Online sensing method for running state of water chilling unit
By training an energy efficiency prediction degradation model with real-time running data, and combining a long short-term memory neural network and an attention mechanism, the problem of accuracy in assessing the performance degradation of chiller units has been solved. This enables online sensing and graded assessment of chiller unit performance, reducing energy consumption and improving equipment stability.
Patent Information
- Application Number
- CN202511108297.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-08
- Publication Date
- 2025-11-14
AI Technical Summary
Existing methods for diagnosing chiller unit faults rely on health data to train evaluation models, which leads to discrepancies between real-time data and health data. This results in an inability to accurately reflect the actual operating status and performance degradation of the unit, leading to increased energy consumption and equipment stability risks.
An energy efficiency prediction degradation model is trained using real-time operating data of chiller units. Through long short-term memory artificial neural networks and attention mechanisms, combined with a standard database, backtesting is performed to predict energy efficiency values and calculate deviations, thereby achieving online perception and graded assessment of the performance status of chiller units.
It improves the accuracy of fault diagnosis for chiller units, can reflect performance degradation trends across time scales, reduces energy consumption costs, and requires no additional hardware resources, making it suitable for chiller units and intelligent building control systems.
Smart Images
Figure CN120950918A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the technical field of chiller units, and specifically relates to an online sensing method for the operating status of chiller units. Background Technology
[0002] Chillers, as highly efficient heat and cold source devices, are widely used in temperature control systems of various buildings, industrial processes, and special environments. With their excellent cooling efficiency and system controllability, chillers have become an indispensable key component in central air conditioning systems.
[0003] With the rapid development of information technology, the demand for environmental temperature control in data centers has increased dramatically. The dense deployment of servers within data centers generates a significant amount of heat during operation. Failure to manage this heat effectively and promptly can lead to decreased server performance, increased failure rates, and even systemic downtime. In practical applications, chiller units operate continuously for extended periods, and over time, internal components inevitably experience wear, aging, and fouling. For example, compressor bearings wear under prolonged high-load operation, leading to decreased efficiency; condensers and evaporators easily accumulate dirt or scale, reducing heat exchange efficiency; refrigerant leaks and valve wear also weaken the refrigeration cycle performance. These internal degradation phenomena often do not occur drastically in the short term but accumulate gradually, resulting in hidden performance decline.
[0004] A direct manifestation of internal degradation in chiller units is a decline in the Coefficient of Performance (COP), leading to increased energy consumption across the entire cooling system. This is particularly problematic in energy-efficient environments like data centers, where the hidden deterioration of chiller operating conditions not only increases energy costs but can also cause insufficient cooling capacity in certain areas, impacting the stability of IT equipment. Traditional maintenance methods typically rely on periodic inspections or manual checks, lacking real-time awareness and dynamic assessment of the chiller's actual operating status. This can easily result in undetected equipment malfunctions, over- or delayed maintenance, and wasted maintenance resources.
[0005] Current condition monitoring or fault diagnosis methods use health data as a training database. This results in a healthy model being trained. When backtesting with real-time data, the discrepancy between real-time and health data leads to a degraded model that cannot accurately reflect the unit's true operating status and its deviation from the healthy state. During diagnosis, introducing real-time data into the trained health model creates an inherent bias between the training data, the method / model, and the real-time data. This is an inherent flaw that current diagnostic methods cannot overcome at the model level, resulting in subsequent monitoring results that cannot accurately reflect the actual state. Summary of the Invention
[0006] The purpose of this invention is to provide an online sensing method for the operating status of chiller units, addressing the aforementioned problems. This method aims to improve upon existing chiller unit fault diagnosis methods that rely on health data to train evaluation models, which fail to intuitively and accurately reflect the performance degradation of the chiller unit relative to its healthy state.
[0007] The technical solution adopted in this invention is as follows: A method for online sensing of the operating status of a chiller unit, the method specifically includes the following: Obtain real-time operating data and real-time energy efficiency values of the chiller system; Align the time series of the data, extract the feature parameters that are highly correlated with the real-time energy efficiency value, and select the operating data that is longer than the set time span threshold. The feature parameters are normalized, and a long short-term memory artificial neural network with an attention mechanism is trained to predict the energy efficiency degradation model, which then predicts the energy efficiency value through the feature parameters. The standard data that meets the current characteristic data conditions in the standard database is called and substituted into the energy efficiency prediction degradation model for backtesting. The steady-state predicted energy efficiency value is selected, and the deviation between the predicted energy efficiency value and the actual energy efficiency value of the standard data is calculated. The performance degradation of the chiller system is graded based on the deviation, the performance status of the chiller system is obtained, and corresponding operations are performed.
