An energy-saving optimization control method for evaporative cooling air conditioning based on deep reinforcement learning

Abstract

The present application relates to the technical field of air conditioning energy-saving control, and specifically to a method for optimizing energy-saving control of an evaporative cooling air conditioner based on deep reinforcement learning. The method comprises: calculating a heat load rise index based on the changing trend and amplitude of the ambient temperature in a local time period at each moment during the operation of the evaporative cooling air conditioner; obtaining an energy consumption peak synchronization index based on the similarity of the peak fluctuations in the fan, water pump, and evaporator power in the local time period at each moment; calculating environmental complexity based on the difference between the heat load rise index and the energy consumption peak synchronization index at each moment; setting a reward function in a deep reinforcement learning algorithm based on the ambient temperature and environmental complexity at each moment; and optimizing the fan power of the evaporative cooling air conditioner using the deep reinforcement learning algorithm based on the current ambient temperature, water pump, and evaporator power. The present application can improve the accuracy of air conditioning energy-saving control using a deep reinforcement learning algorithm.

Description

Technical Field

[0001] The present application relates to the technical field of air-conditioning energy-saving control, and specifically to an evaporative cooling air-conditioning energy-saving optimization control method based on deep reinforcement learning. Background Art

[0002] Evaporative cooling is a highly efficient natural cooling technology that uses water as the cooling medium and utilizes the difference between the dry-bulb and wet-bulb temperatures in the air as a driving force. This cooling technology absorbs the air's latent heat of vaporization as the water evaporates, thereby cooling the air. Evaporative cooling systems can provide fresh air to buildings and, using water as the cooling medium, reduce carbon dioxide emissions. Traditional evaporative cooling control methods rely primarily on a single control strategy. These strategies are insensitive to ambient temperature changes and often exhibit lag when facing dynamically changing heat loads.

[0003] Currently, deep reinforcement learning algorithms are commonly used to optimize cooling tower fan power through continuous learning and optimization of control strategies, achieving significant energy savings and possessing high adaptability. However, during the operation of evaporative cooling air conditioners, collecting operational data from fans, water pumps, and evaporators is difficult, and the accuracy of the data collection is difficult to control. As a result, the performance of the reward function in deep reinforcement learning algorithms is limited by the accuracy of the input parameters, causing the algorithm to learn incorrect control modes, affecting the energy-saving control accuracy of the air conditioner. Summary of the Invention

[0004] In order to solve the above technical problems, the present application provides an evaporative cooling air conditioning energy-saving optimization control method based on deep reinforcement learning to solve the existing problems.

[0005] The present application discloses an energy-saving optimization control method for evaporative cooling air conditioning based on deep reinforcement learning, which adopts the following technical solutions:

[0006] An embodiment of the present application provides an energy-saving optimization control method for an evaporative cooling air conditioner based on deep reinforcement learning, the method comprising the following steps:

[0007] S1, obtain the ambient temperature and the power of the fan, water pump, and evaporator at each moment during the operation of the evaporative cooling air conditioner;

[0008] S2, calculate the heat load rise index based on the change trend and change range of the ambient temperature in the local period at each moment;

[0009] S3, according to the similarity of the peak fluctuations of the fan, water pump and evaporator power in the local period at each moment, obtain the energy consumption peak synchronization index at each moment;

[0010] S4, calculates the environmental complexity based on the difference between the heat load rise index and the energy consumption peak synchronization index at each moment; comprehensively considers the ambient temperature and environmental complexity at each moment to set the reward function in the deep reinforcement learning algorithm;

[0011] S5, using the deep reinforcement learning algorithm to optimize and control the fan power of the evaporative cooling air conditioner based on the current ambient temperature, the power of the water pump, and the evaporator.

[0012] Furthermore, the heat load rise index is calculated based on the change trend and change range of the ambient temperature in the local time period at each moment, including:

[0013] At time t, the sequence of ambient temperatures at a preset number of moments before time t is recorded as the neighboring temperature monitoring sequence at time t;

[0014] The existence of an upward trend or downward trend in the sequence is used as the hypothesis test condition in the Wilcoxon signed rank test algorithm. The Wilcoxon signed rank test algorithm is used to perform a statistical test on the neighboring temperature monitoring sequence at time t, and the p-value of the existence of an upward trend or downward trend in the neighboring temperature monitoring sequence at time t is obtained;

[0015] The heat load rising index at time t is calculated by comprehensively considering the p-values ​​of the upward and downward trends and the degree of change of the neighboring temperature monitoring series at time t.

