Compressor starting and stopping point parameter adjusting system based on Bayesian optimization
By using a Bayesian-optimized compressor start-stop point parameter adjustment system, the problem of low efficiency in refrigerator control parameter adjustment in existing technologies is solved. This system achieves adaptive global optimization of the refrigerator's start-stop point, reduces testing costs, and improves energy efficiency.
Patent Information
- Application Number
- CN202511199679.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-26
- Publication Date
- 2025-11-07
AI Technical Summary
Existing compressor control technology is inefficient when adjusting refrigerator control parameters, making it difficult to achieve global optimization. Furthermore, the PID control method has poor adaptability in complex environments, which can easily lead to increased energy consumption and unstable temperature.
A compressor start-stop point parameter adjustment system based on Bayesian optimization is adopted. Real-time and historical data are acquired through the data acquisition module, a Gaussian process regression model is constructed, and the start-stop point is adjusted using the Bayesian optimization function to achieve automated global optimal adjustment.
It significantly reduced laboratory testing costs, achieved adaptive global optimization of refrigerator start-up and shutdown point parameters, and improved temperature stability and energy efficiency.
Smart Images
Figure CN120906784A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of compressor control, and particularly relates to a compressor start-stop point parameter adjustment system based on Bayesian optimization. BACKGROUND
[0002] As indispensable refrigeration equipment in modern families and commercial places, refrigerators are of various types and functions, and have been widely applied to food cold storage and preservation, frozen storage and intelligent home integrated systems. The control quality of the internal temperature of a refrigerator is directly related to food preservation effect, equipment energy consumption level and whole machine operation reliability, and the start-stop control strategy of a compressor is particularly crucial. Core control parameters such as the start temperature, stop temperature and hysteresis temperature of the compressor determine multiple important performance indexes such as the temperature fluctuation range in the refrigerator, the start-stop frequency of the compressor, the system energy consumption and the mechanical life of components.
[0003] At present, the method for adjusting the control parameters of a refrigerator in a laboratory environment mainly depends on a traditional manual trial and error method or an adjustment strategy based on classic proportional-integral-derivative (PID) control. These methods still have obvious limitations in actual application.
[0004] Firstly, the manual trial and error method highly depends on the experience level of a debugging personnel and lacks systematicness and theoretical guidance. An operator usually needs to perform multiple rounds of repeated tests, and the whole debugging cycle is long, and the global optimality of the parameter combination cannot be guaranteed. The method is low in efficiency, high in test cost, and difficult to meet the requirements of modern production on debugging efficiency and consistency.
[0005] Secondly, although the PID control method has certain self-adaptive ability, the parameters thereof are usually set under fixed working conditions, and it is difficult to adapt to complex and changeable environmental conditions in actual use, such as environmental temperature fluctuation caused by seasonal change, change of the load in the refrigerator and system performance aging in long-term operation. In addition, the PID control is sensitive to sensor noise, and is easy to cause frequent start-stop of the compressor or control deviation, thereby causing increased energy consumption and decreased temperature stability. SUMMARY
[0006] The present application provides a compressor start-stop point parameter adjustment system based on Bayesian optimization, to solve the problem of low efficiency in the global optimal point search process of a compressor by using existing compressor control technology.
[0007] The system comprises:
[0008] a data acquisition module configured to acquire real-time data of a compressor to be adjusted, an observation space, and an action space; the observation space and the action space are generated based on historical data of the compressor to be adjusted, the historical data being of the same data type as the real-time data; the observation space is used to represent an environment state and a device state of the compressor to be adjusted, and the action space is used to represent an on-off point of the compressor to be adjusted;
[0009] an on-off point adjustment module configured to perform model training according to the observation space, the action space, and a Gaussian process regression model to obtain a target Gaussian process regression model; the Gaussian process regression model is constructed based on a Bayesian optimization function;
[0010] an on-off execution module configured to generate target on-off data from the real-time data and the target Gaussian process regression model, and adjust the on-off point of the compressor to be adjusted through the target on-off data.
