Water chiller operation control and model training method and electronic device

Abstract

The application discloses a water chilling unit operation control and model training method and electronic equipment. The water chilling unit operation control method comprises the following steps: obtaining a target relationship model between the energy efficiency ratio and the load rate of a target water chilling unit; determining the target load rate required for the target water chilling unit to reach the target energy efficiency ratio based on the target relationship model; and adjusting the operation parameters of the target water chilling unit based on the target load rate. The target relationship model comprises a target polynomial function module and a first correction network module. The target polynomial function module is constructed based on a reference polynomial function module in at least one reference relationship model. The reference relationship model is trained based on at least one first sample data. The target relationship model is trained by performing parameter training and adjustment on the target polynomial function and the first correction network module based on second sample data of the target water chilling unit at at least one historical time point.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method and electronic device for chiller unit operation control and model training. Background Technology

[0002] In building energy systems, the operating efficiency of chiller units has a decisive impact on overall energy consumption. Therefore, to achieve high energy efficiency, it is necessary to determine the energy efficiency ratio (EER) of chiller units under different load rates. Currently, the EER is generally determined based on historical operating data of the chiller units to rationally control their operation. However, historical operating data for newly deployed chiller units is limited, making it impossible to accurately determine the relationship between load rate and EER based on historical data, naturally leading to difficulties in rationally controlling the operation of the chiller units. Summary of the Invention

[0003] On the one hand, this application provides a method for controlling the operation of a chiller unit, including:

[0004] Obtain the target relationship model between the energy efficiency ratio and load rate of the target chiller unit;

[0005] Based on the target relationship model, the target load rate required for the target chiller unit to achieve the target energy efficiency ratio is determined;

[0006] Based on the target load rate, adjust the operating parameters of the target chiller unit;

[0007] The target relation model includes: a target polynomial function module and a first correction network module;

[0008] The objective polynomial function module is constructed based on the reference polynomial function module in at least one reference relation model; the reference relation model is trained based on at least one first sample data pair, the first sample data pair including: the sample energy efficiency ratio and sample load rate of the sample chiller unit;

[0009] The target relationship model is obtained by training the target polynomial function and the first correction network module with parameters based on the second sample data pair of the target chiller unit at at least one historical time point. The second sample data pair includes the historical energy efficiency ratio and historical load rate of the target chiller unit at the historical time point.

[0010] In one possible implementation, the polynomial functions corresponding to both the target polynomial function module and the reference polynomial function module are quadratic polynomial functions.

[0011] The first correction network module is a residual network module.

[0012] In another possible implementation, the first loss function used to train the target relation model includes a first regularization term, which includes the third derivative of the target relation function corresponding to the target relation model.

[0013] In another possible implementation, the first loss function further includes: a second regularization term;

[0014] The second regularization term includes: the difference between the first coefficient value of the coefficient term in the objective polynomial function and the mean value of the coefficients corresponding to the coefficient term;

[0015] Wherein, the target polynomial function is the polynomial function corresponding to the target polynomial function module;

[0016] The mean value of the coefficients corresponding to the coefficient terms is the average value of the second coefficient values ​​of the coefficient terms in at least one reference polynomial function; the reference polynomial function is the polynomial function corresponding to the reference polynomial function module.

[0017] In another possible implementation, the reference relation model includes the reference polynomial function module and the second correction network module;

[0018] The second loss function used to train the reference relationship model includes a third regularization term, which includes the third derivative of the reference relationship function corresponding to the reference relationship model.

[0019] In another possible implementation, the chiller unit operation control method further includes:

[0020] Determine the current actual energy efficiency ratio and actual load rate of the target chiller unit;

[0021] The actual energy efficiency ratio and the actual load rate are stored as the historical energy efficiency ratio and historical load rate of the target chiller unit, respectively, to obtain the second sample data pair corresponding to the current time point;

[0022] In response to the satisfaction of the model update conditions, the target relationship model is updated based on at least one second sample data pair obtained in the most recent time period.

[0023] Furthermore, this application also provides a model training method, including:

[0024] At least one trained reference relation model is obtained, the reference relation model comprising: a reference polynomial function module, the reference relation model being trained based on at least one first sample data pair, the first sample data pair comprising: the sample energy efficiency ratio and sample load rate of the sample chiller unit;

[0025] Based on the reference polynomial function module of the aforementioned reference relationship model, the target polynomial function module is determined.