[0016] Furthermore, the calculation formula of the heat load rise index at time t is:

[0017]

[0018] in, represents the heat load rising index at time t, 、 The p-values ​​represent whether the neighboring temperature monitoring series at time t has an upward trend or a downward trend respectively; represents the sum of all positive numbers in the first-order difference temperature series at time t; Represents the absolute value of the sum of all negative numbers in the first-order difference temperature series at time t; 1 is a preset constant greater than 0.

[0019] Furthermore, the method for obtaining the energy consumption peak synchronization index at each moment includes:

[0020] At time t, the sequence of wind turbine power at time t and within the preset period before it is subjected to first-order difference processing to obtain the wind turbine power difference sequence at time t; all maximum values ​​in the wind turbine power difference sequence are arranged in chronological order to obtain the wind turbine power peak sequence at time t;

[0021] Obtain the water pump power peak sequence and the evaporator power peak sequence at time t; record the set consisting of the fan power peak sequence, the water pump power peak sequence, and the evaporator power peak sequence at time t as the feature set at time t;

[0022] For any two sequences in the feature set, obtaining the peak similarity between the two sequences based on the similarity measurement between the two sequences;

[0023] The sum of the peak similarities between all pairs of sequences in the feature set is calculated as the energy consumption peak synchronization index at time t.

[0024] Furthermore, the method for obtaining the peak similarity between the two sequences includes: calculating the dynamic time warping distance between the two sequences; and using the inverse normalized value of the dynamic time warping distance as the peak similarity between the two sequences.

[0025] Furthermore, the environmental complexity is calculated by calculating the absolute value of the difference between the normalized value of the heat load rise index and the normalized value of the energy consumption peak synchronization index at each moment as the environmental complexity at each moment.

[0026] Furthermore, the reward function in the deep reinforcement learning algorithm is set as follows:

[0027]

[0028] in, represents the complexity of the environment at time t; represents the ambient temperature at time t; Indicates the preset comfortable temperature; Represents the reward function in the deep reinforcement learning algorithm at time t; exp() is an exponential function with a natural constant as the base.

[0029] Furthermore, the deep reinforcement learning algorithm is used to optimize the fan power of the evaporative cooling air conditioner according to the current ambient temperature, the power of the water pump, and the evaporator, including:

[0030] In the deep reinforcement learning algorithm, the operating range of the evaporative cooling air conditioner fan power is divided into a first preset number of intervals, and each interval is used as an action space parameter group. The ambient temperature, water pump, and evaporator power are used as state space parameters. The state space parameters within the first preset period before the current moment are clustered to obtain a second preset number of state space parameter groups.

[0031] The action space parameter group, state space parameter group, and reward function of the wind turbine power at each moment in the first preset period before the current moment are trained using the Q-Learning reinforcement learning algorithm to obtain a trained Q function.

[0032] The state space parameter data of the evaporative cooling air conditioner at the current moment is input into the Q function to obtain the optimal fan power at the current moment, thereby realizing energy-saving optimization control of the evaporative cooling air conditioner.

[0033] Furthermore, the state space parameter data of the evaporative cooling air conditioner at the current moment is input into the Q function to obtain the optimal fan power at the current moment, thereby realizing energy-saving optimization control of the evaporative cooling air conditioner, including:

[0034] The state space parameters at the current moment are classified into the state space parameter group. The action space parameter group corresponding to the maximum reward function value in the state space parameter group to which the state space parameters at the current moment belong is used as the execution action group. The middle value of the execution action group is used as the optimal fan power of the evaporative cooling air conditioning cooling tower.

[0035] Furthermore, the method for classifying the current state space parameters into the state space parameter groups is: using the KNN classification algorithm to classify the current state space parameters and the clustered state space parameter groups to obtain the state space parameter group to which the current state space parameters belong.

