A central air conditioning energy-saving optimization control method and system

Abstract

The application relates to the technical field of central air conditioner control, in particular to a central air conditioner energy-saving optimization control method and system. The method comprises the following steps: constructing a Q value table, obtaining current time parameter data, selecting corresponding actions according to the state section corresponding to the current time parameter data and the Q value table, controlling the working state of the central air conditioner, obtaining a reward value and a new state according to the selected actions, updating the Q value, and using the updated Q value for next time control. That is, the scheme can intelligently adjust the working state of the central air conditioner according to the indoor environment and user demand, realize more accurate environment control, and improve the user experience.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of central air conditioning control. More particularly, the present application relates to a central air conditioning energy-saving optimization control method and system. BACKGROUND

[0002] As an important environmental conditioning equipment in modern buildings, the central air conditioning system is mainly composed of a cold and heat source system and an air conditioning system, which uses the principle of liquid vaporization refrigeration to provide the required cold quantity for the air conditioning system to offset the indoor environmental heat load; at the same time, provides the required heat to offset the indoor environmental cold load, so as to maintain the indoor temperature within the human comfort range.

[0003] With the improvement of people's living standards, the demand for indoor temperature comfort is increasing, and users expect the air conditioning system to dynamically adjust according to real-time environment and human perception (such as temperature, humidity, etc.), to realize personalized comfort experience, rather than the traditional fixed temperature set point, which often leads to excessive refrigeration or heating, causing unnecessary energy waste.

[0004] In terms of energy saving, the central air conditioning system is a major component of building energy consumption, accounting for a large proportion of the entire building energy consumption, especially in large commercial buildings or data centers, the operation energy consumption can reach more than 40% of the total energy consumption. The traditional control method usually adopts independent control of each component (such as air cooling heat pump unit, refrigerated water circulating pump, etc.), although it can apply variable frequency energy-saving technology to improve local efficiency (such as reducing water pump energy consumption by adjusting motor speed), but it is difficult to coordinate the logical relationship between multiple targets, and cannot realize overall system optimization, resulting in low efficiency and rising energy consumption.

[0005] In addition, under the background of the "double carbon" goal (carbon peak, carbon neutralization), the carbon emissions of the heating, ventilation and air conditioning system in the building field are huge, and it is urgent to reduce energy consumption through data-driven intelligent control strategies.

[0006] The limitations of the prior art mainly lie in the poor model analysis effect, which is difficult to dynamically adapt to environmental changes and user needs, resulting in insufficient comprehensive control strategy and poor balance between energy efficiency and effect. SUMMARY

[0007] The present application aims to provide a central air conditioning energy-saving optimization control method and system to solve the problem that the control method in the prior art is difficult to dynamically adapt to environmental changes and user needs, resulting in insufficient comprehensive control strategy and poor balance between energy efficiency and effect. To this end, the present application provides solutions in the following two aspects.

[0008] The present application provides a central air conditioning energy-saving optimization control method, comprising:

[0009] A Q-value table is constructed, a state space in the Q-value table comprises a plurality of state segments, and an action space comprises a plurality of operating states;

[0010] Each state segment is: obtaining historical parameter data of the central air conditioner in a historical working period, the historical parameter data comprising parameters in multiple dimensions, parameters of the same dimension at different times forming a parameter sequence, the multiple dimensions comprising indoor temperature, user behavior, and a time label, the user behavior representing the activity level of indoor personnel; segmenting each parameter sequence by using a dimension segmentation point, and forming a state segment by using all dimension local segments in the same historical period; the segmentation point being each dimension parameter corresponding to a time point when the central air conditioner switches between different operating states in the historical working period.

[0011] Obtaining current time parameter data, and selecting a corresponding action according to the current time parameter data and the Q-value table to control the working state of the central air conditioner.