[0008] It should be noted that the energy efficiency prediction degradation model is trained using measured data from operating units, i.e., data that has already shown degradation. Hence, it is called the energy efficiency prediction degradation model.
[0009] This method uses actual operating data of the chiller as the degradation dataset, trains an energy efficiency prediction degradation model using the degradation dataset, and then uses healthy data as the backtesting dataset for backtesting. The healthy data serves as the standard data, and is derived from data from healthy chiller units. This overcomes the shortcoming of traditional models that rely on healthy data for training and cannot accurately assess the degradation of the unit's true energy efficiency value relative to its healthy state.
[0010] Furthermore, the real-time energy efficiency value of the chiller system is taken as the Coefficient of Performance (COP) value: ; ; in, This refers to the heat absorbed by the chiller from the heat source or the heat dissipated by the chiller to the external environment. This is the sum of all input electrical power of the chiller system. The specific heat capacity of water, The water flow rate of the evaporator or condenser; This refers to the temperature difference between the inlet and outlet water of the evaporator or the temperature difference between the inlet and outlet water of the condenser.
[0011] Furthermore, time series alignment of the data includes the following: Based on the time step of data acquisition according to different operating parameters, align the data timestamps and unify the time step; Outlier data in the time series is removed, and the z-score method with a sliding window is used. The sliding window time step is 5-15, and the z-score is calculated as follows: ; in, for Time data, for The mean of the window before time step [time]. Let be the standard deviation of the window before time , if ,but If the time-based data contains outliers, these outliers will be removed. To fill in outliers, linear interpolation is used to calculate new data to fill in the removed outliers. The calculation formula is as follows: ; in, For a moment, for Time data.
[0012] Furthermore, the Pearson correlation coefficient method is used to extract feature parameters that are highly correlated with real-time energy efficiency values, the correlation coefficient between each operating parameter and the energy efficiency value is calculated, and the parameters that are highly correlated with the energy efficiency value are extracted as feature parameters. Using operating parameters as independent variables and energy efficiency value as the dependent variable, the Pearson correlation coefficient between the independent and dependent variables is calculated as follows: ; in, As the independent variable, For the observed values of the dependent variable, The mean of the independent variable. Let the mean of the dependent variable be... If the absolute value of the Pearson correlation coefficient is taken as the independent variable, then the high correlation characteristic parameter is taken.
[0013] Furthermore, the feature parameters are normalized to obtain the normalized parameters. The formula for its calculation is: ; in, The maximum value of the characteristic parameter. It represents the minimum value of the characteristic parameter.
[0014] Furthermore, the normalized feature parameters The sequence is fed into the input layer of the neural network as the input sequence of the Long Short-Term Memory (LSTM) neural network. First, it is processed by the LSTM layer to match the time series data. Then, the output of the LSTM is input into the attention layer, which uses the attention mechanism to uncover high-dimensional correlations in the time series data. Finally, the data processed by the attention mechanism is... The input data is processed by the softmax function in the fully connected layer. The summation is performed, and the final predicted energy efficiency value is output by the output layer. An energy efficiency prediction degradation model was obtained.
[0015] Furthermore, the standard database includes the following data: preset standard operating characteristic data or early operating data of the chiller unit.
[0016] Furthermore, standard data from the standard database that conforms to the current characteristic data operating conditions is called and substituted into the energy efficiency prediction degradation model for backtesting to predict the energy efficiency value of the standard data and obtain the corresponding predicted energy efficiency value. The energy efficiency prediction degradation model is used to predict the energy efficiency value of the healthy unit under the corresponding operating conditions, select the predicted energy efficiency value of the stable state, and calculate the deviation between the predicted energy efficiency value and the actual energy efficiency value of the standard data.
[0017] It should be noted that the energy efficiency value of the standard data under the corresponding operating conditions is the same as the energy efficiency value of the standard data that conforms to the current characteristic data operating conditions.