[0036] This application has at least the following beneficial effects:

[0037] This application constructs a heat load rise index by analyzing the trend change characteristics of ambient temperature, evaluates the relationship between ambient temperature and heat load, and facilitates the optimization of the operation strategy of evaporative cooling air conditioners to achieve energy-saving goals; analyzes the similarity of power peak changes between key refrigeration components of evaporative cooling air conditioners to construct an energy consumption peak synchronization index, which is convenient for analyzing the overall cooling effect of the refrigeration system; further, in order to improve the accuracy of energy-saving control, analyzes the reflection of the difference between the heat load rise index and the energy consumption peak on environmental changes, adjusts the reward function in the deep reinforcement learning algorithm, and adjusts the reward function in the deep reinforcement learning algorithm. The obtained reward function takes into account the environmental comfort and energy-saving effect, so that the improved deep reinforcement learning algorithm can save the power consumption of evaporative cooling air conditioners while meeting the comfort level, realize energy-saving optimization control of evaporative cooling air conditioners, and improve the accuracy of energy-saving control of air conditioners using deep reinforcement learning algorithms. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present application or the prior art, the following is a brief introduction to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0039] Figure 1 A flowchart of the steps of an evaporative cooling air conditioning energy-saving optimization control method based on deep reinforcement learning provided in this application;

[0040] Figure 2 This is a flow chart for energy-saving control of evaporative cooling air conditioning using the Q-Learning reinforcement learning algorithm. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, 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 in this application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0042] Unless defined otherwise, 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 belongs.

[0043] The following describes in detail a specific scheme of an evaporative cooling air conditioning energy-saving optimization control method based on deep reinforcement learning provided by this application with reference to the accompanying drawings.

[0044] An embodiment of the present application provides an evaporative cooling air conditioner energy-saving optimization control method based on deep reinforcement learning. Specifically, the following evaporative cooling air conditioner energy-saving optimization control method based on deep reinforcement learning is provided. Figure 1 , the method comprises the following steps:

[0045] Step S1, obtaining the ambient temperature and the power of the fan, water pump, and evaporator at each moment during the operation of the evaporative cooling air conditioner.

[0046] One specific implementation scenario for the embodiments of this application is optimizing energy-saving control during the operation of an evaporative cooling air conditioner. When using a deep reinforcement learning algorithm to control an evaporative cooling air conditioner, the performance of the reward function in the algorithm is limited by the measurement accuracy of operating status data collected from the fan, water pump, and evaporator. Therefore, to consistently improve the energy-saving control accuracy of the air conditioner, this embodiment sets a reward function by analyzing ambient temperature changes and the interactions between different components.

[0047] First, every x seconds, the temperature sensor collects the ambient temperature of the evaporative cooling air conditioner's environment, and the power sensor collects the operating power of the fan, water pump, and evaporator. In this embodiment, x = 1, but the user can set it based on actual conditions.

[0048] During the data collection process, the collected data may contain anomalies such as missing values ​​due to factors such as environmental interference. To prevent missing values ​​from affecting the processing of subsequent steps, this application uses a regression filling method to fill in the missing values. At the same time, to prevent the dimensional differences of different data from affecting the data calculation results, this application uses a maximum and minimum normalization method to normalize the filled data. The regression filling method and the maximum and minimum normalization method are well-known technologies, and the specific implementation process will not be repeated here.

[0049] Step S2, calculating the heat load rise index according to the change trend and change range of the ambient temperature in the local time period at each moment.

[0050] During evaporative cooling operation, as ambient temperature rises, the heat load increases, forcing the evaporative cooling system to generate more cooling capacity to maintain a cool and comfortable indoor temperature. As ambient temperature drops, the evaporative cooling system must reduce power to conserve energy. Therefore, the trend of heat load changes caused by ambient temperature plays a proactive role in energy-saving control of evaporative cooling systems.