[0012] The reward value when the corresponding action is selected is the inverse of the weighted sum of a feature value, energy consumption, and fan noise of an indoor unit in the current state; the feature value representing the difference in indoor temperature change in the current state; the sum of the weights of the feature value, the energy consumption, and the fan noise being 1 when the weighted sum is performed, the weight of the feature value being adjusted by a first adjustment coefficient to a corresponding initial weight, the first adjustment coefficient being the difference change of the average indoor temperature and the target temperature in the current state; the weight of the fan noise being adjusted by a second adjustment coefficient to a corresponding initial weight, the second adjustment coefficient being positively correlated with the noise level relative change and the time label in the current state, the noise level relative change being the absolute value of the ratio of a first difference value and a noise threshold, the first difference value being the difference between the fan noise and the noise threshold.

[0013] The above scheme is based on historical parameter data of the central air conditioner in a historical working period, a Q-value table is constructed in advance, and then the central air conditioner can intelligently adjust the working state according to different indoor environments and user needs in combination with the constructed Q-value table, to achieve more accurate environmental control and improve user experience. At the same time, the energy consumption and the fan noise are comprehensively considered, so that the central air conditioner can effectively reduce the energy consumption and the noise level while ensuring the temperature regulation effect, to achieve energy saving and environmental protection and comfortable operation. That is, compared with the traditional single factor control mode, the scheme of the present application can more comprehensively meet the demand for indoor environment regulation.

[0014] Preferably, the plurality of operating states at least include power on / off, and speed adjustment mode in cooling / heating mode.

[0015] Preferably, the method further comprises: before calculating the reward value, standardizing the indoor temperature, the energy consumption, and the fan noise of the indoor unit to obtain standardized indoor temperature, energy consumption, and fan noise of the indoor unit.

[0016] Preferably, the reward value is:

[0017] ;

[0018] Where r is the reward value, denoted as the characteristic value under the current state, which is the average difference between the indoor temperature and the target temperature at all times under the current state, P is the energy consumption under the current state, N is the fan noise under the current state, and w1, w2, and w3 are the weights of the characteristic value, energy consumption, and fan noise, respectively.

[0019] The above provides a method for accurately calculating reward values.

[0020] Preferably, the weights for the eigenvalues ​​and fan noise are as follows:

[0021] ;

[0022] ;

[0023] Where w1 and w3 are the weights of the eigenvalue and the fan noise, respectively. The initial weights for the eigenvalues, This is the first adjustment coefficient under the current state t. ; The average indoor temperature under the current state t. The target temperature; ≤0.1, Here, m is the initial weight for fan noise, and m is the second adjustment coefficient. S0 is a time tag, with a value of 1 or 0, where 1 represents night and 0 represents day. Noise threshold The noise level is the fan noise in the current state t.

[0024] The initial weights are adjusted based on parameters from the actual scenario (such as user feedback), enabling adaptive adjustment of the importance of different parameters.

[0025] Preferably, the process of obtaining the user behavior is as follows:

[0026] Obtain the number of people and noise levels in the target area at different times;

[0027] By combining the number of people and the noise level, the intensity of user activity at each time point can be obtained;

[0028] The average user activity intensity under each window is taken as the average user activity at the corresponding time at the end of the window, and the rate of change between the average user activity at two adjacent time points is taken as the user behavior at the next time point; the noise level is noise other than fan noise and background noise.

[0029] By introducing user behavior as a parameter, the impact of user activities on indoor environmental control can be fully considered.

[0030] The preferred rule for updating the Q value is as follows:

[0031] ;

[0032] in, This is a discount factor used to reduce the importance of future rewards. For state-action pairs >Value function, For state-action pairs >Reward value, For state-action pairs >Value function, For state The corresponding action, For the nth state, This is the (n+1)th state. For state The corresponding action.

[0033] Preferably, the energy consumption is the product of the power of the central air conditioner and the operating time; the fan noise is obtained by extracting the characteristic frequency range of fan operation from the total environmental noise through spectrum analysis.

[0034] Preferably, it also includes outdoor temperature; wherein both indoor and outdoor temperatures are obtained by temperature sensors; the number of people is obtained by optical sensors; and the total environmental noise is obtained by microphones.

