

[0007] One embodiment of this application provides an optimization control method for a water-cooled chiller system based on deep reinforcement learning, the method comprising the following steps:

[0008] S1: Real-time acquisition of the input power and operating frequency of each cooling device in the cooling system of the water-cooled refrigeration station; real-time acquisition of the ambient temperature inside the water-cooled refrigeration station;

[0009] S2, After the cooling system issues a control command, it analyzes the degree of synchronous execution of the control command by the cooling system based on the trend and magnitude of the change in ambient temperature, and obtains the command response strength value.

[0010] S3. Calculate the environmental fluctuation value based on the stable ambient temperature during a local time period before the control command is issued; analyze the numerical differences between the command response strength values ​​of all control processes to obtain the response outlier value for each control process; combine the environmental fluctuation value before the current control command is issued with the response outlier value of its previous control process to obtain the anti-interference strength index.

[0011] S4, adjust the convergence factor of the optimization algorithm in the current control process according to the anti-interference strength index; use the adjusted optimization algorithm to optimize the control of all cooling devices according to the input power and operating frequency of all cooling devices.

[0012] Furthermore, the step of analyzing the degree of synchronous execution of control commands by the cooling system based on the changing trend and magnitude of ambient temperature, and obtaining the command response intensity value, includes:

[0013] For the t-th control command issued by the cooling system, the first moment after the t-th control command is issued is recorded as the response moment; the preset time window starting from the response moment is recorded as the response monitoring window; based on the cumulative instantaneous temperature change within the response monitoring window, the instantaneous temperature change accumulation value after the t-th control command is issued is obtained.

[0014] Obtain the command temperature from the control command; obtain the difference between the command temperature and the ambient temperature at the response time; record the difference as the adjustment temperature difference for the tth time; calculate the command response intensity value for the tth time based on the synchronization between the instantaneous temperature change accumulation value and the overall change value of the ambient temperature, and the synchronization between the overall change trend of the ambient temperature and the adjustment temperature difference for the tth time.

[0015] Furthermore, the method for obtaining the instantaneous temperature change accumulation value after the t-th control command is issued includes:

[0016] The sequence of all ambient temperatures within the response monitoring window is denoted as the response temperature sequence; the first-order difference sequence of the response temperature sequence is obtained; the sum of the absolute values ​​of all data in the first-order difference sequence is denoted as the instantaneous temperature change accumulation value after the t-th control command is issued.

[0017] Furthermore, the formula for calculating the intensity value of the t-th instruction response is as follows:

[0018]

[0019] in, This represents the strength value of the command response at time t. Represents a symbolic function. This indicates the temperature difference adjusted in the t-th iteration. , Let represent the nth and 1st elements in the response temperature sequence corresponding to the t-th control command, respectively, where n represents the length of the response temperature sequence. This represents the instantaneous accumulated temperature change after the t-th control command is issued; This is a preset constant greater than 0.

[0020] Furthermore, the method for calculating the environmental fluctuation value includes:

[0021] The sequence of all ambient temperatures within a preset time period before the control command is issued is recorded as the pre-control data; the pre-control data is subjected to first-order difference processing to obtain the difference data; the formula for calculating the environmental fluctuation value is: ;in, This represents environmental fluctuation values. It is the mean of the absolute values ​​of all elements in the differenced data; is the standard deviation of all elements in the differenced data.

[0022] Furthermore, the analysis examines the numerical differences between the command response strength values ​​of all control processes to obtain the outlier values ​​of each control process, including:

[0023] For the i-th control, cluster all instruction response intensity values ​​up to and including the i-th control to obtain a preset number of clusters; calculate the mean of all instruction response intensity values ​​within each cluster as the feature value of each cluster; denote the cluster with the largest feature value as the instruction execution cluster; and denote the absolute value of the difference between the instruction response intensity value of the i-th control and the feature value of the instruction execution cluster as the outlier value of the i-th control.