[0051] In this embodiment, for moment t, the sequence composed of the ambient temperatures at n moments before moment t is recorded as the neighbor temperature monitoring sequence at moment t, and the sequence length is n. In this embodiment, n is 20. The neighbor temperature monitoring sequence at moment t is used as the input of the Wilcoxon signed rank test algorithm. Assume that test condition 1 is that the sequence has an upward trend. The output of the algorithm is the p-value of the sequence that has an upward trend. The size of the p-value reflects the reliability of the existence of an upward trend in the neighbor temperature monitoring sequence. Assume that test condition 2 is changed to that the neighbor temperature monitoring sequence has a downward trend, and perform the Wilcoxon signed rank test again. The output result is the p-value of the sequence that has a downward trend. The size of the p-value reflects the reliability of the existence of a downward trend in the neighbor temperature monitoring sequence. Among them, the Wilcoxon signed rank test algorithm calculates the p-value of the sequence change trend based on the difference between adjacent points. The specific calculation process is a well-known technology and will not be repeated here.

[0052] Furthermore, the heat load rising index at time t is calculated by integrating the p-value and degree of change of the neighboring temperature monitoring series at time t, which shows an upward trend or a downward trend. .

[0053] In this embodiment, the first-order difference processing is performed on the neighboring temperature monitoring sequence at time t to obtain the first-order difference temperature sequence at time t. The calculation formula of the heat load rise index is:

[0054]

[0055] in, represents the heat load rising index at time t, 、 The p-values ​​represent whether the neighboring temperature monitoring series at time t has an upward trend or a downward trend respectively; represents the sum of all positive numbers in the first-order difference temperature series at time t; represents the absolute value of the sum of all negative numbers in the first-order difference temperature sequence at time t; 1 is a constant greater than 0 preset in this embodiment, which is used to prevent the numerator or denominator from being zero and affecting the calculation result.

[0056] If the degree of the upward trend of the neighboring temperature monitoring series is more obvious, that is, The bigger, The smaller the time, The bigger; at this time, The bigger, The smaller it is, the greater the increase in ambient temperature, the higher the heat load intensity in the environment, and the more the evaporative air conditioner needs to increase the cooling power to lower the ambient temperature; conversely, the evaporative air conditioner should reduce the cooling power to achieve energy saving.

[0057] By analyzing the changing relationship between ambient temperature and heat load, precise control of evaporative cooling air conditioning can be achieved, thereby minimizing energy consumption while meeting indoor comfort.

[0058] Step S3, obtaining the energy consumption peak synchronization index at each moment according to the similarity of the peak fluctuations of the fan, water pump, and evaporator power in the local time period at each moment.

[0059] Evaporative cooling air conditioning systems consist of multiple key components, including cooling tower fans, cooling water pumps, evaporators, and refrigeration units. These components are interdependent and mutually influential during operation, collectively determining the system's overall energy consumption and cooling performance. In other words, the power consumption of each component is affected not only by its own operating state but also by the operating conditions of other components. The speed of the cooling tower fan directly affects the temperature of the cooling water, which in turn affects the cooling efficiency of the evaporator. The flow rate of the cooling water pump determines the cooling water circulation rate and plays a key role in the heat exchange capacity of the entire system. Therefore, performing similarity analysis on the characteristics of the fan, water pump, and evaporator power data can quantify the degree of mutual influence between components.

[0060] In this embodiment, the sequence of wind turbine power at time t and within a preset period N before it is processed by first-order difference to obtain the wind turbine power difference sequence at time t; in this embodiment, N=1min. All the maximum values ​​in the wind turbine power difference sequence are arranged in chronological order to obtain the wind turbine power peak sequence at time t. Similarly, for the water pump and evaporator power at time t and within 1 minute, the water pump power peak sequence is obtained according to the method of obtaining the fan power peak sequence. , evaporator power peak sequence The sequence obtained after the first-order difference represents the variation characteristics of the original data, and the peak values ​​are extracted as eigenvalues ​​to form the fan power peak sequence, water pump power peak sequence, and evaporator power peak sequence that reflect the variation characteristics of the data.

[0061] Furthermore, the set consisting of the fan power peak sequence, the water pump power peak sequence, and the evaporator power peak sequence at time t is recorded as the feature set at time t; for any two sequences in the feature set, the peak similarity between the two sequences is obtained based on the similarity measurement between the two sequences.