[0035] This invention provides a central air conditioning energy-saving optimization control system, comprising:

[0036] processor;

[0037] The memory stores computer instructions for energy-saving optimization control of central air conditioning. When the computer instructions are executed by the processor, the system performs the aforementioned energy-saving optimization control method for central air conditioning.

[0038] The beneficial effects of this invention are as follows:

[0039] The solution of this invention can analyze the historical operating status of the central air conditioning system, construct a Q-value table, and comprehensively consider multiple factors such as environmental parameters, user behavior, and time when analyzing historical data. Compared with the traditional single-factor control method, it can realize the dynamic adjustment of the central air conditioning system in the future, meet the user's temperature comfort requirements, minimize global energy consumption, improve system efficiency, and reduce carbon emissions. Attached Figure Description

[0040] Figure 1 The flowchart illustrating the steps of a central air conditioning energy-saving optimization control method in this embodiment is shown in the illustration. Detailed Implementation

[0041] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0042] like Figure 1 As shown in this embodiment, a central air conditioning energy-saving optimization control method includes the following steps:

[0043] Step S1: Establish a control model for the central air conditioning system based on the Q-learning algorithm. The control model includes a Q-value table.

[0044] Q-learning is a model-free reinforcement learning algorithm. In reinforcement learning, an agent changes the state of an environment by performing actions and receives a reward based on the state transitions. The agent's goal is to maximize the long-term cumulative reward, which typically involves learning a policy—a rule for choosing the best action given a state.

[0045] Q-learning falls under the category of reinforcement learning. The core of Q-learning is the Q-value table, a two-dimensional table where rows represent states and columns represent actions. The Q-value represents the expected cumulative reward for taking an action in a given state. A higher Q-value indicates a better long-term reward for taking that action in that state.

[0046] The process of obtaining the Q-value table in this embodiment is as follows:

[0047] S11, Initialize the Q-value table. The Q-values ​​in the initial Q-value table are random numbers or small values.

[0048] S12, construct the state space S and action space a in the Q-value table.

[0049] The central air conditioning system includes multiple operating states, including on / off and fan speed levels in cooling / heating modes. These fan speed levels include at least no fan speed, low, medium, and high.

[0050] At this time, there are 10 operating states, namely, power on, power off, 4 fan speed levels in cooling mode, and 4 fan speed levels in heating mode.

[0051] Taking summer as an example, the multiple operating states of the action space in this embodiment are power on, power off, and different wind speed levels corresponding to the cooling mode.

[0052] Furthermore, central air conditioning systems also have air purification functions, such as HEPA filters and negative ion generators. Therefore, the operating state can be different combinations of fan speed levels and filtration modes in cooling mode. There are at least 8 combinations and at most 12 combinations (4 of these 12 combinations are to account for the fact that some central air conditioning systems can use filtration modes in parallel, such as HEPA + negative ion generator).

[0053] It should be noted that the specific combination methods mentioned above can be determined based on the functions of the central air conditioning system. Through these different operating modes, the central air conditioning system can meet various indoor environmental needs.

[0054] The process of obtaining the state space is as follows:

[0055] First, obtain the parameter sequence of the central air conditioning system in multiple dimensions during historical operating periods.

[0056] This includes multiple dimensions such as environmental parameters, user behavior, and time stamps. Parameters at all times within the same dimension constitute a parameter sequence.

[0057] Specifically, the temperature sensor built into the central air conditioning system continuously collects environmental parameters (including outdoor temperature and indoor temperature).

[0058] In this embodiment, the number of people indoors is obtained through an optical sensor; the total ambient noise and background noise are obtained through a microphone installed indoors. The background noise is the average noise level measured over a period of time (such as one day or several days) when there is no human activity and the central air conditioning is not running.

[0059] Data from each dimension is collected every minute, and historical data for a period of time (such as one week) is stored for analysis.

[0060] User behavior represents the activity level of indoor personnel during historical working periods. Specifically, user behavior is acquired as follows:

[0061] The number of people and noise levels at each time point within the acquired historical working period are standardized. The standardized number of people and noise levels are then fused to obtain the user activity intensity at each time point. The mean of user activity intensity at each window is calculated, and this mean is used as the user activity mean at the end of the window. The rate of change between the user activity mean at two adjacent time points is used as the user behavior at the next time point.