[0024] Furthermore, the method for calculating the anti-interference strength index is as follows:

[0025]

[0026] in, This represents the anti-interference strength index of the i-th control. The environmental fluctuation value for the i-th control; The outlier value in the response of the (i-1)th control; It is a linear normalization function.

[0027] Furthermore, the convergence factor of the optimization algorithm adjusted according to the anti-interference strength index in the current control process includes:

[0028] Calculate the difference between the value 1 and the anti-interference strength index of the current control, and record it as the first difference; obtain the original convergence factor in the Grey Wolf optimization algorithm; multiply the original convergence factor and the first difference as the convergence factor of the Grey Wolf optimization algorithm in the current control process.

[0029] Furthermore, the step of optimizing the control of all cooling devices using an adjusted optimization algorithm based on the input power and operating frequency of all cooling devices includes:

[0030] The Grey Wolf optimization algorithm is initialized based on the input power and operating frequency of all cooling devices; the convergence factor of the Grey Wolf optimization algorithm in the current control process is used for iterative calculation to obtain the optimal combination of control parameters for all cooling devices in the current control process;

[0031] The optimal control parameters of all cooling devices are combined and input into the execution unit to optimize the control of each cooling device.

[0032] Furthermore, the initialization of the gray wolf optimization algorithm includes: using a vector composed of the operating frequencies of all cooling devices as the position of the gray wolf in the gray wolf optimization algorithm; and using the sum of the input power of all cooling devices after normalization as the objective function in the gray wolf optimization algorithm.

[0033] This application has at least the following beneficial effects:

[0034] This application analyzes the trend of ambient temperature changes after a control command is issued, compares the expected temperature change direction of the control command with the actual temperature change direction, and then constructs a command response intensity value, which accurately reflects the degree of synchronous execution of the control command by the cooling system. Furthermore, it combines the command response intensity value representing the execution state during the response execution phase between two control commands, and the ambient temperature stability during the stable state phase, to construct an anti-interference intensity index to evaluate the stability of the cooling system and accurately reflect its current controllability. Furthermore, based on the anti-interference intensity index, it dynamically determines the convergence factor in the optimization algorithm, obtaining the optimal parameter combination adapted to the current complex environment. This solves the problem that traditional fixed-parameter optimization algorithms are difficult to adapt to and have low computational accuracy due to the inherent variable operating conditions of water-cooled refrigeration station cooling systems, thus achieving optimized control of the water-cooled refrigeration station cooling system. Attached Figure Description

[0035] To more clearly illustrate the technical solutions and advantages in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0036] Figure 1 A flowchart illustrating the steps of the optimized control method for a water-cooled refrigeration station cooling system based on deep reinforcement learning provided in this application;

[0037] Figure 2 This is a flowchart for obtaining the anti-interference strength index. Detailed Implementation

[0038] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. In the absence of conflict, the embodiments of this application and the technical features in the embodiments can be combined with each other. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0039] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0040] The following description, in conjunction with the accompanying drawings, details the specific scheme of the optimization control method for the cooling system of a water-cooled refrigeration station based on deep reinforcement learning provided in this application.

[0041] This application provides an embodiment of an optimization control method for a water-cooled chiller system based on deep reinforcement learning. Specifically, the following optimization control method for a water-cooled chiller system based on deep reinforcement learning is provided. Please refer to [link / reference]. Figure 1 The method includes the following steps:

[0042] Step S1: Real-time acquisition of the input power and operating frequency of each cooling device in the cooling system of the water-cooled refrigeration station; real-time acquisition of the ambient temperature inside the water-cooled refrigeration station.