[0062] In this embodiment, the dynamic time warping distance between two sequences is calculated; the reverse normalized value of the dynamic time warping distance is used as the peak similarity between the two sequences. The reverse normalization method is: ;in, is the peak similarity between the i-th and j-th sequences; is the dynamic time warping distance between the i-th and j-th sequences. Calculation of the DTW dynamic time warping distance between two sequences of unequal lengths requires processing the two sequences. The specific method is a well-known technique of the DTW algorithm and will not be described in detail.

[0063] During the operation of evaporative cooling air conditioners, if the maximum points of power changes between key components are more similar, the DTW distance between sequences in the feature set is smaller, and the calculated peak similarity is greater.

[0064] The sum of the peak similarities between all pairwise sequences in the feature set is calculated as the energy consumption peak synchronization index at time t. The greater the peak similarity between the key components, the better the overall cooling effect of the system, and thus the calculated energy consumption peak synchronization index is larger.

[0065] Step S4: Calculate the environmental complexity based on the difference between the heat load rise index and the energy consumption peak synchronization index at each moment; and set the reward function in the deep reinforcement learning algorithm based on the ambient temperature and environmental complexity at each moment.

[0066] During normal operation of an evaporative air conditioner, the ambient temperature should remain stable and comfortable. However, if the ambient temperature begins to rise sharply and fluctuate significantly due to factors such as sunlight exposure or the movement of people inside the room, the heat load rise index calculated in step S2 will be large. At this time, the relevant refrigeration components of the evaporative air conditioner have not yet adjusted to this change, and the peak energy consumption synchronization index remains stable. To enable the relevant refrigeration components to respond to environmental changes as quickly as possible, this embodiment considers the difference between the heat load rise index and the peak energy consumption synchronization index when adjusting the reward function in the deep reinforcement learning algorithm.

[0067] Specifically, the absolute value of the difference between the normalized value of the heat load rise index and the normalized value of the energy consumption peak synchronization index at each moment is calculated as the environmental complexity at each moment. The normalization method is the Z-Score normalization method. In other embodiments of this application, the Min-Max normalization method may also be used, which is not limited in this application. The smaller the difference between the normalized heat load rise index and the energy consumption peak synchronization index, the more stable the environment, and the output power of the refrigeration components can be reduced to achieve energy-saving effects.

[0068] To ensure the cooling effect of evaporative cooling during energy-saving control, this embodiment constructs a reward function based on the difference between ambient temperature and perceived comfort temperature, combined with environmental complexity. Specifically, the reward function in the deep reinforcement learning algorithm is set based on the ambient temperature and environmental complexity at each moment.

[0069] In this embodiment, the Q-Learning reinforcement learning algorithm is used. The reward function in the Q-Learning reinforcement learning algorithm is calculated as follows:

[0070]

[0071] in, represents the complexity of the environment at time t; represents the ambient temperature at time t; Indicates the preset comfortable temperature, which is 26°C in this embodiment; Represents the reward function in the deep reinforcement learning algorithm at time t; exp() is an exponential function with the natural constant e as the base.

[0072] When the ambient temperature rises, if the operating power of the evaporative cooling air conditioner's cooling components increases accordingly, it indicates that the evaporative cooling air conditioner has achieved its cooling target and has not experienced abnormal energy consumption. When the ambient temperature drops, excessive cooling is no longer required. If the operating power of the evaporative cooling air conditioner's cooling components decreases accordingly, it indicates that the evaporative cooling air conditioner is in energy-saving mode. The less complex the environment and the closer the temperature is to the most comfortable temperature, the higher the environmental comfort level and the better the evaporative cooling air conditioner's performance, thus increasing the calculated reward function value.

[0073] Step S5: Based on the current ambient temperature, the power of the water pump and the evaporator, the deep reinforcement learning algorithm is used to optimize the control of the fan power of the evaporative cooling air conditioner.

[0074] The reward function in the Q-Learning reinforcement learning algorithm, set up through the above steps, is dynamic and takes into account the changing environment and operating conditions of the evaporative cooler's components. Therefore, when using the Q-Learning reinforcement learning algorithm to control the evaporative cooler's fan power based on ambient temperature, water pump, and evaporator power, it can promptly respond to environmental changes, optimizing the evaporative cooler's energy-saving control while maintaining a comfortable ambient temperature, thereby improving control accuracy.

