A method and apparatus for controlling a refrigeration appliance

Abstract

A method and apparatus for controlling a chiller are provided, including training one or more models using historical data of the chiller and determining a model from the one or more models based on a training result, controlling an outlet chilled water temperature of the chiller based at least in part on the determined model, determining whether a retraining is to be performed, and in response to determining that the retraining is to be performed, retraining the one or more models using updated historical data of the chiller and determining an updated model from the one or more models based on a retraining result.

Description

Technical Field

[0001] In general, the present invention relates to refrigeration in engineering, and more specifically, to methods and apparatus for controlling refrigeration equipment. Background Technology

[0002] Today, refrigeration systems are widely deployed in manufacturing plants and buildings worldwide. These systems typically consume significant amounts of energy and electricity, hindering the achievement of goals such as energy conservation, emission reduction, low carbon emissions, and carbon neutrality. How to minimize energy consumption while meeting cooling needs has long been a challenging problem for the engineering community.

[0003] Thanks to the development of digitalization and sensor technology, many data-driven algorithms and digital building energy systems have emerged, making it easier to monitor, analyze, and optimize various energy consumption links in the manufacturing process, thereby revealing the parts or links that can provide energy savings.

[0004] Currently, a class of data-driven optimization methods for refrigeration systems focuses on researching accurate models based on energy behavior to control and schedule components within buildings to minimize energy consumption. Examples include global optimization architectures based on data-driven power prediction models for refrigeration equipment, and predictive control of cooling loads through building energy consumption simulation and global optimization. These methods can provide some energy savings by using power prediction models built for the main components of the refrigeration system to globally optimize the power used by each main component to minimize the total power consumption of the refrigeration system, while meeting constraints such as required cooling loads. However, the quality of the optimization results is usually highly dependent on the prediction model used and the search space.

[0005] The goal is to provide a more efficient and simple method for optimizing refrigeration systems, thereby further improving energy efficiency. Summary of the Invention

[0006] The following provides a brief overview of one or more embodiments to provide a basic understanding of these embodiments. This overview is not a generalization of all contemplated embodiments, nor is it intended to identify key or essential elements of all embodiments or to describe the scope of any or all embodiments. Its purpose is solely to provide some concepts of one or more embodiments in a simplified form as an introduction to the more detailed description provided below.

[0007] In one aspect of this disclosure, a method for controlling a refrigeration device is provided, comprising: training one or more models using historical data of the refrigeration device, and determining a model from the one or more models based on the training results; controlling the outlet chilled water temperature of the refrigeration device based at least in part on the determined model; determining whether retraining is required; and in response to determining that retraining is required, retraining the one or more models using updated historical data of the refrigeration device, and determining an updated model from the one or more models based on the retraining results.

[0008] In another aspect of this disclosure, an apparatus for controlling a refrigeration device is provided, including a memory; and at least one processor coupled to the memory and configured to perform the following operations: training one or more models using historical data of the refrigeration device, and determining a model from the one or more models based on the training results; controlling the outlet chilled water temperature of the refrigeration device based at least in part on the determined model; determining whether retraining is required; and in response to determining that retraining is required, retraining the one or more models using updated historical data of the refrigeration device, and determining an updated model from the one or more models based on the retraining results.

[0009] In another aspect of this disclosure, a computer program product for controlling a refrigeration device is provided, comprising processor-executable computer code for performing the following operations: training one or more models using historical data of the refrigeration device, and determining a model from the one or more models based on the training results; controlling the outlet chilled water temperature of the refrigeration device based at least in part on the determined model; determining whether retraining is required; and in response to determining that retraining is required, retraining the one or more models using updated historical data of the refrigeration device, and determining an updated model from the one or more models based on the retraining results.