[0011] Preferably, the data acquisition module is further configured to:
[0012] acquire historical data of the compressor to be adjusted;
[0013] preprocess the historical data to obtain the observation space;
[0014] construct the action space according to an on-off specification of the compressor to be adjusted.
[0015] Preferably, the data acquisition module comprises:
[0016] a data acquisition unit configured to acquire the real-time data and the historical data of the compressor to be adjusted;
[0017] a preprocessing unit configured to preprocess the historical data to obtain preprocessed data; the preprocessing includes noise reduction processing and abnormal data filtering processing;
[0018] a first space construction unit configured to perform first data space construction according to the preprocessed data to obtain the observation space;
[0019] a second space construction unit configured to perform second data space construction according to an on-off specification of the compressor to be adjusted to obtain the action space.
[0020] Preferably, the system is applied to a compressor in a refrigerator.
[0021] Preferably, the observation space comprises indoor temperature, indoor humidity, set temperature of refrigerator compartment, temperature of refrigerator compartment, set temperature of freezer compartment, temperature of freezer compartment, temperature of freezing evaporator, defrosting heater state, door opening state, compressor rotation speed, valve state, fan state, compressor state, compressor current value and compressor power value in the environment where the compressor to be adjusted is located;
[0022] The action space comprises refrigerator compartment start-up point temperature, freezer compartment start-up point temperature, refrigerator compartment shutdown point temperature and freezer compartment shutdown point temperature.
[0023] Preferably, the start-up and shutdown point adjustment module is further configured to:
[0024] build a proxy model of the Gaussian process regression model and a kernel function;
[0025] set a target function and a collection function of the proxy model;
[0026] input the observation space and the action space into the proxy model and perform iterative training; the target function is used to adjust the start-up and shutdown point weight of the proxy model, the collection function is used to detect the training degree of the proxy model after this round of iterative training, and the kernel function is used for data input.
[0027] Preferably, the start-up and shutdown point adjustment module is further configured to:
[0028] stop iterative training when the proxy model reaches the maximum number of iterations or the target function converges, and output the target Gaussian process regression model.
[0029] Preferably, the formula of the kernel function is:
[0030]
[0031] In the formula, x i and x j are temperature values; is input temperature point variance, is used to control the overall amplitude; l is a length scale parameter, which is used to adjust the smoothness; (x i ,x j ) 2 is the square distance between input temperature points.
[0032] Preferably, the formula of the target function is:
[0033] f(x)=w1×E+w2×ΔT+w3N;
[0034] In the formula, E is cumulative energy consumption, which is calculated by power integration; AT is interchamber temperature standard deviation; N is the number of compressor start-stop times, and the values of weights w1-w3 are determined by the analytic hierarchy process.
[0035] Preferably, the formula of the acquisition function is:
[0036] a(x) = (f * (x) - u(x)) / σ(x) * exp(-λ * c(x));
[0037] In the formula, a(x) is the acquisition function value; f * (x) is the maximum value of the current objective function; u(x) is the average prediction value of the proxy model parameter x; σ(x) * is the standard deviation of the proxy model; λ * is the control parameter; and c(x) is the single experiment time corresponding to the parameter x.
[0038] From the above, the application provides a compressor start-stop point parameter adjustment system based on Bayesian optimization, which comprises a data acquisition module configured to acquire real-time data of a compressor to be adjusted, an observation space, and an action space; the observation space and the action space are generated based on historical data of the compressor to be adjusted, and the historical data are of the same data type as the real-time data; the observation space is used to represent the environment state and the equipment state of the compressor to be adjusted, and the action space is used to represent the start-stop point of the compressor to be adjusted; a start-stop point adjustment module configured to perform model training according to the observation space, the action space, and a Gaussian process regression model to obtain a target Gaussian process regression model; the Gaussian process regression model is constructed based on a Bayesian optimization function; and a start-stop execution module configured to generate target start-stop data from the real-time data and the target Gaussian process regression model, and adjust the start-stop point of the compressor to be adjusted through the target start-stop data. The application solves the problem of low efficiency in searching for a global optimal point of a compressor by using existing compressor control technology through the above system. BRIEF DESCRIPTION OF DRAWINGS
[0039] In order to more clearly illustrate the technical solutions of the application, the drawings needed in the embodiments will be briefly introduced below. Obviously, other drawings can also be obtained by those skilled in the art without creative labor on the basis of these drawings.