[0075] Specifically, the Q-Learning reinforcement learning algorithm is used to perform energy-saving control on evaporative cooling air conditioners. Figure 2 As shown, it is divided into the following steps:

[0076] Based on the normal operating range of cooling tower fan power, the fan is divided into M intervals, each of which is considered an action space parameter group. In this embodiment, M is a first preset number, which is set to 5. The implementer can select the value based on actual conditions. For example, the cooling tower fan power of a small evaporative air conditioner is typically between 0.37 kW and 5.5 kW, which is divided into 5 action space parameter groups.

[0077] The ambient temperature, water pump, and evaporator power are used as state-space parameters in the algorithm and discretized. The specific steps are as follows: The state-space parameters of the evaporative cooling air conditioner for the period T minutes before the current moment, where T = 30, can be selected by the implementer based on actual circumstances. The state-space parameters within this period are used as input to a K-means clustering algorithm. The number of clusters, K, is set to 5 in this application and can be selected based on actual circumstances. The algorithm outputs the clustered state-space parameter groups. The K-means clustering algorithm is well known and will not be described in detail here.

[0078] The wind turbine power, state space parameters and reward function within 30 minutes before the current moment are used as the training set, and the Q-Learning reinforcement learning algorithm is used for training to obtain the trained Q function. The training process is as follows:

[0079] During the training process, the Q-Learning reinforcement learning algorithm generates a table with state space parameter groups as rows and action space parameter groups as columns. Each cell element is the value of the corresponding reward function. The table is continuously updated according to the value of the reward function at each moment within 30 minutes before the current moment obtained in the above steps to generate the Q function.

[0080] The state space parameter data of the evaporative cooling air conditioner at the current moment is used as the input of the Q function, and the output is the optimal fan power of the evaporative cooling air conditioner cooling tower at the current moment. The state space parameters at the current moment are classified into the state space parameter group; specifically: the state space parameters at the current moment and the clustered state space parameter groups are classified using the KNN classification algorithm to obtain the state space parameter group to which the state space parameters at the current moment belong. The action space parameter group corresponding to the maximum Q function value in the state space parameter group to which the state space parameters at the current moment belong is used as the execution action group; the middle value of the execution action group is used as the optimal fan power of the evaporative cooling air conditioner cooling tower to achieve energy-saving optimization control of the evaporative cooling air conditioner. The Q-Learning reinforcement learning algorithm and the KNN classification algorithm are well-known technologies and will not be described in detail here.

[0081] Through the above description of the implementation method in combination with the accompanying drawings, technical personnel in the relevant field can understand that for the convenience and simplicity of description, only the division of the above-mentioned functional modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by 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.

[0082] The above content is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A method for energy-saving optimization control of evaporative cooling air conditioner based on deep reinforcement learning, characterized in that: The method comprises the following steps: S1, obtain the ambient temperature and the power of the fan, water pump, and evaporator at each moment during the operation of the evaporative cooling air conditioner; S2, calculate the heat load rise index based on the change trend and change range of the ambient temperature in the local period at each moment; S3, according to the similarity of the peak fluctuations of the fan, water pump and evaporator power in the local period at each moment, obtain the energy consumption peak synchronization index at each moment; S4, calculates the environmental complexity based on the difference between the heat load rise index and the energy consumption peak synchronization index at each moment; comprehensively considers the ambient temperature and environmental complexity at each moment to set the reward function in the deep reinforcement learning algorithm; S5, optimizing and controlling the fan power of the evaporative cooling air conditioner using the deep reinforcement learning algorithm based on the current ambient temperature, the power of the water pump, and the evaporator; The heat load rise index is calculated based on the change trend and change range of the ambient temperature in the local time period at each moment, including: At time t, the sequence of ambient temperatures at a preset number of moments before time t is recorded as the neighboring temperature monitoring sequence at time t; The existence of an upward trend or downward trend in the sequence is used as the hypothesis test condition in the Wilcoxon signed rank test algorithm. The Wilcoxon signed rank test algorithm is used to perform a statistical test on the neighboring temperature monitoring sequence at time t, and the p-value of the existence of an upward trend or downward trend in the neighboring temperature monitoring sequence at time t is obtained; The heat load rising index at time t is calculated by comprehensively considering the p-values ​​of the upward and downward trends and the degree of change of the neighboring temperature monitoring series at time t. The calculation formula of the heat load rising index at time t is: in, represents the heat load rising index at time t, 、 The p-values ​​represent whether the neighboring temperature monitoring series at time t has an upward trend or a downward trend respectively; represents the sum of all positive numbers in the first-order difference temperature series at time t; Represents the absolute value of the sum of all negative numbers in the first-order difference temperature series at time t; 1 is a preset constant greater than 0; The method for calculating the environmental complexity is as follows: calculating the absolute value of the difference between the normalized value of the heat load rise index and the normalized value of the energy consumption peak synchronization index at each moment as the environmental complexity at each moment; The reward function in the deep reinforcement learning algorithm is set as follows: in, represents the complexity of the environment at time t; represents the ambient temperature at time t; Indicates the preset comfortable temperature; Represents the reward function in the deep reinforcement learning algorithm at time t; exp() is an exponential function with a natural constant as the base.