[0062] The above fusion can be the average of the standardized number of people and the noise level, or it can be a weighted sum of the standardized number of people and the noise level by setting different fusion weights.

[0063] The noise levels mentioned above exclude fan noise and background noise. Fan noise refers to the noise generated during the operation of the central air conditioning system.

[0064] In one embodiment, the process of obtaining fan noise and noise level is as follows:

[0065] First, obtain the total environmental noise data for the historical working period.

[0066] Specifically, for example, during peak daytime activity periods, ensure that user activity and fan noise coexist, i.e., collect audio signals from the target area as the total environmental noise.

[0067] Secondly, different noise sources are separated by signal processing techniques (such as adaptive filtering and blind source separation) to obtain fan noise and noise levels.

[0068] For example, taking adaptive filtering as an example, the process of obtaining fan noise and noise level is as follows:

[0069] First, determine the reference signal.

[0070] Specifically, a reference signal for fan noise can be obtained, such as by recording a segment of "pure" noise from the fan running near the fan, or by indirectly generating a periodic reference signal from the fan speed (RPM) signal.

[0071] Secondly, an adaptive filter is constructed.

[0072] The filter is designed using an adaptive algorithm (such as LMS or NLMS). The input to the filter is the mixed signal x(t'), and the output is the estimated fan noise y(t'), where t' is the time. That is, y(t') = w(t') × r(t'), where w(t') is the filter weight and r(t') is the reference signal.

[0073] Then, error calculation and weight update.

[0074] Specifically, the error signal e(t') is: e(t') = x(t') - y(t').

[0075] Update the filter weights using the LMS algorithm: w(t'+1)=w(t')+μ×e(t')×r(t'), where μ is the learning rate, and its value ranges from 0.01 to 0.1.

[0076] Finally, the fan noise was extracted, and a fast Fourier transform was performed on the separated fan noise to analyze its spectral characteristics.

[0077] In this embodiment, the filter output y(t') is the separated fan noise signal.

[0078] Here, e(t') includes the noise levels of both environmental noise and user activity. Based on the frequency range of the user activity noise level, the time-frequency spectrum of the error signal e(t') is extracted to obtain the noise level of the user activity. The time-frequency spectrum is obtained through a short-time Fourier transform.

[0079] User activity noise (such as talking) is typically concentrated in the 200Hz-4kHz range, footsteps in the 50-500Hz range, while ambient noise (such as background white noise) is more evenly distributed. Therefore, frequency domain analysis of the error signal is performed to obtain the noise level.

[0080] Since adaptive filtering and blind source separation are existing technologies, they will not be elaborated on here.

[0081] In another embodiment, the fan noise may also be the difference between the total ambient noise at each moment and the background noise when the fan is not running, or the average of the differences at all moments.

[0082] The above window can be set for either half an hour or 15 minutes.

[0083] The above standardization can be the maximum and minimum value normalization method or the Z-Score normalization method.

[0084] Secondly, obtain the split points of the parameter sequence in multiple dimensions.

[0085] The dividing point in this embodiment is the parameters of each dimension at the corresponding moment when the central air conditioning switches between different operating states during the statistical historical working period.

[0086] Specifically, by statistically analyzing the parameters of each dimension at the corresponding moment when the central air conditioning switches between different operating states, we can obtain all dimension data at all switching moments, and thus obtain the segmentation point.

[0087] For example, when the wind speed is switched from low to high, parameters such as temperature, humidity, user behavior, and time stamps are recorded. Through statistical analysis of a large amount of historical data, relatively accurate segmentation points are obtained.

[0088] Then, the parameter sequences are segmented using all the split points. Multiple local segments from different dimensions within the same historical time period are combined to form a state segment, and these multiple state segments form a state space. For example, the temperature sequence is divided into several segments based on the split points, and other parameter sequences such as humidity and user behavior are also segmented in the same way. For example, within a specific one-hour historical time period, if the temperature is in one segment, the humidity is in another segment, and the user behavior is in another segment, with a time label of 1 (night), then these local segments are combined into a single state segment.