[0026] Construct a target relation model to be trained, the target relation model including: the target polynomial function module to be trained and the first correction network module to be trained;

[0027] Obtain at least one second sample data pair of the target chiller unit in the most recent time period, the second sample data pair including: the historical energy efficiency ratio and historical load rate of the target chiller unit at a historical time point;

[0028] Based on the at least one second sample data pair, the parameters of the target polynomial function module and the first correction network module in the target relation model are trained and adjusted to obtain the trained target relation model.

[0029] In one possible implementation, the step of training and adjusting the parameters of the target polynomial function module and the first correction network module in the target relation model based on the at least one second sample data pair includes:

[0030] Using the historical load rate of the target chiller unit as training data and the historical energy efficiency ratio of the target chiller unit as the training objective, the parameters of the target polynomial function module and the first correction network module in the target relationship model are trained and adjusted in combination with the first loss function.

[0031] The first loss function includes at least one of a first regularization term and a second regularization term;

[0032] Wherein, the first regularization term includes: the third derivative of the target relation function corresponding to the target relation model;

[0033] The second regularization term includes: the difference between the first coefficient value of the coefficient term in the target polynomial function and the mean value of the coefficient corresponding to the coefficient term, wherein the target polynomial function is the polynomial function corresponding to the target polynomial function module; the mean value of the coefficient corresponding to the coefficient term is the average value corresponding to the second coefficient value of the coefficient term in at least one reference polynomial function, wherein the reference polynomial function is the polynomial function corresponding to the reference polynomial function module.

[0034] In yet another possible implementation, obtaining at least one trained reference relation model includes:

[0035] Obtain at least one first sample data pair corresponding to each of the sample chiller units;

[0036] Using the sample load rate corresponding to the chilled water sample group as training data and the sample energy efficiency ratio corresponding to the chilled water sample group as the training objective, a reference relationship model is trained by combining the second loss function to obtain a trained reference relationship model corresponding to at least one sample chiller unit.

[0037] The second loss function includes a third regularization term, which includes the third derivative of the reference relation function corresponding to the reference relation model.

[0038] In another aspect, this application also provides an electronic device, including: a memory and a processor;

[0039] The memory is used to store computer programs;

[0040] The processor is configured to execute the computer program to perform the following steps:

[0041] Obtain the target relationship model between the energy efficiency ratio and load rate of the target chiller unit;

[0042] Based on the target relationship model, the target load rate required for the target chiller unit to achieve the target energy efficiency ratio is determined;

[0043] Based on the target load rate, adjust the operating parameters of the target chiller unit;

[0044] The target relation model includes: a target polynomial function module and a first correction network module;

[0045] The objective polynomial function module is constructed based on the reference polynomial function module in at least one reference relation model; the reference relation model is trained based on at least one first sample data pair, the first sample data pair including: the sample energy efficiency ratio and sample load rate of the sample chiller unit;

[0046] The target relationship model is obtained by training the target polynomial function and the first correction network module with parameters based on the second sample data pair of the target chiller unit at at least one historical time point. The second sample data pair includes the historical energy efficiency ratio and historical load rate of the target chiller unit at the historical time point. Attached Figure Description

[0047] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0048] Figure 1A schematic flowchart of the chiller unit operation control method provided in this application;

[0049] Figure 2 A flowchart illustrating the model training method provided in this application;

[0050] Figure 3 This is a schematic diagram illustrating a data preprocessing process for candidate data pairs.

[0051] Figure 4 This is an example diagram showing the distribution of candidate data pairs for chiller unit A and chiller unit B respectively.

[0052] Figure 5 This is an example diagram showing the distribution of data pairs after data preprocessing for chiller unit A and chiller unit B respectively.

[0053] Figure 6 Another flowchart illustrating the model training method provided in this application;

[0054] Figure 7 This is an example graph showing the relationship between energy efficiency ratio and load rate as represented by the reference relationship model trained using the scheme of this application;

[0055] Figure 8 This is a schematic diagram illustrating one implementation process of updating the target relation model in this application;

[0056] Figure 9 An example diagram showing the determination of the optimal energy efficiency ratio and the corresponding target load rate based on the target relationship model trained in this application;

[0057] Figure 10 A schematic diagram of the component architecture of the electronic device provided in this application. Detailed Implementation

[0058] The embodiments of this application are described below with reference to the accompanying drawings. The terminology used in the implementation section of this application is only for explaining specific embodiments and is not intended to limit the application. Those skilled in the art will recognize that, with technological advancements and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0059] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms are interchangeable where appropriate; this is merely a way of distinguishing objects with the same attributes in the embodiments of this application. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of elements is not necessarily limited to those elements, but may include other elements not explicitly listed or inherent to those processes, methods, products, or apparatuses.