2. The energy-saving optimization control method for evaporative cooling air conditioning based on deep reinforcement learning according to claim 1, characterized in that: The method for obtaining the energy consumption peak synchronization index at each moment includes: At time t, the sequence of wind turbine power at time t and within the preset period before it is subjected to first-order difference processing to obtain the wind turbine power difference sequence at time t; all maximum values ​​in the wind turbine power difference sequence are arranged in chronological order to obtain the wind turbine power peak sequence at time t; Obtain the water pump power peak sequence and the evaporator power peak sequence at time t; record the set consisting of the fan power peak sequence, the water pump power peak sequence, and the evaporator power peak sequence at time t as the feature set at time t; For any two sequences in the feature set, obtaining the peak similarity between the two sequences based on the similarity measurement between the two sequences; The sum of the peak similarities between all pairs of sequences in the feature set is calculated as the energy consumption peak synchronization index at time t.

3. The energy-saving optimization control method for evaporative cooling air conditioning based on deep reinforcement learning according to claim 1, characterized in that: The method for obtaining the peak similarity between the two sequences includes: calculating the dynamic time warping distance between the two sequences; and using the inverse normalized value of the dynamic time warping distance as the peak similarity between the two sequences.

4. The energy-saving optimization control method for evaporative cooling air conditioning based on deep reinforcement learning according to claim 1, characterized in that: The method of optimizing the fan power of the evaporative cooling air conditioner using the deep reinforcement learning algorithm according to the current ambient temperature, the power of the water pump, and the evaporator includes: In the deep reinforcement learning algorithm, the operating range of the evaporative cooling air conditioner fan power is divided into a first preset number of intervals, and each interval is used as an action space parameter group. The ambient temperature, water pump, and evaporator power are used as state space parameters. The state space parameters within the first preset period before the current moment are clustered to obtain a second preset number of state space parameter groups. The action space parameter group, state space parameter group, and reward function of the wind turbine power at each moment in the first preset period before the current moment are trained using the Q-Learning reinforcement learning algorithm to obtain a trained Q function. The state space parameter data of the evaporative cooling air conditioner at the current moment is input into the Q function to obtain the optimal fan power at the current moment, thereby realizing energy-saving optimization control of the evaporative cooling air conditioner.

5. The energy-saving optimization control method for evaporative cooling air conditioning based on deep reinforcement learning according to claim 4, characterized in that: The state space parameter data of the evaporative cooling air conditioner at the current moment is input into the Q function to obtain the optimal fan power at the current moment, thereby realizing energy-saving optimization control of the evaporative cooling air conditioner, including: The state space parameters at the current moment are classified into the state space parameter group. The action space parameter group corresponding to the maximum reward function value in the state space parameter group to which the state space parameters at the current moment belong is used as the execution action group. The middle value of the execution action group is used as the optimal fan power of the evaporative cooling air conditioning cooling tower.

6. The energy-saving optimization control method for evaporative cooling air conditioning based on deep reinforcement learning according to claim 5, characterized in that: The method for classifying the current state space parameter into the state space parameter group is: using the KNN classification algorithm to classify the current state space parameter and each clustered state space parameter group to obtain the state space parameter group to which the current state space parameter belongs.

Images