[0043] One specific implementation scenario of this application is the optimization control scenario of the cooling system of a water-cooled refrigeration station during operation. When using a deep reinforcement learning algorithm to control an evaporative air conditioning system, directly using the original gray wolf optimization algorithm for online optimization requires evaluating the quality of the current solution in each iteration of its standard iterative optimization mechanism. Due to the aforementioned thermal inertia and time delay, inaccurate fitness evaluations can severely mislead the wolf pack's search direction, resulting in low computational accuracy. Therefore, to steadily improve the energy-saving control accuracy of the cooling system, this embodiment analyzes the temperature data change patterns and the synchronization degree between data changes and control commands monitored during the control process, and rationally sets the convergence factor in the optimization algorithm.

[0044] First, the input power and operating frequency of the cooling equipment are collected in real time via an electricity meter; specifically, the input power of the chiller, cooling water pump, and cooling tower fan; the operating frequency of the chiller, cooling water pump, and cooling tower fan is collected via a frequency converter; and the ambient temperature within the water-cooled refrigeration station is collected in real time via a temperature sensor. In this application, the data collection interval is 1 second, which the implementer can set according to actual conditions.

[0045] To ensure data quality and the accuracy of subsequent calculations, the data collected in the above steps needs to be preprocessed accordingly. Specifically: missing values ​​caused by network latency or sensor malfunction are filled in using linear interpolation to ensure data integrity; to eliminate random noise interference from the sensor itself, the collected data is smoothed using a moving average filter with a window size of 10; to eliminate the influence of different value ranges, the data of different dimensions is normalized using the max-min normalization method. The linear interpolation method, moving average filter, and max-min normalization method are well-known techniques, and their specific implementation processes will not be elaborated further.

[0046] Step S2: After the cooling system issues a control command, the degree of synchronous execution of the control command by the cooling system is analyzed based on the trend and magnitude of the change in ambient temperature, and the command response intensity value is obtained.

[0047] The unique thermal inertia and time delay characteristics of water-cooled refrigeration systems mean that when responding to new control commands, energy consumption fluctuations are not chaotic but rather manifest as a directional and significant transition process. For example, when a cooling system receives a command to increase cooling capacity, its temperature does not immediately jump but gradually approaches a new stable point along an upward trajectory. During this process, the trend and magnitude of the temperature drop reflect the strength of the transient response. In other words, a rapid response to a control command results in energy consumption changes that are not only large in magnitude but also uniform in direction and highly monotonic. Conversely, a weak response or a system nearing steady state exhibits small energy consumption changes, and the trend may be less pronounced due to minor disturbances.

[0048] Based on the above analysis, this application analyzes the trend and magnitude of ambient temperature change after the control command is issued during the operation of the cooling system, and constructs a command response intensity value to reflect the intensity of the transient response. The specific process is as follows:

[0049] Taking the t-th control command issued by the cooling system as an example, after the control command is issued, the execution unit of the cooling system will immediately respond and execute it, completing the execution within a certain period of time. Since the execution completion time is not fixed, this embodiment selects the trend and magnitude of ambient temperature change over a period of time to analyze the response; specifically, the first moment after the t-th control command is issued is recorded as the response moment; a preset time window starting from the response moment is recorded as the response monitoring window. In this embodiment, the preset time window contains n=60 moments. Based on the cumulative instantaneous temperature change within the response monitoring window, the instantaneous temperature change accumulation value after the t-th control command is issued is obtained.

[0050] Specifically, the sequence of ambient temperature data within the response monitoring window is denoted as the response temperature sequence. A first-order difference sequence is then performed on the response temperature sequence to obtain a first-order difference sequence. This first-order difference sequence represents the magnitude of the ambient temperature change within the response monitoring window. The sum of the absolute values ​​of all data in the first-order difference sequence is denoted as the instantaneous temperature change accumulation value after the t-th control command is issued. The larger the instantaneous temperature change accumulation value, the more significant the overall trend of change in the elements of the response temperature sequence within the corresponding response monitoring window, and the more likely it is a change caused by the execution command.