[0010] In another aspect of this disclosure, a computer-readable medium is provided storing computer code for controlling a refrigeration device, the computer code, when executed by a processor, causing the processor to perform the following operations: training one or more models using historical data from the refrigeration device, and determining a model from the one or more models based on the training results; controlling the outlet chilled water temperature of the refrigeration device based at least in part on the determined model; determining whether retraining is required; and in response to determining that retraining is required, retraining the one or more models using updated historical data from the refrigeration device, and determining an updated model from the one or more models based on the retraining results.

[0011] Other aspects or variations of this disclosure will become clearer when the following detailed description and accompanying drawings are taken into consideration. Attached Figure Description

[0012] Figure 1 An exemplary block diagram of a refrigeration system 100 according to one or more aspects of this disclosure is shown;

[0013] Figure 2 This is a flowchart illustrating an exemplary method 200 for controlling a refrigeration device according to one or more aspects of this disclosure;

[0014] Figure 3 This is a flowchart illustrating an exemplary method 300 for training one or more models according to one or more aspects of this disclosure;

[0015] Figure 4 This is a flowchart illustrating an exemplary method 400 for controlling the outlet chilled water temperature of a refrigeration unit based at least in part on a determined model, according to one or more aspects of this disclosure.

[0016] Figure 5 This is a flowchart illustrating an exemplary method 500 for determining whether to retrain, according to one or more aspects of this disclosure;

[0017] Figure 6 An example of a hardware implementation of a device 600 for controlling a refrigeration equipment according to one or more aspects of this disclosure is shown. Detailed Implementation

[0018] Several embodiments will now be described with reference to the accompanying drawings, wherein like elements are indicated herein by the same reference numerals. In the following description, numerous specific details are set forth for ease of explanation in order to provide a thorough understanding of one or more embodiments. However, it will be apparent that the embodiments may also be implemented without these specific details. In other instances, well-known structures and devices are shown in block diagram form to facilitate the description of one or more embodiments.

[0019] Figure 1An exemplary block diagram of a refrigeration system 100 is shown. The refrigeration system 100 may include a chiller 102, a cooling tower 106 in an external circulation 101, and an indoor air cooling device 104 in an internal circulation 103. The internal circulation 103 may push chilled water from the chiller 102 into the interior spaces of a building for air cooling. The chilled water may absorb heat from the building and return to the chiller 102 at a higher temperature. The external circulation 101 may push condensate from the chiller 102 to the outdoor cooling tower 106, which releases heat to the outdoor environment, thereby allowing the chilled water to be returned to the chiller 102. The external circulation 101 may operate independently of the internal circulation 103. Heat exchange between the internal circulation 103 and the external circulation 101 may be performed by the chiller 102.

[0020] In one aspect of this disclosure, the refrigeration system 100 may further include one or more sensors and a database for collecting and maintaining monitoring data from the one or more sensors. The one or more sensors may be used to monitor one or more parameters of the chiller 102, including the chilled water outlet temperature at the outlet of the chiller 102, the return water temperature at the return outlet of the chiller 102, and the chilled water flow rate and / or velocity in the internal circulation 103 of the refrigeration system 100. In another aspect of this disclosure, the refrigeration system 100 may further include and / or utilize a centralized data platform.

[0021] In other aspects of this disclosure, the refrigeration system 100 may also include one or more additional or alternative components.

[0022] This disclosure provides a method for controlling refrigeration equipment. This method uses a data-driven model to control one or more parameters of the refrigeration equipment. The model used for control is not fixed and can be varied according to changes in the external environment (e.g., external temperature) and / or the indoor environment (e.g., production line layout). Compared to existing techniques and methods that typically involve building power prediction models for the major components of the refrigeration system or simulating building energy consumption performance based on global optimization, the method proposed in this disclosure omits the complex process of building such a power prediction model. Instead, it uses a predictive model of the cooling load, and the variability of this model improves adaptability to the environment, thereby further enhancing the energy-saving effect on the refrigeration system.