[0040] Figure 1 FIG. 1 is a schematic diagram of a compressor start-stop point parameter adjustment system based on Bayesian optimization of the application;
[0041] Figure 2 A process diagram for model training in a compressor start-stop point parameter adjustment system based on Bayesian optimization is provided in the present application.
[0042] Figure 3 An observation space and action space data schematic diagram in a compressor start-stop point parameter adjustment system based on Bayesian optimization is provided in the present application. DETAILED DESCRIPTION
[0043] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.
[0044] It should be noted that the brief description of the terms in the present application is only for the convenience of understanding the subsequently described embodiments, and is not intended to limit the embodiments of the present application. Unless otherwise specified, these terms should be understood according to their ordinary and general meanings.
[0045] It should be noted that in the present application, the words such as “exemplary” or “for example” are used to represent an example, illustration or description. Any embodiment or design scheme described as “exemplary” or “for example” in the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. Rather, the words such as “exemplary” or “for example” are intended to present the relevant concept in a specific manner.
[0046] Based on the above problems, the present application provides the following embodiments.
[0047] Figure 1 A schematic diagram of a compressor start-stop point parameter adjustment system based on Bayesian optimization is provided in the present application.
[0048] Referring to Figure 1 It can be seen that the present embodiment provides a compressor start-stop point parameter adjustment system based on Bayesian optimization, which comprises:
[0049] The data acquisition module 100 is configured to acquire real-time data of a compressor to be adjusted, an observation space and an action space; the observation space and the action space are generated based on historical data of the compressor to be adjusted, and the historical data and the real-time data are of the same data type; the observation space is used to represent the environment state and the equipment state of the compressor to be adjusted, and the action space is used to represent the start-stop point of the compressor to be adjusted.
[0050] Specifically, in the embodiment, before the start-stop point adjustment of the compressor is performed, the related parameter data of the compressor need to be acquired first, and therefore the real-time data of the compressor to be adjusted, the observation space and the action space are acquired by the data acquisition module 100. The real-time data can be understood as the current data of the compressor, and the real-time observation space and the real-time action space can be understood as the parameter data of the compressor during the previous operation.
[0051] The observation space can be understood as the data observed or acquired by a sensor or the like, that is, the operation, environment and other data of the compressor to be adjusted.
[0052] The action space is the data generated by the adjustment of the compressor, and the action space can be understood as the data required for the adjustment of the compressor.
[0053] The system further comprises:
[0054] The start-stop point adjustment module 200 is configured to perform model training according to the observation space, the action space and the Gaussian process regression model to obtain a target Gaussian process regression model. The Gaussian process regression model is constructed based on a Bayesian optimization function.
[0055] Specifically, in the embodiment, unlike the existing adjustment scheme of the compressor, the target Gaussian process regression model based on the Bayesian optimization function is set to adjust the start-stop point of the compressor to be adjusted. The target Gaussian process regression model is trained by the observation space, the action space and the initially constructed Gaussian process regression model.
[0056] The system further comprises:
[0057] The start-stop execution module 300 is configured to generate target start-stop data by the real-time data and the target Gaussian process regression model, and adjust the start-stop point of the compressor to be adjusted by the target start-stop data.
[0058] Specifically, in the embodiment, when the training of the target Gaussian process regression model is completed, the start-stop point adjustment of the compressor to be adjusted is performed by the start-stop execution module 300 using the target Gaussian process regression model.
[0059] The adjustment of the start-stop point of the compressor by the start-stop execution module 300 is performed by inputting the real-time data into the target Gaussian process regression model, generating the target start-stop data for automatically adjusting the start-stop point of the compressor, and adjusting the start-stop point of the compressor to be adjusted by the target start-stop data.