[0018] Furthermore, a visual aid is used to manually select a stationary data segment as the predicted energy efficiency value for the steady state, and the mean relative error (MRE) of the stationary data segment is used as the deviation. ; in, The standard data is the actual COP value. The standard data is substituted into the COP of the energy efficiency prediction degradation model to predict the energy efficiency value.
[0019] Furthermore, the performance status of the chiller system includes optimal state, deviating from optimal state, deviating from stable state, and failure state. When the performance status of the chiller system is not in the optimal state, the operator will be notified through the chiller monitoring terminal or corresponding prompting equipment.
[0020] It should be noted that the chiller unit monitoring terminal or corresponding prompting device is an existing terminal or device.
[0021] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. This invention uses an energy efficiency prediction degradation model trained with real-time operating data of chiller units. This model overcomes the shortcomings of traditional models that rely on health data for training and cannot accurately assess the degradation of the unit's true energy efficiency value compared to its healthy state. At the same time, the backtested data is a fixed dataset, which can effectively reflect the performance degradation trend of chiller units across time scales. This improves the accuracy of chiller unit fault diagnosis methods. 2. The standard database of this invention can be provided by the chiller manufacturer or a small amount of healthy operating status data can be collected in the standard database during the initial operation of the chiller. This allows the method to run without the need to collect corresponding fault datasets for different fault types, thus achieving full-cycle monitoring and performance perception of the chiller. 3. The method of the present invention can be directly integrated into a chiller unit or an intelligent building control system. It utilizes existing hardware and software resources, without the need to add other computing power equipment, and can complete the training of the backtest model with only a small amount of computing power and time.
[0022] 4. The dataset of this invention uses the actual operating data of the chiller as the degradation dataset, trains the energy efficiency prediction degradation model using the degradation dataset, and then uses the healthy data as the backtesting dataset for backtesting, and then compares the results. This avoids the discrepancy between the actual operating data and the healthy data, and improves the accuracy of the method. Attached Figure Description
[0023] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a diagram of the energy efficiency prediction degradation model of the long short-term memory artificial neural network with an attention mechanism introduced in this invention. Figure 3 The chart shows a comparison of backtesting results when the standard data of this invention is substituted into the energy efficiency prediction degradation model. Detailed Implementation
[0024] The present invention will now be described in detail with reference to the accompanying drawings.
[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.
[0026] In existing technologies, common chiller unit fault detection techniques can usually only diagnose faults after they occur, and cannot perceive the degradation of the chiller unit in real time or the impact of the fault on the chiller unit's performance in the early stages. In addition, the fixed energy efficiency range and static ratio judgment method has shortcomings such as limited adaptability, coarse diagnostic granularity, difficulty in quantifying the degree of degradation, and difficulty in accurately judging when the unit is operating under conditions exceeding those included in the model.
[0027] Traditional methods of training health models using health data cannot accurately reflect the actual operating status of the unit and the deviation from the healthy state when backtesting with real-time data. This is because real-time data is biased compared to health data; therefore, degradation models trained using real-time data cannot accurately reflect the actual operating status of the unit and the deviation from the healthy state.
[0028] Therefore, in order to solve the above-mentioned technical problems, this invention proposes an online sensing method for the operating status of chiller units, which can realize the sensing and prediction of the real-time operating energy efficiency of chiller units, and further classify the system degradation, so as to realize the early detection and warning of the performance degradation of chiller units.
[0029] It should be noted that the following embodiments and specific technical features are merely illustrative examples of the present invention and are not intended to limit the scope of protection of the present invention. The core concept of the present invention lies in using an energy efficiency prediction degradation model trained with real-time operating data of the chiller unit. This overcomes the shortcomings of traditional models that rely on healthy data for training and cannot accurately assess the degradation of the unit's true energy efficiency value relative to its healthy state. Furthermore, the backtested data is a fixed dataset, which can effectively reflect the performance degradation trend of the chiller unit across time scales, thereby improving the accuracy of chiller unit fault diagnosis methods. Based on this concept, the indicators predicted by the training model of the present invention can be not only COP but also other general refrigeration system performance indicators. The system and method of the present invention are not limited to chiller units but are also applicable to other refrigeration systems, such as building HVAC and data center cooling. Any aspect involving reduced energy efficiency ratio and system heat dissipation falls within the scope of protection of the present invention.