[0023] Figure 2 This is a flowchart illustrating an exemplary method 200 for controlling a refrigeration device according to one or more aspects of this disclosure. It can be seen, for example, in... Figure 1Method 200 is implemented in the refrigeration system 100 shown. At step 202, one or more models are trained using historical data from the refrigeration equipment to determine a model from among the one or more models. At step 204, the outlet chilled water temperature of the refrigeration equipment is controlled, at least in part, based on the model determined at step 202. In one aspect of this disclosure, the control of the refrigeration equipment based on the model determined at step 202 is real-time control and requires no human intervention. At step 206, it is determined whether retraining is required, and in response to the determination that retraining is required, the process returns to step 202, where the one or more models are retrained using updated historical data from the refrigeration equipment to determine an updated model from among the one or more models; or, in response to the determination that retraining is not required, the process returns to step 204, where the determined model continues to be used to control the refrigeration equipment.

[0024] In one aspect of this disclosure, the refrigeration device in method 200 may include, for example: Figure 1 The refrigerator 102 is shown in the diagram. In another aspect of this disclosure, historical data may include external temperatures for one or more time periods and the actual cooling load corresponding to those time periods. For example, historical data may include hourly external temperature values ​​recorded for a previous month and hourly cooling loads actually provided by the refrigeration equipment for that month. In one aspect of this disclosure, the actual cooling load provided by the refrigeration equipment may be derived based on measurements monitored by one or more sensors deployed in the refrigeration system 100, for example, based on one or more of the following: the temperature of the chilled water at the outlet of the refrigerator 102, the return water temperature at the return outlet of the refrigerator 102, and the chilled water flow rate and / or velocity in the internal circulation 103 of the refrigeration system 100. In one or more aspects of this disclosure, historical data may be updated monthly, for example, once a month. In other aspects of this disclosure, historical data may be updated at periods longer or shorter than a month, or may be updated irregularly. In another aspect of this disclosure, the data used in method 200 (e.g., historical data on actual cooling load and external temperature, etc.) may be based on a centralized data platform included in and / or coupled to the refrigeration system 100.

[0025] Figure 3This is a flowchart illustrating an exemplary method 300 for training one or more models according to one or more aspects of this disclosure. For example, method 200 may perform method 300 at step 202. At step 302, data received from a database, such as a refrigeration system 100, is preprocessed to generate historical data. In one example, the preprocessing may include downsampling measurements from one or more sensors collected in the database to obtain downsampled measurements. In another example, the preprocessing may include format conversion of measurements from one or more sensors collected in the database. In one aspect of this disclosure, preprocessing enables the generation of historical data using measurements from one or more sensors having the same sampling rate and / or format.

[0026] At step 304, one or more models are trained using the historical data generated at step 302. This historical data can be used as a training set and / or a test set for training the models. In one aspect of this disclosure, the external temperature and actual cooling load of the previous time period can be used as input features of the model, and the cooling load predicted by the model for the next time period can be used as the model's output. In one example, one or more models may include support vector regression, stochastic regression prediction, decision tree regression, ridge regression, Gaussian process regression, linear regression, Adaboost regression, gradient boosting regression, etc. In another aspect of this disclosure, one or more models can be trained and evaluated based on time series cross-validation.

[0027] At step 306, based on the training results obtained in step 304, a model is determined from one or more models. For example, the model with the smallest error can be selected from one or more models. In one example, the metric used to measure the error may include the mean absolute error percentage (MAPE), which is expressed as:

[0028]

[0029] Among them, y i This is the actual value. This is the prediction result.

[0030] In one or more aspects of this disclosure, method 300 can also be used to determine the updated model. In one example, determining the updated model may include selecting a model different from the current model, for example, updating from the current support vector regression model to a stochastic regression prediction model. In another example, determining the updated model may include updating the model parameters without updating the model type (e.g., still using the current support vector regression model).