[0051] The control command issued by the control unit of the cooling system specifies the temperature to be controlled. The control command temperature is then retrieved; the difference between the control command temperature and the ambient temperature at the response time is also retrieved; this difference is recorded as the adjustment temperature difference for the t-th time.

[0052] Furthermore, based on the synchronization between the instantaneous temperature change accumulation value within the response monitoring window and the overall change value of the ambient temperature, and the synchronization between the overall trend of the ambient temperature change and the temperature difference adjusted for the tth time, the command response intensity value for the tth time is calculated. In this embodiment, the method for calculating the command response intensity value is as follows:

[0053]

[0054] in, This represents the strength value of the instruction response at the t-th time, and sgn() represents the sign function. This indicates the temperature difference adjusted in the t-th iteration. , Let represent the nth and 1st elements in the response temperature sequence corresponding to the t-th control command, respectively, where n represents the length of the response temperature sequence. This represents the instantaneous accumulated temperature change after the t-th control command is issued; This is a preset constant greater than 0, used to avoid the denominator being zero and thus unable to be calculated. In this application, the value is 1, but implementers can select it according to the actual situation.

[0055] If the cooling system's execution unit has a strong ability to execute the control command issued for the tth time, then in the corresponding response monitoring window, the overall trend of the elements in the response temperature sequence should be relatively significant, and the trend should conform to the trend of the temperature difference adjustment. For example, when the ambient temperature is continuously decreasing, the numerator term... With denominator They are quite close, and the sign of the numerator is similar to... The same sign corresponds to a larger calculated command response strength value; if the execution capability of the issued control command is weak, the overall trend of the response temperature sequence in the response monitoring window is less significant, and the corresponding calculated command response strength value is smaller, fluctuating around 0; if the execution direction of the issued control command is opposite, such as issuing a cooling command but actually a heating process, the sign of the numerator is the same as the previous value. The opposite sign makes the calculated command response strength value much less than 0.

[0056] Step S3: Calculate the environmental fluctuation value based on the stable ambient temperature during the local time period before the control command is issued; analyze the numerical differences between the command response strength values ​​of all control processes to obtain the response outlier value of each control process; combine the environmental fluctuation value before the current control command is issued and the response outlier value of its previous control process to obtain the anti-interference strength index.

[0057] The command response strength value obtained through the above steps can intuitively reflect the cooling system's ability to execute control commands from one stable state to another after receiving them. During the operation of a water-cooled refrigeration station, it is necessary to maintain stable ambient temperature before receiving the next control command, meaning strong resistance to environmental changes is crucial. However, the environment is constantly changing; factors such as people entering and exiting the room, opening and closing doors and windows, and changes in outdoor sunlight intensity can all disturb the indoor temperature, causing changes in the system's stable state and consequently affecting the command response strength value during the adjustment process. When using optimization algorithms for temperature control, the interference of environmental changes on temperature control should be fully considered, and the interference experienced by the control process should be analyzed to improve the anti-interference capability in subsequent control operations.

[0058] Based on the above analysis, during the execution of control commands, the ambient temperature should change towards the commanded temperature. Once the commanded temperature is reached, the control command is no longer executed; instead, the temperature remains constant, maintaining a new stable state. There are two phases between the two control command occurrences: the response execution phase and the state stabilization phase. If the ambient temperature fluctuates during the state stabilization phase, it indicates that the fluctuation is caused by environmental instability. Figure 2 As shown, the anti-interference strength index is calculated based on the analysis results of the data characteristics of the two stages. First, the environmental fluctuation value is calculated based on the ambient temperature change during the stable state stage. Specifically, when the control command is issued, the environmental fluctuation value is calculated based on the ambient temperature stability during the local time period prior to the issuance of the control command.

[0059] In this embodiment, the sequence of all ambient temperatures within a preset time period before the control command is issued is recorded as the pre-control data; wherein, the preset time period is 30 seconds before the control command is issued.