[0030] Example 1 like Figure 1 As shown, one embodiment of the present invention is a method for online sensing of the operating status of a chiller unit, characterized in that the method specifically includes the following: Step S100: Obtain real-time operating data and real-time energy efficiency values of the chiller system; The real-time operating data of the chiller system includes the operating parameters of components such as the chiller compressor, heat exchanger, throttling device, and lubrication device. The specific operating parameters are shown in Table 1.
[0031] Table 1 Summary of Operating Parameters
[0032] Step S200: Align the time series of the data, extract the feature parameters that are highly correlated with the real-time energy efficiency value, and select the operating data that is greater than the set time span threshold; Among them, the time span threshold Typically, there are more than 500 time steps; Step S300: Normalize the feature parameters, train the long short-term memory artificial neural network energy efficiency prediction degradation model with attention mechanism, and predict the energy efficiency value through the feature parameters; Step S400: Call the standard data in the standard database that matches the current characteristic data conditions and substitute it into the energy efficiency prediction degradation model for backtesting. Select the steady-state predicted energy efficiency value and calculate the deviation between the predicted energy efficiency value and the actual energy efficiency value of the standard data. Step S500: Based on the deviation, assess the performance degradation of the chiller system to obtain the performance status of the chiller system, and perform the corresponding operations.
[0033] This embodiment acquires real-time operating data and real-time energy efficiency values of the chiller system, extracts feature parameters, and uses these parameters to predict energy efficiency values using an energy efficiency prediction degradation model. Then, it backtests the model using standard data to further predict energy efficiency values, selecting the predicted energy efficiency value for a stable state. Based on the deviation between the predicted energy efficiency value and the actual energy efficiency value from the standard data, a system performance degradation grading assessment is obtained. This solves the problem that existing models trained on healthy data cannot intuitively and accurately reflect the performance degradation of the chiller unit relative to its healthy state, and uses backtesting data as a fixed dataset to effectively reflect the performance degradation trend of the chiller unit across time scales.
[0034] Example 2 In another embodiment of the present invention, the real-time energy efficiency value of the chiller system is taken as the coefficient of performance (COP) value: ; ; in, This refers to the heat absorbed by the chiller from the heat source or the heat dissipated by the chiller to the external environment. This is the sum of all input electrical power of the chiller system. The specific heat capacity of water, The water flow rate of the evaporator or condenser; This refers to the temperature difference between the inlet and outlet water of the evaporator or the temperature difference between the inlet and outlet water of the condenser.
[0035] Generally, the heat absorbed by the chiller from the heat source or the heat dissipated by the chiller to the external environment can be calculated based on the inlet and outlet water flow rates and temperature differences of the chiller's evaporator or condenser. (This embodiment...) Using condenser water flow rate FWC, The condenser inlet and outlet water temperature difference TCO-TCI is adopted.
[0036] Example 3 Another embodiment of the present invention includes the following: Time series alignment of data. Based on the time step of data acquisition according to different operating parameters, align the data timestamps and unify the time step; Outlier data in the time series are removed, and the z-score method with a sliding window is used. The sliding window has 5 time steps, and the z-score is calculated as follows: ; in, for Time data, for The mean of the window before time step [time]. Let be the standard deviation of the window before time , if ,but If the time-based data contains outliers, these outliers will be removed. Outliers are filled in using linear interpolation to calculate new data to fill in the removed outliers or missing values. The formula is as follows: ; in, For a moment, for Time data.
[0037] Example 4 Another embodiment of the present invention is that, after the time series of data is aligned, the Pearson correlation coefficient method is used to extract feature parameters that are highly correlated with the real-time energy efficiency value, the correlation coefficient between each operating parameter and the energy efficiency value is calculated, and the parameters that are highly correlated with the energy efficiency value are extracted as feature parameters. Using operating parameters as independent variables and energy efficiency value as the dependent variable, the Pearson correlation coefficient between the independent and dependent variables is calculated as follows: ; in, As the independent variable, For the observed values of the dependent variable, The mean of the independent variable. The absolute value of the correlation coefficient is taken as the mean of the dependent variable. The independent variables were selected as highly correlated feature parameters, and the filtered feature parameters are shown in Table 2.
[0038] Table 2. Feature parameters and correlation coefficients after screening
[0039] Set time span threshold For 500 time steps, based on Extract the required runtime data.