[0089] Each state segment represents a state (theoretically, the parameters in each dimension do not fluctuate significantly within a defined timeframe). Different state segments correspond to different states, and the transition from one state segment to another indicates that one or more dimensions of the parameters have experienced substantial fluctuations. Each state segment provides rich and detailed state information for the control model.

[0090] S12, construct the reward value, update the initial Q-value table, and obtain the updated Q-value table.

[0091] The reward value is negatively correlated with the characteristic value, energy consumption, and fan noise under the current state. This reward value comprehensively considers environmental changes, energy consumption, and fan noise to guide the control model to make better decisions.

[0092] Before calculating the reward value, the characteristic values, energy consumption, and fan noise under the current state need to be standardized to obtain standardized characteristic values, energy consumption, and fan noise, thus eliminating the dimensions.

[0093] Specifically, the reward value is:

[0094] ;

[0095] Where r is the reward value, Let w1 be the characteristic value under the current state, which is the average difference between the indoor temperature and the target temperature at all times. Let P be the energy consumption under the current state, and N be the fan noise under the current state. w1, w2, and w3 are the weights of the characteristic value, energy consumption, and fan noise, respectively.

[0096] The energy consumption mentioned above is the product of the power of the central air conditioning system and the operating time. Specifically, for a power time series, the energy consumption can be the product of the average power and the operating time.

[0097] In one embodiment, the weights of the eigenvalues ​​and fan noise can be obtained by adjusting the initial weights, specifically as follows:

[0098] The weights for the difference, energy consumption, and fan noise are as follows:

[0099] ;

[0100] ;

[0101] ;

[0102] in, The initial weights for the eigenvalues, This is the first adjustment coefficient; ≤0.1, The initial weight for fan noise is set to 0.2, and m is the second adjustment coefficient. The initial weight for the feature value is 0.5.

[0103] The first adjustment coefficient mentioned above is the absolute value of the relative change in environmental parameters.

[0104] Specifically, the first adjustment coefficient is: ; The average indoor temperature under the current state t. The target temperature.

[0105] The target temperature mentioned above is set according to the actual situation, but the value is not 0 degrees Celsius. For example, in summer, the target temperature can be 25 degrees Celsius.

[0106] The second adjustment coefficient is positively correlated with the relative change in noise level under the current state and the time label; specifically, the second adjustment coefficient m is: m S0 is a time tag, with a value of 1 or 0, where 1 represents night and 0 represents day. This is the noise threshold, which can be 30dB. The noise level is the fan noise in the current state t. This represents the relative change in noise levels.

[0107] The noise thresholds mentioned above represent the noise levels that users can tolerate, such as 40 dB.

[0108] By adjusting the initial weights as described above, the adjusted weights can better reflect the requirements of the actual scenario. For example, when it is nighttime, users are often more sensitive to noise, so the weights should be set higher; conversely, when it is daytime, users are less sensitive to noise, so the weights can be set lower.

[0109] In another embodiment, the initial weights of the aforementioned eigenvalues, energy consumption, and fan noise can also be the information gains of the eigenvalues, energy consumption, and fan noise calculated separately, that is, each information gain is used as the initial weight.

[0110] Since the acquisition of information gain is a current technology, it will not be elaborated here.

[0111] The aforementioned bonus value incorporates energy consumption and fan noise, enabling accurate assessment of the central air conditioning system's operating status.

[0112] In this embodiment, based on the set reward value and combined with the status and actions during historical working periods, a Q-value table of the central air conditioning system during historical working periods can be obtained.

[0113] Step S2: Obtain the parameter data at the current moment, and select the corresponding action according to the state segment and Q value table corresponding to the parameter data at the current moment to control the working state of the central air conditioner. Obtain the reward value and new state according to the selected action, update the Q value, and use the updated Q value for control in the next moment.

[0114] Specifically, after constructing the Q-value table, the specific control process is as follows:

[0115] Step S21: Determine the state segment to which the current time period parameter data belongs based on the indoor current time parameter data.