[0060] The pre-control data is processed using first-order differencing to obtain differencing data; the formula for calculating environmental fluctuation values ​​is: ;in, This represents environmental fluctuation values. It is the mean of the absolute values ​​of all elements in the differenced data; is the standard deviation of all elements in the differenced data.

[0061] Differential data represents the magnitude of change in the data before control over time. Under stable environmental conditions, this value should decrease gradually until it stabilizes at 0. and It should get smaller and smaller; if and The larger the sum of the values, the more it indicates temperature changes caused by the environment, and the greater the environmental fluctuation value.

[0062] Furthermore, for the i-th control, the outlier value of the i-th control process is obtained based on the numerical difference between the i-th command response strength value and the command response strength values ​​of all previous control processes.

[0063] Specifically, all instruction response intensity values ​​up to and including the i-th instruction are clustered to obtain a predetermined number of clusters; in this embodiment, the K-means clustering algorithm is used, and the number of clusters K is set to 2. The mean of all instruction response intensity values ​​within each cluster is calculated and used as the feature value of each cluster. The cluster with the largest feature value is denoted as the instruction execution cluster, and the cluster with the smallest feature value is denoted as the reverse execution cluster. The absolute value of the difference between the instruction response intensity value of the i-th control and the feature value of the instruction execution cluster is denoted as the outlier value of the i-th control response.

[0064] The outlier in the response reflects the degree of difference between the command response strength of the i-th control process and other control processes. If there is no environmental influence, the command response strength will be at a similarly high level in each control process without interference. Therefore, the larger the outlier in the response, the greater the interference intensity during the execution of the command.

[0065] By comprehensively analyzing the outlier values ​​of the response during the execution phase, which reflect the intensity of interference encountered during instruction execution, and the environmental fluctuation values ​​during the stable phase, which reflect temperature stability, an anti-interference strength index is constructed. The specific calculation method is as follows:

[0066]

[0067] in, This represents the anti-interference strength index of the i-th control. The environmental fluctuation value for the i-th control; The outlier value in the response of the (i-1)th control; It is a linear normalization function.

[0068] During the period following the issuance of a control command, the system is in the command execution phase and belongs to the command execution cluster. Once the command execution is complete, a new stable state is reached. If the anti-interference capability is strong, the command response strength value during the command execution phase should belong to the command execution cluster. That is, the smaller the response outlier, the larger the anti-interference strength index for the next control. If the anti-interference capability is poor, the ambient temperature will be unstable and the environmental fluctuation value will be large before the next control command is issued, resulting in a smaller anti-interference strength index for the next control.

[0069] Step S4: Adjust the convergence factor of the optimization algorithm in the current control process according to the anti-interference strength index; and use the adjusted optimization algorithm to optimize the control of all cooling devices according to the input power and operating frequency of all cooling devices.

[0070] When using the Grey Wolf optimization algorithm to optimize the cooling system of a water-cooled refrigeration station, a large anti-interference index indicates that the cooling system is in a stable state with strong anti-interference capability, and the current control parameters are very close to the optimal solution. In this case, the convergence speed should be increased during the optimization process to converge to the optimal solution as quickly as possible. Conversely, a small anti-interference index indicates that the cooling system is in an unstable state, such as when the indoor environment is open (e.g., doors or windows are open and not closed), and is greatly affected by the external environment. This means that the current combination of control parameters cannot cope with the instability of the current environment. In this case, the global search capability needs to be improved, i.e., the convergence speed should be reduced to find a more stable optimal solution.

[0071] Based on the above analysis, this application improves the convergence factor in the gray wolf optimization algorithm based on the anti-interference strength index. The calculation method is as follows: ;in This represents the convergence factor of the Grey Wolf optimization algorithm in the i-th control process. This represents the original convergence factor in the Grey Wolf optimization algorithm. This represents the anti-interference strength index of the i-th control.