[0040] Example 5 Another embodiment of the present invention involves selecting data from real-time online datasets that meet a time span threshold. The data is processed by normalizing the feature parameters to obtain the normalized parameters. The formula for its calculation is: ; in, The maximum value of the characteristic parameter. It represents the minimum value of the characteristic parameter.
[0041] like Figure 2 As shown, the normalized feature parameters The sequence is fed into the input layer of the neural network as the input sequence of the Long Short-Term Memory (LSTM) neural network. First, it is processed by the LSTM layer to match the time series data. Then, the output of the LSTM is input into the attention layer, which uses the attention mechanism to uncover the high-dimensional correlations in the time series data. Finally, the data processed by the attention mechanism is... The input data is processed by the softmax function in the fully connected layer. The summation is performed, and the final predicted energy efficiency value is output by the output layer. An energy efficiency prediction degradation model was obtained.
[0042] By incorporating normalized parameters into a long-short-term artificial neural network energy efficiency prediction degradation model based on an attention mechanism, accurate assessments can be made under out-of-range operating conditions based on real-time operating data of the current chiller system.
[0043] Example 6 Another embodiment of the present invention includes data backtesting as follows: The standard data that meets the current characteristic data conditions in the standard database is called and substituted into the energy efficiency prediction degradation model for backtesting. The energy efficiency value of the standard data is predicted and the corresponding predicted energy efficiency value is obtained. The energy efficiency prediction degradation model is used to predict the energy efficiency value of the healthy unit under the corresponding operating conditions. The predicted energy efficiency value of the stable state is selected and the deviation between the predicted energy efficiency value and the actual energy efficiency value of the standard data is calculated. Figure 3 A comparison chart of backtesting results was generated by substituting standard data into an energy efficiency prediction degradation model. The standard database includes: preset standard operating characteristic data or early-stage operating data of chiller units.
[0044] Visual aids are used to manually select stable data segments, or the moving standard deviation method is used to screen stable data segments as predicted energy efficiency values for the steady state. The mean relative error (MRE) of the stable data segments is used as the deviation. ; in, The standard data is the actual COP value. The standard data is substituted into the COP of the energy efficiency prediction degradation model to predict the energy efficiency value.
[0045] Example 7 Another embodiment of the present invention is as follows: Step S500: Based on the deviation, the performance degradation classification of the chiller system is evaluated to obtain the performance status of the chiller system, and corresponding operations are performed.
[0046] The performance status of the chiller system includes optimal state, deviating from optimal state, deviating from stable state, and failure state. When the performance status of the chiller system is not in the optimal state, the operator will be notified through the chiller monitoring terminal or corresponding prompting equipment.
[0047] In the performance degradation grading assessment of chiller units, if the deviation is <5%, the chiller unit is considered to be in optimal performance, and the chiller unit terminal will not perform any other actions; if the deviation is 5% << 15%, the chiller unit is considered to be deviating from optimal performance, and the chiller unit monitoring terminal will indicate that the unit is in a deviated from optimal performance; if the deviation is 15% << 25%, the chiller unit is considered to be deviating from stable performance, and the chiller unit monitoring terminal will indicate that the unit is in a deviated from stable performance; if the deviation is > 25%, the chiller unit is considered to be in a failure state, and the chiller unit monitoring terminal will indicate that the unit is in a failure state. The deviation threshold can be set to other fixed values or changed in real time according to the status of the chiller unit.
[0048] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for online sensing of the operating status of a chiller unit, characterized in that, The specific methods include the following: Obtain real-time operating data and real-time energy efficiency values of the chiller system; Align the time series of the data, extract the feature parameters that are highly correlated with the real-time energy efficiency value, and select the operating data that is longer than the set time span threshold. The feature parameters are normalized, and a long short-term memory artificial neural network with an attention mechanism is trained to predict the energy efficiency degradation model, which then predicts the energy efficiency value through the feature parameters. The standard data that meets the current characteristic data conditions in the standard database is called and substituted into the energy efficiency prediction degradation model for backtesting. The steady-state predicted energy efficiency value is selected, and the deviation between the predicted energy efficiency value and the actual energy efficiency value of the standard data is calculated. The performance degradation of the chiller system is graded based on the deviation, the performance status of the chiller system is obtained, and corresponding operations are performed.