[0116] Specifically, the current indoor parameter data is matched with each state segment to obtain the corresponding state segment.

[0117] In one embodiment, if all dimensions of the parameter data at the current moment are within a certain state segment, then the parameter data at the current moment is considered to belong to the state of the corresponding state segment.

[0118] In another embodiment, parameter data from the most recent times preceding the current time parameter data can be obtained to form the current time period parameter data. The DTW algorithm is then used to calculate the shortest distance between each parameter sequence in the current time period parameter data and the corresponding parameter sequence in each state segment, thus obtaining the state segment where the mean of the shortest distance is minimized across all dimensions.

[0119] Step S22: The agent selects an action based on the state segment and Q-value table corresponding to the parameter data at the current moment.

[0120] The agent can use a greedy algorithm to select the action with the highest Q value. For example, in the current state segment, the action with the highest Q value recorded in the Q value table is the low wind speed + HEPA filtration mode, so the agent selects that action.

[0121] In step S23, the agent executes the selected action, observes the response of the environment, and calculates the reward value and the new state.

[0122] For example, the central air conditioning system performs a selected action, such as switching to a low fan speed + HEPA filter mode. Simultaneously, sensors collect environmental response data in real time, such as energy consumption and fan noise. Based on this data, a reward value is calculated, and a new state segment is determined. For instance, after performing the action, the temperature drops from 30 degrees Celsius to 20 degrees Celsius, energy consumption is 50W, and fan noise is 30dB. A reward signal is calculated using a function of the reward value, and a new state segment is determined based on the new parameter data.

[0123] Step S24: Update the Q value based on the reward signal and the new state.

[0124] The update rule for the Q value is as follows:

[0125] Set an initial value function and perform initialization;

[0126] Based on the set reward function, the reward value corresponding to each action selected by the agent is obtained until the sum of the reward values ​​is maximized, and the initial value function is updated; specifically, the update is as follows:

[0127] ;

[0128] in, This is a discount factor used to reduce the importance of future rewards. For state-action pairs >Value function, For state-action pairs >Reward value, For state-action pairs >Value function, For state The corresponding action, For the nth state, This is the (n+1)th state. For state The corresponding action.

[0129] In step S25, the agent continuously executes the above steps, interacts with the environment, learns and improves the Q-value function until the stopping condition is met.

[0130] The above stopping condition can be reaching the maximum number of iterations.

[0131] The solution of this invention can make full use of multi-dimensional parameter data and use the Q-learning algorithm to achieve intelligent, precise and efficient control of the central air conditioning working status. While improving environmental control, it optimizes energy consumption and noise control, creating a more comfortable, healthy and energy-saving indoor environment for users.

[0132] Furthermore, in practical applications, regular sensor calibration should be considered to ensure the accuracy of the collected data. Simultaneously, reward values, weight settings, and state segment divisions can be further optimized based on user feedback or big data analysis, continuously improving the intelligence level and user experience of the central air conditioning system.

[0133] Furthermore, the present invention also provides a central air conditioning energy-saving optimization control system, including a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-described central air conditioning energy-saving optimization control method according to the present invention is implemented.

[0134] Since the embodiments of a central air conditioning energy-saving optimization control method have been described in detail above, they will not be repeated here. In the description of this specification, "multiple" means at least two, such as two, three or more, etc., unless otherwise explicitly specified.

[0135] While various embodiments of the invention have been shown and described in this specification, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention.