[0060] Further, in some embodiments, the data acquisition module 100 is further configured to:
[0061] acquire historical data of the compressor to be adjusted;
[0062] preprocess the historical data to obtain the observation space;
[0063] construct the action space according to the start-stop specifications of the compressor to be adjusted.
[0064] Specifically, in the present embodiment, it can be known from the above that the observation space and the action space are acquired by the data acquisition module 100, and both the observation space and the action space are generated or integrated based on historical data. Therefore, the historical data of the compressor to be adjusted needs to be acquired before the observation space and the action space are acquired.
[0065] After the historical data is acquired, the observation space and the action space can be acquired. The acquisition methods of the observation space and the action space are different. For the observation data, the corresponding historical data needs to be preprocessed to obtain the observation space. For the action space, it can be directly constructed according to the start-stop specifications of the compressor to be adjusted in the historical data to obtain the action space.
[0066] Further, in some embodiments, the data acquisition module 100 comprises:
[0067] a data acquisition unit 110 configured to acquire real-time data and the historical data of the compressor to be adjusted.
[0068] Specifically, in the present embodiment, the real-time data and the historical data of the compressor to be adjusted are acquired by the data acquisition unit 110. The data acquisition unit 110 is only used for data acquisition, and the means of data acquisition can be direct input by an operator or data acquisition by a sensor or the like.
[0069] The data acquisition module 100 further comprises:
[0070] a preprocessing unit 120 configured to preprocess the historical data to obtain preprocessed data; the preprocessing includes noise reduction processing and abnormal data filtering processing.
[0071] Specifically, in the present embodiment, after the historical data is acquired, the preprocessing unit 120 is used to preprocess the historical data to filter abnormal data in the historical data and to perform noise reduction processing on the data.
[0072] It should be noted that while the historical data is preprocessed, the historical data is backed up for subsequent data processing.
[0073] The data acquisition module 100 further comprises:
[0074] The first space construction unit 130 is configured to perform first data space construction according to the preprocessed data, to obtain the observation space.
[0075] Specifically, in this embodiment, the preprocessed data is constructed by the first space construction unit 130 to integrate the preprocessed data together, thereby obtaining the multi-dimensional observation space.
[0076] Figure 3 An observation space and action space data schematic diagram in a compressor start-stop point parameter adjustment system based on Bayesian optimization.
[0077] Referring to Figure 3 It can be seen that the present embodiment is applicable to the compressor in the refrigerator, so the data in the corresponding observation space also corresponds to the related data in the refrigerator. The observation data includes indoor temperature, indoor humidity, set temperature of the refrigerator compartment, refrigerator compartment temperature, set temperature of the freezer compartment, freezer compartment temperature, freeze evaporator temperature, defrosting heater state, door opening and closing state, compressor speed, valve state, fan state, compressor state, compressor current value, and power value of the compressor.
[0078] It should be noted that according to different application scenarios, the corresponding data is also different, so the system provided by the present embodiment is not only applicable to the refrigerator. If applied to other scenarios, only the type of data needs to be adjusted.
[0079] The data acquisition module 100 further comprises:
[0080] The second space construction unit 140 is configured to perform second data space construction according to the start-stop specifications of the compressor to be adjusted, to obtain the action space.
[0081] Specifically, in this embodiment, the start-stop specifications of the compressor to be adjusted in the historical data that has not been preprocessed are constructed by the second space construction unit 140, thereby obtaining the multi-dimensional action space.
[0082] Referring to Figure 3 It can be seen that the action space includes the refrigerator compartment start point temperature, the freezer compartment start point temperature, the refrigerator compartment stop point temperature, and the freezer compartment stop point temperature.
[0083] Figure 2 Figure 1 is a process diagram of model training in a compressor start-stop point parameter adjustment system based on Bayesian optimization according to an embodiment of the present application.
[0084] Referring to Figure 2 It can be further understood that, in some embodiments, the start-stop point adjustment module 200 is further configured to:
[0085] construct a surrogate model of the Gaussian process regression model and a kernel function.