[0031] Figure 4This is a flowchart illustrating an exemplary method 400 for controlling the outlet chilled water temperature of a refrigeration unit based at least in part on a determined model, according to one or more aspects of this disclosure. Method 400 can be performed at step 204 of method 200. At step 402, the determined model uses the external temperature and actual cooling load of the previous time period as input to predict and output the cooling load for the next time period. In one aspect of this disclosure, the model used in step 402 can be determined by method 300. In another aspect of this disclosure, the cooling load that the refrigeration unit 102 will provide in the next hour can be predicted at the current time using the external temperature of the previous hour and the actual cooling load provided by the refrigeration unit 102.

[0032] At step 404, based on a comparison between the predicted cooling load for the next time period and the actual cooling load for the next time period, a control command is generated to perform fuzzy control on the setpoint of the outlet chilled water temperature of refrigeration equipment such as chiller 102. In one aspect of this disclosure, it is desirable to maintain a certain amount of cooling load in the refrigeration system 100 to meet the cooling needs of manufacturing plants and / or buildings. However, as factors such as external temperature change, the amount of power required to provide the same cooling load may vary, providing opportunities for power savings and improved energy efficiency. For example, when the cooling load for the next period, predicted by the determined model based on the external temperature and actual cooling load of the previous period, is greater than the actual cooling load of the next period, a control command is generated to increase the setpoint of the outlet chilled water temperature of the chiller 102 in the next period after that, in order to save unnecessary power consumption; or, when the cooling load for the next period, predicted by the determined model based on the external temperature and actual cooling load of the previous period, is less than the actual cooling load of the next period, a control command is generated to decrease the setpoint of the outlet chilled water temperature of the chiller 102 in the next period after that, in order to ensure that the cooling demand can be met.

[0033] At step 406, based on the control command generated at step 404, the setpoint for the outlet chilled water temperature of the chiller 102 in the next time period is adjusted. For example, the outlet chilled water temperature of the chiller 102 in the next time period can be adjusted by a programmable logic controller (PLC) for adjusting the setpoint of the outlet chilled water temperature of the chiller 102. In one example, the outlet chilled water temperature of the chiller 102 in the next hour can be increased or decreased in steps of 0.25°C (degrees Celsius) according to the control command. In one aspect of this disclosure, the outlet chilled water temperature of the chiller 102 can be initially set at a predetermined setpoint, and this temperature can be adjusted within a certain range (e.g., 8°C to 12°C).

[0034] At step 408, the actual cooling load for each time period is calculated using measurements monitored by one or more sensors deployed in the refrigeration system 100, and the calculated actual cooling load is used as input for steps 402 and 404 in the next cycle for the next time period.

[0035] In one aspect of this disclosure, steps 402 to 408 can be repeated at a predetermined period (e.g., hourly) to achieve real-time control of the chiller 102. In other aspects of this disclosure, steps 402 to 408 can also be repeated in a non-periodic manner.

[0036] Figure 5 This is a flowchart illustrating an exemplary method 500 for determining whether retraining should be performed, according to one or more aspects of this disclosure. Method 500 can be performed at step 206 of method 200. At step 502, the error between the cooling load predicted by the determined model for one or more time periods and the actual cooling load corresponding to the one or more time periods is calculated. In one example, the one or more cooling load predictions for one or more time periods (e.g., hourly) at step 402 of method 400 are compared with one or more actual cooling load values ​​measured by sensors for the one or more time periods at step 408 of method 400 to calculate the error between the model prediction and the actual measurement. This error can be represented by the mean absolute error percentage (MAPE). In one example, this error can be calculated periodically, for example, monthly. In another example, the error can also be calculated non-periodicly.

[0037] At step 504, the calculated error is compared with an error threshold (e.g., 5%) to determine whether retraining is necessary. For example, if the calculated error is greater than the error threshold, retraining is determined, and if the calculated error is less than or equal to the error threshold, retraining is determined not to be performed, and the model determined by method 300 is continued.