2. The method for online sensing of the operating status of a chiller unit according to claim 1, characterized in that, The real-time energy efficiency value of a chiller system is taken as the coefficient of performance (COP) value: ； ； in, This refers to the heat absorbed by the chiller from the heat source or the heat dissipated by the chiller to the external environment. This is the sum of all input electrical power of the chiller system. The specific heat capacity of water, The water flow rate of the evaporator or condenser; This refers to the temperature difference between the inlet and outlet water of the evaporator or the temperature difference between the inlet and outlet water of the condenser.
3. The method for online sensing of the operating status of a chiller unit according to claim 1, characterized in that, Time series alignment of data includes the following: Based on the time step of data acquisition according to different operating parameters, align the data timestamps and unify the time step; Outlier data in the time series is removed, and the z-score method with a sliding window is used. The sliding window time step is 5-15, and the z-score is calculated as follows: ； in, for Time data, for The mean of the window before time step [time]. Let be the standard deviation of the window before time , if ,but If the time-based data contains outliers, these outliers will be removed. To fill in outliers, linear interpolation is used to calculate new data to fill in the removed outliers. The calculation formula is as follows: ； in, For a moment, for Time data.
4. The method for online sensing of the operating status of a chiller unit according to claim 3, characterized in that, The Pearson correlation coefficient method was used to extract feature parameters that are highly correlated with real-time energy efficiency values. The correlation coefficients between each operating parameter and the energy efficiency value were calculated, and the parameters that are highly correlated with the energy efficiency value were extracted as feature parameters. Using operating parameters as independent variables and energy efficiency value as the dependent variable, the Pearson correlation coefficient between the independent and dependent variables is calculated as follows: ； in, As the independent variable, For the observed values of the dependent variable, The mean of the independent variable. Let the mean of the dependent variable be... If the absolute value of the Pearson correlation coefficient is taken as the independent variable, then the high correlation characteristic parameter is taken.
5. The method for online sensing of the operating status of a chiller unit according to claim 1, characterized in that, Feature parameter normalization process to obtain normalized parameters The formula for its calculation is: ； in, The maximum value of the characteristic parameter. It represents the minimum value of the characteristic parameter.
6. The method for online sensing of the operating status of a chiller unit according to claim 5, characterized in that, Normalized feature parameters The sequence is fed into the input layer of the neural network as the input sequence of the Long Short-Term Memory (LSTM) neural network. First, it is processed by the LSTM layer to match the time series data. Then, the output of the LSTM is input into the attention layer, which uses the attention mechanism to uncover the high-dimensional correlations in the time series data. Finally, the data processed by the attention mechanism is... The input data is processed by the softmax function in the fully connected layer. The summation is performed, and the final predicted energy efficiency value is output by the output layer. An energy efficiency prediction degradation model was obtained.
7. The method for online sensing of the operating status of a chiller unit according to claim 1, characterized in that, The standard database includes the following data: preset standard operating characteristic data or early operating data of chiller units.
8. The method for online sensing of the operating status of a chiller unit according to claim 7, characterized in that, The standard data that meets the current characteristic data conditions in the standard database is called and substituted into the energy efficiency prediction degradation model for backtesting. The energy efficiency value of the standard data is predicted and the corresponding predicted energy efficiency value is obtained. The energy efficiency prediction degradation model is used to predict the energy efficiency value of the healthy unit under the standard data in the corresponding operating conditions. The predicted energy efficiency value in the stable state is selected and the deviation between the predicted energy efficiency value and the actual energy efficiency value of the standard data is calculated.
9. The method for online sensing of the operating status of a chiller unit according to claim 8, characterized in that, Visual aids were used to manually select stationary data segments as the predicted energy efficiency values for the steady state, and the mean relative error (MRE) of the stationary data segments was used as the deviation. ； in, The standard data is the actual COP value. The standard data is substituted into the COP of the energy efficiency prediction degradation model to predict the energy efficiency value.
10. The method for online sensing of the operating status of a chiller unit according to claim 1, characterized in that, The performance status of the chiller system includes optimal state, deviating from optimal state, deviating from stable state, and failure state. When the performance status of the chiller system is not in the optimal state, the operator will be notified through the chiller monitoring terminal or corresponding prompting equipment.