[0072] In the process of using the Grey Wolf optimization algorithm to perform real-time optimization control of the cooling system of a water-cooled refrigeration station, if the cooling system has strong anti-interference ability, the control parameters are close to the optimal solution, and the convergence rate should be increased to make the calculated convergence factor smaller. If the cooling system is in an unstable state and the anti-interference strength index is small, the control parameters cannot cope with the complex environment. In this case, the convergence rate needs to be reduced and the global search capability needs to be improved to find the optimal combination of control parameters that can cope with the current situation, so that the calculated convergence factor is larger.

[0073] Furthermore, by obtaining the convergence factor for optimizing the cooling system of the water-cooled refrigeration station through the above steps, the Grey Wolf optimization algorithm can be used to perform real-time optimization control of the cooling system. The specific process is as follows:

[0074] The relevant parameters and values ​​in the Grey Wolf optimization algorithm are as follows: the population size is preset to 30, the maximum number of iterations is preset to 100, and the initial value of the convergence factor is 2. Implementers can set these parameters according to their actual needs.

[0075] The position of each gray wolf in the population is randomly initialized, where the wolf's position is a vector composed of monitoring parameters, specifically the operating frequencies of the chilled water pump, cooling water pump, and cooling tower fan. The objective function is the sum of the normalized input power of the chiller unit, cooling water pump, and cooling tower fan. During the iterative optimization process, the convergence factor at each iteration is calculated through the above steps, ensuring that the optimized combination of control parameters adapts to the current environment and guarantees the stable operation of the water-cooled chiller station. The iterative process of the gray wolf optimization algorithm is a well-known technique and will not be elaborated here.

[0076] After obtaining the optimal combination of control parameters adapted to the current environment through the above steps, it is input into the central controller of the execution unit. The central controller controls the relevant equipment in the cooling system to achieve real-time optimized control of the water-cooled refrigeration station cooling system.

[0077] Through the above description of the embodiments in conjunction with the accompanying drawings, those skilled in the art will understand that, for the sake of convenience and brevity, the above division of functional modules is only used as an example. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0078] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for optimizing and controlling a water-cooled refrigeration station cooling system based on deep reinforcement learning, characterized in that, The method includes the following steps: S1: Real-time acquisition of the input power and operating frequency of each cooling device in the cooling system of the water-cooled refrigeration station; real-time acquisition of the ambient temperature inside the water-cooled refrigeration station; S2, After the cooling system issues a control command, it analyzes the degree of synchronous execution of the control command by the cooling system based on the trend and magnitude of the change in ambient temperature, and obtains the command response strength value. For the t-th control command issued by the cooling system, the first moment after the t-th control command is issued is recorded as the response moment; the preset time window starting from the response moment is recorded as the response monitoring window; based on the cumulative instantaneous temperature change within the response monitoring window, the instantaneous temperature change accumulation value after the t-th control command is obtained. Obtain the command temperature from the control command; obtain the difference between the command temperature and the ambient temperature at the response time; record the difference as the adjustment temperature difference at the t-th time; the formula for calculating the command response intensity value is: in, This represents the strength value of the command response at time t. Represents a symbolic function. This indicates the temperature difference adjusted in the t-th iteration. , Let represent the nth and 1st elements in the response temperature sequence corresponding to the tth control command, respectively, where n represents the length of the response temperature sequence. This represents the instantaneous accumulated temperature change after the t-th control command is issued; This is a preset constant greater than 0; S3. Calculate the environmental fluctuation value based on the stable ambient temperature during a local time period before the control command is issued. The specific calculation process for the environmental fluctuation value is as follows: perform differential processing on the ambient temperature during the local time period before the control command is issued, and calculate the environmental fluctuation value based on the data dispersion of the differential results; analyze the numerical differences between the command response intensity values ​​of all control processes to obtain the response outlier value for each control process; the calculation process for the response outlier value is as follows: cluster the command response intensity values ​​of all control processes, divide the execution command clusters based on the clustering results, and calculate the response outlier value based on the difference between the data in the execution command cluster and the command response intensity; combine the environmental fluctuation value before the current control command is issued and the response outlier value of its previous control process to obtain the anti-interference strength index. S4, adjust the convergence factor of the optimization algorithm in the current control process according to the anti-interference strength index; use the adjusted optimization algorithm to optimize the control of all cooling devices according to the input power and operating frequency of all cooling devices.