[0038] Assume, using Figure 4The method 400 shown derives, via step 408, the actual cooling load for period k-1 as 1 refrigeration ton (RT) and the actual cooling load for period k as 0.8 refrigeration ton (RT), and uses these as inputs to steps 402 and 404 in the next cycle (i.e., for period k+1). At step 402, based on the actual cooling load of 1 refrigeration ton (RT) for period k-1 and the external temperature for period k-1, the determined model predicts the cooling load for period k as 1.1 refrigeration ton (RT). At step 404, since the predicted 1.1 refrigeration ton (RT) for period k is greater than the actual cooling load of 0.8 refrigeration ton (RT) for period k, a control command is generated to increase the outlet chilled water temperature. At step 406, according to this control command, the outlet chilled water temperature for period k+1 is increased in steps of 0.25°C. At step 408, the actual cooling load for period k+1, measured by the sensor, is 0.7 refrigeration tons (RT), and 0.8 refrigeration tons (RT) and 0.7 refrigeration tons (RT) are used as inputs to steps 402 and 404 in the next cycle (i.e., for period k+2). At step 402, based on 0.8 refrigeration tons (RT) and the external temperature for period k, the cooling load for period k+1 is predicted to be 1.2 refrigeration tons (RT) using the determined model. At step 404, since the predicted 1.2 refrigeration tons (RT) for period k+1 is greater than the actual cooling load of 0.7 refrigeration tons (RT) for period k+1, a control command is generated to increase the outlet chilled water temperature. At step 406, according to this control command, the outlet chilled water temperature for period k+2 is increased in steps of 0.25°C. At step 408, the actual cooling load for period k+2, measured by the sensor, is 0.6 refrigeration tons (RT). Returning to step 402, based on the actual cooling load of 0.7 tons of refrigeration (RT) for period k+1 and the external temperature for period k+1, the cooling load for period k+2 is predicted to be 1.3 tons of refrigeration (RT) using the determined model. The above example can be represented by Table 1.

[0039] Table 1

[0040] Forecast value of cooling load 1.1 1.2 1.3 Actual value of cooling load 1 0.8 0.7 0.6

[0041] As shown in Table 1, the relative error percentage between the actual value (0.8) and the predicted value (1.1) for time period k is approximately 38% (i.e., (1.1-0.8) / 0.8), and the relative error percentage between the actual value (0.7) and the predicted value (1.2) for time period k+1 is 71% (i.e., (1.2-0.7) / 0.7). The errors for both time periods exceed the error threshold (e.g., 5%). If the average relative error percentage exceeds the error threshold over a certain period (e.g., one month), it may mean that the currently used model has a large error in predicting changes in factors such as external temperature, or is no longer applicable to the changing environment. For example, in the example in Table 1, a possible scenario is that the external temperature rises rapidly, requiring the chilled water temperature to continuously decrease to provide the required cooling load. However, due to the inadequacy of the currently used model, the outlet chilled water temperature even increases in time periods k+1 and k+2. Therefore, as... Figure 5 Method 500 describes a method that determines that retraining is necessary to identify a more suitable model when the calculated average error MAPE is greater than an error threshold.

[0042] In one aspect of this disclosure, according to the example in Table 1, method 300 may use updated historical data, such as actual cooling loads of 1, 0.8, 0.7, and 0.6 RT for periods k-1, k, k+1, and k+2, and corresponding external temperatures for these periods, to retrain one or more models such that the updated model is better adapted to recent environmental changes compared to the currently used model (which has larger errors between its predicted and actual values, e.g., 38% and 71%).

[0043] According to one or more aspects of this disclosure, by performing one or more of methods 200, 300, 400, and / or 500, predictions of cooling load can be provided using historical data specific to the refrigeration equipment and external temperatures, and the model can be adapted to changing environments by retraining the model.

[0044] Figure 6 An example of a hardware implementation of an apparatus 600 for controlling a refrigeration device according to one or more aspects of this disclosure is shown. The apparatus 600 for controlling the refrigeration device may include a memory 610 and at least one processor 620. The processor 620 may be coupled to the memory 610 and configured to perform the above-described references. Figure 2 , Figure 3 , Figure 4 and Figure 5One or more of the described methods 200, 300, 400, and / or 500. Processor 620 may be a general-purpose processor, or it may be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors combined with a DSP core, or other such structures. Memory 610 may store input data, output data, data generated by processor 620, and / or instructions executed by processor 620.