[0060] like Figure 1 The diagram illustrates a flow chart of a chiller unit operation control method provided in this application. The method of this embodiment can be applied to electronic devices, such as computers, servers, or other control devices used to control the operation of chiller units, without limitation.

[0061] The method in this embodiment may include:

[0062] S101, obtain the target relationship model between the energy efficiency ratio and load rate of the target chiller unit.

[0063] A chiller is a device that achieves cooling or heating through vapor compression refrigeration cycles or absorption refrigeration cycles. In building energy systems, chillers are the core equipment of central chiller stations, responsible for the centralized production of cooling or heating capacity.

[0064] The Energy Efficiency Ratio (EER) of a chiller is a core indicator for measuring its energy efficiency. For example, the EER can be the ratio of the cooling (or heating) capacity produced by the chiller to the total input power consumed by the chiller. In some cases, the EER can also be the Coefficient of Performance (COP).

[0065] Load factor refers to the ratio of the current actual cooling capacity (or heating capacity) of a chiller unit to its rated cooling capacity (or rated heating capacity). The actual cooling capacity (or heating capacity) of the chiller unit is related to the chilled water flow rate and the temperature difference between the chilled water inlet and outlet. For example, the actual cooling capacity of the chiller unit can be the product of the specific heat capacity of water, the density of water, the chilled water flow rate, and the temperature difference between the chilled water inlet and outlet.

[0066] In this application, for ease of distinction, the chiller unit that needs to be controlled is referred to as the target chiller unit.

[0067] In this application, the target relational model includes: a target polynomial function module and a first correction network module.

[0068] The objective polynomial function module is a module containing polynomial functions. Therefore, running this module essentially means running the polynomial functions to perform computational processing based on them. For ease of distinction, the polynomial functions included in this module are referred to as the objective polynomial functions. There are no restrictions on the types of objective polynomial functions. For example, in one possible implementation, research has shown that the curve relationship between the energy efficiency ratio and load rate of a chiller unit more closely matches the curve relationship represented by a quadratic polynomial function. Based on this, the objective polynomial function can be a quadratic polynomial function.

[0069] The first correction network module is used to correct the deviation of the target polynomial function module. This first correction network module can be a neural network model or other network models, without limitation. It is understood that, based on the physical characteristics of the chiller unit, a polynomial function can reflect the main variation law between the chiller unit's energy efficiency ratio and load rate. However, the variation law between the energy efficiency ratio and load rate reflected solely by the polynomial function still has some deviations. Therefore, this application introduces a first correction network module into the target relationship model to capture and correct the deviation between the nonlinear relationship represented by the polynomial function and the energy efficiency law of the chiller unit, so that the target relationship model can be used to represent the energy efficiency law of the chiller unit.

[0070] In this application, the specific model type of the first correction network module can be set according to actual needs and is not limited thereto. In one possible implementation, in order to reduce the complexity and resource consumption of the target relationship model, the first correction network module can be a residual network module. The model size of the residual network model is relatively small, which can both initiate the correction effect and avoid the target relationship model from becoming too complex.

[0071] It is understandable that when historical operating data for the target chiller unit is lacking or relatively scarce, the polynomial function relationship between the energy efficiency ratio and load rate in the target chiller unit cannot be accurately determined based on this historical operating data. Therefore, in this application, the target polynomial function module in the target relationship model is constructed based on a reference polynomial function module in at least one reference relationship model. Specifically, the reference polynomial function module includes reference polynomial functions; that is, the reference polynomial function module can be a module that includes polynomial functions.

[0072] In this application, the reference relationship model is trained based on at least one first sample data pair. The first sample data pair includes: the sample energy efficiency ratio (EER) and sample load rate of the sample chiller unit. The sample EER and sample load rate are the EER and load rate generated from the historical operation of the sample chiller unit. The sample chiller unit can be a chiller unit that has been running for a period of time and for which sufficient EER and load rate data can be collected. To ensure that the constructed objective polynomial function module can more accurately reflect the relationship between the EER and load rate of the target chiller unit, in this application, the sample chiller unit can be a chiller unit of the same model as the target chiller unit.

[0073] In this application, different reference relationship models can be trained based on at least one first sample data pair corresponding to different sample chiller units. Therefore, different reference relationship models correspond to different sample chiller units, and the reference relationship model is used to reflect the relationship between the energy efficiency ratio and the load rate of its corresponding sample chiller unit.

[0074] In this application, constructing a target polynomial function based on the reference polynomial function modules corresponding to at least one reference relation model can be achieved by constructing the reference polynomial functions in the at least one reference polynomial function modules corresponding to the at least one reference relation model into a target polynomial function, thereby obtaining a target polynomial function module including the target polynomial function.