[0086] Specifically, in this embodiment, the surrogate model of the Gaussian process regression model can be understood as an untrained initial model, which is constructed based on a Bayesian optimization function. The kernel function is used for data input. The kernel function of the Gaussian process regression model is:
[0087]
[0088] In the formula, x i and x j are temperature values. is the variance of the input temperature point, is used to control the overall amplitude; l is the length scale parameter, which is used to adjust the smoothness; (x i ,x j ) 2 is the squared distance between input temperature points.
[0089] The Bayesian optimization function is a global optimization method based on Bayesian inference and Gaussian process, which is designed to handle high-cost optimization problems of target functions. The core idea is to find the global optimal solution with the least number of evaluations by constructing a probability model of the target function.
[0090] The start-stop point adjustment module 200 is further configured to:
[0091] set the target function of the surrogate model and the acquisition function.
[0092] Specifically, in this embodiment, during model training, the corresponding target function and acquisition function need to be set. The target function is used to adjust the start-stop point weight of the surrogate model, and the acquisition function is used to detect the training degree of the surrogate model after this round of iterative training.
[0093] The formula of the target function is:
[0094] f(x) = w1 x E + w2 x AT + w3 N;
[0095] In the formula, E is cumulative energy consumption, calculated by power integration; AT is interchamber temperature standard deviation; N is the number of compressor start-stop times, and the values of weights w1-w3 are determined by the analytic hierarchy process.
[0096] The objective function is a black box function (such as a loss function on a machine learning model validation set) that needs to be optimized, and can only obtain output values by evaluation, and cannot be directly analyzed.
[0097] The formula of the acquisition function is:
[0098] a(x)=(f * (x)-u(x)) / σ(x) * exp(-λ * c(x));
[0099] In the formula, a(x) is the acquisition function value; f * (x) is the maximum value of the current objective function; u(x) is the average prediction value of the proxy model parameter x; σ(x) * is the standard deviation of the proxy model; λ * is a control parameter; and c(x) is the single experiment time corresponding to the parameter x.
[0100] The proxy model usually uses Gaussian Process (GP), which describes the hypothesis of the objective function through the prior distribution, updates the posterior distribution combined with the observation data, and predicts the function value and uncertainty of the unsampled point.
[0101] The acquisition function is based on the strategy of balancing exploration (high uncertainty area) and utilization (area near the current optimal value). Common methods include:
[0102] Expected Improvement (EI): measures the expectation that a new point is better than the current optimal value.
[0103] Upper Confidence Bound (UCB): selects potential optimal points based on the confidence interval.
[0104] The start-stop point adjustment module 200 is further configured to:
[0105] Input the observation space and the action space into the proxy model and perform iterative training.
[0106] Specifically, in this embodiment, when the proxy model reaches the maximum number of iterations or the objective function converges, the iterative training is stopped, and the target Gaussian process regression model is output.
[0107] This embodiment has the following advantages:
[0108] This embodiment introduces Bayesian optimization into the laboratory refrigerator on-off point setting, forming a "data-model-decision" closed loop; based on the multi-objective function joint objective function design of energy consumption, temperature fluctuation and start-stop number, the global optimal of the refrigerator on-off point parameter is realized, and the laboratory test cost is significantly reduced.
[0109] The above description has been presented in conjunction with specific embodiments thereof. However, the foregoing discussion is not intended to be exhaustive or to limit the embodiments to the precise forms disclosed. Various modifications and changes can be made by those skilled in the art using the teachings of the above disclosure. The selection and arrangement of various elements described in the above embodiments are intended to provide a better understanding of the disclosure, and to enable those skilled in the art to better use the embodiments.
Claims
1. A compressor on-off point parameter adjustment system based on Bayesian optimization, characterized in that, The system comprises: a data acquisition module (100) configured to acquire real-time data of a compressor to be adjusted, an observation space and an action space; the observation space and the action space are generated based on historical data of the compressor to be adjusted, the historical data being of the same data type as the real-time data; the observation space is used to represent the environment state and the equipment state of the compressor to be adjusted; and the action space is used to represent the start-stop points of the compressor to be adjusted; a start-stop point adjustment module (200) configured to perform model training according to the observation space, the action space and a Gaussian process regression model to obtain a target Gaussian process regression model; the Gaussian process regression model is constructed based on a Bayesian optimization function; a start-stop execution module (300) configured to generate target start-stop data by the real-time data and the target Gaussian process regression model, and adjust the start-stop points of the compressor to be adjusted through the target start-stop data.