2. The optimization control method for a water-cooled refrigeration station cooling system based on deep reinforcement learning as described in claim 1, characterized in that, The method for obtaining the instantaneous temperature change accumulation value after the t-th control command is issued includes: The sequence of all ambient temperatures within the response monitoring window is denoted as the response temperature sequence; the first-order difference sequence of the response temperature sequence is obtained; the sum of the absolute values ​​of all data in the first-order difference sequence is denoted as the instantaneous temperature change accumulation value after the t-th control command is issued.

3. The optimization control method for a water-cooled refrigeration station cooling system based on deep reinforcement learning as described in claim 1, characterized in that, The method for calculating the environmental fluctuation value includes: The sequence of all ambient temperatures within a preset time period before the control command is issued is recorded as the pre-control data; the pre-control data is subjected to first-order difference processing to obtain the difference data; the formula for calculating the environmental fluctuation value is: ;in, This represents environmental fluctuation values. It is the mean of the absolute values ​​of all elements in the differenced data; is the standard deviation of all elements in the differenced data.

4. The optimization control method for a water-cooled refrigeration station cooling system based on deep reinforcement learning as described in claim 1, characterized in that, The analysis examines the numerical differences between the command response strength values ​​of all control processes to obtain the outlier values ​​of each control process, including: For the i-th control, cluster all instruction response intensity values ​​up to and including the i-th control to obtain a preset number of clusters; calculate the mean of all instruction response intensity values ​​within each cluster as the feature value of each cluster; denote the cluster with the largest feature value as the instruction execution cluster; denote the absolute value of the difference between the instruction response intensity value of the i-th control and the feature value of the instruction execution cluster as the outlier value of the i-th control.

5. The optimization control method for a water-cooled refrigeration station cooling system based on deep reinforcement learning as described in claim 1, characterized in that, The method for calculating the anti-interference strength index is as follows: in, This represents the anti-interference strength index of the i-th control. The environmental fluctuation value for the i-th control; The outlier value in the response of the (i-1)th control; It is a linear normalization function.

6. The optimization control method for a water-cooled refrigeration station cooling system based on deep reinforcement learning as described in claim 1, characterized in that, The convergence factor of the optimization algorithm adjusted according to the anti-interference strength index in the current control process includes: Calculate the difference between the value 1 and the anti-interference strength index of the current control, and record it as the first difference; obtain the original convergence factor in the Grey Wolf optimization algorithm; multiply the original convergence factor and the first difference as the convergence factor of the Grey Wolf optimization algorithm in the current control process.

7. The optimized control method for a water-cooled refrigeration station cooling system based on deep reinforcement learning as described in claim 6, characterized in that, The step involves optimizing the control of all cooling devices using an adjusted optimization algorithm based on their input power and operating frequency, including: The Grey Wolf optimization algorithm is initialized based on the input power and operating frequency of all cooling devices; the convergence factor of the Grey Wolf optimization algorithm in the current control process is used for iterative calculation to obtain the optimal combination of control parameters for all cooling devices in the current control process; The optimal control parameters of all cooling devices are combined and input into the execution unit to optimize the control of each cooling device.

8. The optimized control method for a water-cooled refrigeration station cooling system based on deep reinforcement learning as described in claim 7, characterized in that, The initialization of the gray wolf optimization algorithm includes: using a vector composed of the operating frequencies of all cooling devices as the position of the gray wolf in the gray wolf optimization algorithm; and using the sum of the input power of all cooling devices after normalization as the objective function in the gray wolf optimization algorithm.

Images