[0045] The various operations, models, and networks described in conjunction with this disclosure can be implemented as hardware, processor-executed software, firmware, or any combination thereof. According to one or more aspects of this disclosure, a computer program product for controlling a refrigeration device may include programs for performing the aforementioned operations. Figure 2 , Figure 3 , Figure 4 and Figure 5 The processor-executable computer code of one or more of the described methods 200, 300, 400, and / or 500. According to other aspects of this disclosure, a computer-readable medium can store computer code for controlling a refrigeration device, which, when executed by a processor, causes the processor to perform the aforementioned described code. Figure 2 , Figure 3 , Figure 4 and Figure 5 One or more of the described methods 200, 300, 400, and / or 500. Computer-readable media includes both non-transitory computer storage media and communication media, with communication media including any medium that facilitates the transfer of a computer program from one location to another. Any connection may be appropriately referred to as computer-readable media.

[0046] The above description of the present disclosure is provided to enable those skilled in the art to use or implement various embodiments. Various modifications to the above embodiments will be apparent to those skilled in the art, and the basic principles defined herein can be applied to other embodiments without departing from the scope of this disclosure. Therefore, the scope of the claims is not intended to be limited to the embodiments disclosed herein, but is to be given the broadest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for controlling a refrigeration device, comprising: A model is determined from multiple models, wherein the multiple models are trained on historical data including external temperature and the actual cooling load of the refrigeration equipment, and the determined model is derived from the multiple models based on the training results; and The outlet chilled water temperature of the refrigeration equipment is controlled, at least in part, based on the determined model. The control further includes: Provide the determined model with the current external temperature and actual cooling load to obtain the cooling load for the next period predicted by the determined model; Compare the predicted cooling load for the next period with the actual cooling load for the next period; When the predicted cooling load for the next period is greater than the actual cooling load for the next period, the set point of the outlet chilled water temperature of the refrigeration equipment for the next period is increased by one step. When the predicted cooling load for the next time period is less than the actual cooling load for the next time period, the set point of the outlet chilled water temperature of the refrigeration equipment for the next time period is reduced by one step.

2. The method according to claim 1, wherein, The historical data includes external temperatures for one or more time periods and actual cooling loads corresponding to the one or more time periods, and wherein the updated historical data includes external temperatures for one or more time periods following the one or more time periods and actual cooling loads corresponding to the one or more time periods following the one or more time periods.

3. The method according to claim 1, wherein: Based on the error between the cooling load predicted by the determined model for one or more time periods and the actual cooling load corresponding to the one or more time periods, it is determined whether to retrain.

4. The method according to claim 3, wherein, The error includes the mean absolute percentage error (MAPE) between the cooling load predicted by the determined model for one or more time periods and the actual cooling load corresponding to the one or more time periods. And among them, When the MAPE is greater than a threshold, it is determined that retraining is required; when the MAPE is less than or equal to the threshold, it is determined that retraining is not required, and the determined model can continue to be used.

5. The method according to claim 1, wherein, The multiple models include several of the following: Support vector regression, stochastic regression prediction, decision tree regression, ridge regression, Gaussian process regression, linear regression, Adaboost regression, and gradient boosting regression.

6. An apparatus for controlling a refrigeration device, comprising: Memory; as well as At least one processor coupled to the memory and configured to perform the method of any one of claims 1 to 5.

7. A computer program product for controlling a refrigeration device, comprising processor-executable computer code for performing the method of any one of claims 1 to 5.

8. A computer-readable medium storing computer code for controlling a refrigeration device, said computer code, when executed by a processor, causing the processor to perform the method of any one of claims 1 to 5.

Images