2. The compressor start / stop point parameter adjustment system based on Bayesian optimization of claim 1, wherein, The data acquisition module (100) is further configured to: acquire historical data of the compressor to be adjusted; preprocess the historical data to obtain the observation space; construct the action space according to the start-stop specifications of the compressor to be adjusted.
3. The compressor start / stop point parameter adjustment system based on Bayesian optimization of claim 2, wherein, The data acquisition module (100) comprises: a data acquisition unit (110) configured to acquire the real-time data and the historical data of the compressor to be adjusted; a preprocessing unit (120) configured to preprocess the historical data to obtain preprocessed data; the preprocessing includes noise reduction processing and abnormal data filtering processing; a first space construction unit (130) configured to perform first data space construction according to the preprocessed data to obtain the observation space; a second space construction unit (140) configured to perform second data space construction according to the start-stop specifications of the compressor to be adjusted to obtain the action space.
4. A Bayesian optimization based compressor on-off point parameter adjustment system according to any one of claims 1, 2 or 3, characterized in that, The system is applied to a compressor in a refrigerator.
5. The compressor start / stop point parameter adjustment system based on Bayesian optimization of claim 4, wherein, The observation space comprises indoor temperature, indoor humidity, set temperature of a refrigerator compartment, refrigerator compartment temperature, set temperature of a freezer compartment, freezer compartment temperature, freeze evaporator temperature, defrosting heater state, door opening and closing state, compressor speed, valve state, fan state, compressor state, compressor current value and compressor power value in the environment where the compressor to be adjusted is located; The action space comprises refrigerator compartment start point temperature, freezer compartment start point temperature, refrigerator compartment stop point temperature and freezer compartment stop point temperature.
6. The compressor on-off point parameter adjustment system based on Bayesian optimization of claim 1, wherein, The start-stop point adjustment module (200) is further configured to: construct a surrogate model and a kernel function of the Gaussian process regression model; set a target function and a collection function of the surrogate model; The observation space and the action space are input into the agent model and iteratively trained; the target function is used to adjust the start-stop point weight of the agent model, the collection function is used to detect the training degree of the agent model after this round of iterative training, and the kernel function is used for data input.
7. The compressor start / stop point parameter adjustment system based on Bayesian optimization of claim 6, wherein, The start-stop point adjustment module (200) is further configured to: When the agent model reaches the maximum number of iterations or the target function converges, stop the iterative training and output the target Gaussian process regression model.
8. The compressor start / stop point parameter adjustment system based on Bayesian optimization of claim 6, wherein, The formula of the kernel function is: where x i and x j are temperature values; is the variance of input temperature points, is used to control the overall amplitude; l is a length scale parameter, l is used to adjust the smoothness; (x i ,x j ) 2 is the squared distance between input temperature points.
9. The compressor start / stop point parameter adjustment system based on Bayesian optimization of claim 6, wherein, The formula of the target function is: f(x)=w1×E+w2×ΔT+w3N; In the formula, E is cumulative energy consumption, which is calculated by power integration; ΔT is interchamber temperature standard deviation; N is the number of compressor start-stops, and the values of weights w1-w3 are determined by an analytic hierarchy process.
10. The compressor start / stop point parameter adjustment system based on Bayesian optimization of claim 6, wherein, The formula of the collection function is: a(x) = (f * (x) - u(x)) / σ(x) * exp(-λ * c(x)); In the formula, a(x) is a collection function value; f * (x) is the maximum value of the current target function; u(x) is the average prediction value of the proxy model parameter x; σ(x) * is the standard deviation of the proxy model; λ * is a control parameter; and c(x) is the single experiment time corresponding to the parameter x.