[0075] For example, if there is only one reference relation model, then the reference polynomial function of the reference polynomial function module in that model can be determined as the target polynomial function. If there are multiple reference relation models, then for each coefficient term in the target polynomial function, the value of that coefficient term is the average of the coefficient values ​​of that coefficient term at the corresponding position in the multiple reference polynomial functions corresponding to the multiple reference polynomial function modules. For example, assuming there are two reference polynomial functions with quadratic coefficients of 6 and 8 respectively, then the value of the quadratic coefficient in the target polynomial function is 7.

[0076] The reference polynomial function can be any polynomial function.

[0077] In one possible implementation, the reference polynomial function corresponding to the reference polynomial function module is a quadratic polynomial function. Based on this, the target polynomial function is constructed from at least one quadratic polynomial function corresponding to multiple reference polynomial function modules, ensuring that the constructed target polynomial function is also a quadratic polynomial function.

[0078] In this application, in order to make the target relation model applicable to the target chiller unit, after constructing the target relation model based on the reference polynomial function in the reference relation model, it is also necessary to train the target relation model. Specifically, the target relation model is obtained by training and adjusting the parameters of the target polynomial function and the first correction network module based on second sample data pairs of the target chiller unit at at least one historical time point.

[0079] Each second sample data pair includes: the historical energy efficiency ratio and historical load rate of the target chiller unit at a historical point in time. Different second sample data pairs were collected at different historical points in time.

[0080] Furthermore, considering that the recent energy efficiency ratio and load rate of the target chiller can more accurately reflect the performance characteristics of the target chiller, the second sample data pair used to train the target relationship model is the second sample data pair at each historical time point in the most recent time period.

[0081] S102, based on the target relationship model, determine the target load rate required for the target chiller unit to achieve the target energy efficiency ratio.

[0082] The target energy efficiency ratio can be set according to actual needs.

[0083] For example, the target energy efficiency ratio can be the optimal energy efficiency ratio that the target chiller unit can achieve. In this case, the optimal energy efficiency ratio that the target chiller unit can achieve can be determined by taking the derivative of the target relation function corresponding to the target relation model. Then, the target load rate required to achieve the optimal energy efficiency ratio can be determined based on the target relation model.

[0084] For example, the target energy efficiency ratio can be a set value, such as 70%. Then, the target energy efficiency ratio can be substituted into the target relationship model to obtain the target load factor. Of course, there are other possibilities for the target energy efficiency ratio, without any restrictions.

[0085] The target load rate is the load rate corresponding to the target energy efficiency ratio, determined based on the target relationship model.

[0086] S103, Based on the target load rate, adjust the operating parameters of the target chiller unit.

[0087] The operating parameters are those that affect the load rate of the target chiller unit. For example, as discussed earlier regarding load rate, the chilled water flow rate and the temperature difference between the chilled water inlet and outlet affect the chiller unit's load rate. Therefore, based on the target load rate, this application allows adjustment of one or more of the chilled water temperature and the temperature difference between the inlet and outlet of the target chiller unit. Of course, adjusting the operating parameters of the target chiller unit can also include adjusting its operating time and shutdown time, etc., without any specific limitations.

[0088] The purpose of adjusting the operating parameters of the target chiller unit is to enable the load rate of the target chiller unit to reach the target load rate, thereby enabling the energy efficiency ratio of the target chiller unit to reach the target energy efficiency ratio.

[0089] As can be seen from the above, this application, based on a target relationship model between the energy efficiency ratio (EER) and load rate of a target chiller unit, determines the target load rate required for the target chiller unit to achieve the target EER, and adjusts the operating parameters of the target chiller unit based on the target load rate, thereby enabling the reasonable operation of the target chiller unit and allowing it to achieve the target EER. In this application, the target polynomial function module in the target relationship model is constructed based on the reference polynomial function module in at least one reference relationship model. The reference relationship model is trained on sample data pairs consisting of sample EERs and sample load rates corresponding to sample chiller units. This allows the target polynomial function module constructed based on the reference polynomial function module in at least one reference relationship model to accurately reflect the relationship between the EER and load rate of the target chiller unit to a certain extent. Furthermore, the target relationship model also includes a first correction network module, which can correct deviations in the target polynomial function module. Based on this, this application trains the target polynomial function and the first correction network module in the target relationship model by training and adjusting the parameters based on the historical energy efficiency ratio and historical load rate of the target chiller unit at at least one historical time point. Even when there is limited historical data for the target chiller unit, the trained target relationship model can accurately reflect the relationship between the energy efficiency ratio and load rate of the target chiller unit. This allows for reasonable control of the operation of the target chiller unit based on the target relationship model, thereby reasonably controlling the energy efficiency ratio of the target chiller unit.