Claims

1. A method for optimizing energy-saving control of a central air conditioning system, characterized in that, include: Construct a Q-value table, wherein the state space in the Q-value table includes multiple state segments, and the action space includes multiple running states; Each state segment is as follows: Historical parameter data of the central air conditioning system during historical operating periods is acquired. This historical parameter data includes parameters across multiple dimensions. Parameters of the same dimension at different times constitute a parameter sequence. These multiple dimensions include indoor temperature, user behavior, and time stamps. User behavior represents the activity level of people indoors. Each parameter sequence is segmented using the segmentation points of each dimension, and all local segments of all dimensions within the same historical period constitute a state segment. The segmentation points are the parameters of each dimension at the corresponding time when the central air conditioning system switches between different operating states during historical operating periods. Obtain the parameter data at the current moment, and select the corresponding action based on the parameter data at the current moment and the Q value table to control the working status of the central air conditioning. The reward value for selecting the corresponding action is the negative of the weighted sum of the feature value, energy consumption, and indoor unit fan noise in the current state. The feature value represents the difference in indoor temperature change in the current state. The weighted sum of the feature value, energy consumption, and fan noise is 1. The weight of the feature value is obtained by adjusting the corresponding initial weight through a first adjustment coefficient, which is the difference between the average indoor temperature and the target temperature in the current state. The weight of the fan noise is obtained by adjusting the corresponding initial weight through a second adjustment coefficient, which is positively correlated with the relative change in noise level and the time label in the current state. The relative change in noise level is the absolute value of the ratio of the first difference to the noise threshold, where the first difference is the difference between the fan noise and the noise threshold.

2. The central air conditioning energy-saving optimization control method according to claim 1, characterized in that, The multiple operating states include the power on / off state and the fan speed adjustment level in cooling / heating modes.

3. The central air conditioning energy-saving optimization control method according to claim 1, characterized in that, Also includes: Before calculating the reward value, the indoor temperature, energy consumption, and indoor unit fan noise are standardized to obtain the standardized indoor temperature, energy consumption, and indoor unit fan noise.

4. The central air conditioning energy-saving optimization control method according to claim 3, characterized in that, The reward value is: ； Where r is the reward value, denoted as the characteristic value under the current state, which is the average difference between the indoor temperature and the target temperature at all times under the current state, P is the energy consumption under the current state, N is the fan noise under the current state, and w1, w2, and w3 are the weights of the characteristic value, energy consumption, and fan noise, respectively.

5. The central air conditioning energy-saving optimization control method according to claim 4, characterized in that, The weights for eigenvalues ​​and fan noise are as follows: ； ； Where w1 and w3 are the weights of the eigenvalue and the fan noise, respectively. The initial weights for the eigenvalues, This is the first adjustment coefficient under the current state t. ; The average indoor temperature under the current state t. The target temperature; ≤0.1, Here, m is the initial weight for fan noise, and m is the second adjustment coefficient. S0 is a time tag, with a value of 1 or 0, where 1 represents night and 0 represents day. Noise threshold The noise level is the fan noise in the current state t.

6. The central air conditioning energy-saving optimization control method according to claim 1, characterized in that, The process of obtaining the user behavior is as follows: Obtain the number of people and noise levels in the target area at different times; The standardized number of people and noise levels are combined to obtain the user activity intensity at each time point; The average user activity intensity under each window is taken as the average user activity at the corresponding time at the end of the window, and the rate of change between the average user activity at two adjacent time points is taken as the user behavior at the next time point; the noise level is noise other than fan noise and background noise.

7. The central air conditioning energy-saving optimization control method according to claim 2, characterized in that, The rules for updating the Q value are as follows: ； in, This is a discount factor used to reduce the importance of future rewards. For state-action pairs >Value function, For state-action pairs >Reward value, For state-action pairs >Value function, For state The corresponding action, For the nth state, This is the (n+1)th state. For state The corresponding action.

8. The central air conditioning energy-saving optimization control method according to claim 6, characterized in that, The energy consumption is the product of the power of the central air conditioner and its operating time; the fan noise is obtained by extracting the characteristic frequency range of the fan operation from the total environmental noise through spectrum analysis.

9. A central air conditioning energy-saving optimization control method according to claim 8, characterized in that, It also includes outdoor temperature; both indoor and outdoor temperatures are collected by temperature sensors; the number of people is collected by optical sensors; and the total environmental noise is collected by microphones.

10. A central air conditioning energy-saving optimization control system, characterized in that, include: processor; A memory storing computer instructions for energy-saving optimization control of a central air conditioning system, wherein when the computer instructions are executed by the processor, the system performs an energy-saving optimization control method for a central air conditioning system according to any one of claims 1-9.

Images