[0090] In this application, the specific implementation process for obtaining and training the target relation model is not limited.

[0091] To facilitate understanding, the following explanation uses one method of training a target relation model as an example. For example... Figure 2The diagram illustrates an implementation flow of the model training method provided in this application. The method of this embodiment can be applied to the aforementioned electronic devices or other server devices without limitation.

[0092] The model training method in this embodiment may include:

[0093] S201, Obtain at least one trained reference relation model.

[0094] The reference relation model includes a reference polynomial function module. This reference polynomial function module can include polynomial functions of any type; for example, the polynomial functions in this reference polynomial function module can be quadratic polynomial functions.

[0095] The reference relationship model is trained based on at least one first sample data pair. The first sample data pair includes the sample energy efficiency ratio (EER) and sample load rate of the sample chiller units. Different reference relationship models are trained using first sample data pairs from different sample chiller units, such that different reference relationship models correspond to different sample chiller units. Each reference relationship model characterizes the relationship between the EER and load rate in the sample chiller unit corresponding to that reference relationship model.

[0096] For example, using the sample load rate of the first sample data pair of the sample chiller units as training data, and the sample energy efficiency ratio of the first sample data pair as the training objective, a reference relationship model is trained to obtain the trained reference relationship model. During the training of this reference relationship model, the polynomial coefficients in the reference polynomial function module can be continuously adjusted. The training of this reference relationship model can employ supervised training methods, and the specific process is not restricted.

[0097] In this application, in order to enable the reference polynomial function module in the trained reference relationship model to be used to express the relationship between the energy efficiency ratio and load rate of the target chiller unit more accurately, the sample chiller unit can be the same model or type of chiller unit as the target chiller unit.

[0098] S202, based on the reference polynomial function modules of each reference relation model, determine the target polynomial function module.

[0099] For example, if there is only one reference relation model, the reference polynomial function module in that reference relation model can be used as the target polynomial function module.

[0100] If there are multiple reference relation models, the average value of the polynomial coefficients (i.e., the coefficient values ​​of the coefficient terms) at each position in the reference polynomial function of each reference polynomial function module can be calculated separately. This average value can then be used as the polynomial coefficients at the corresponding positions in the target polynomial function of the target polynomial function module, thus obtaining the target polynomial function module that includes the target polynomial function. Alternatively, the weighted average value of the polynomial coefficients at each position in the reference polynomial function of each reference polynomial function module can be calculated separately, and this weighted average value can then be used as the polynomial coefficients at the corresponding positions in the target polynomial function of the target polynomial function module.

[0101] To facilitate understanding, taking a quadratic polynomial function as an example, we can calculate the average of the quadratic coefficients and the average of the linear coefficients of all quadratic polynomial functions in each reference polynomial function module. The average of the quadratic coefficients is then used as the quadratic coefficients in the target polynomial function module, and the average of the linear coefficients is used as the linear coefficients in the target polynomial function module.

[0102] S203, Construct the target relation model to be trained. The target relation model includes: the target polynomial function module to be trained and the first correction network module to be trained.

[0103] In this application, the target polynomial function module determined in step S202 is combined with a first correction network module to form a target relation model.

[0104] As mentioned earlier, the first correction network module can be any network model module, without any restrictions. In one possible implementation, the first correction network module can be a residual network module.

[0105] For example, if the polynomial function corresponding to the polynomial function module in the target relation model is a quadratic polynomial function, then the target relation function corresponding to the target relation model... This can be expressed as Formula 1 below:

[0106] (Formula 1);

[0107] in, The three coefficient terms of the quadratic polynomial function in the target relation model are called the quadratic coefficient, the linear coefficient, and the constant term, respectively. This is the function expression for the first correction network module, or it could be the function expression for the residual network module. This indicates the load rate of the target chiller unit, with a value ranging from 0 to 100%. 0 indicates that the target chiller unit is off, and 100% indicates that the target chiller unit is running at full load. The scaling factor is set.

[0108] S204, Obtain at least one second sample data pair for the target chiller unit in the most recent time period.

[0109] The time period refers to the data acquisition cycle for collecting the operating data of the target chiller unit. The specific duration of the time period can be set as needed and is not limited.

[0110] The most recent time period includes at least one historical time point, and each historical time point corresponds to a second sample data pair. Different historical time points correspond to different second sample data pairs. For example, the time period can be 7 days, then each day within 7 days can be a historical time point, or each hour or every two hours within 7 days can be a historical time point, etc.

[0111] The second sample data includes the historical energy efficiency ratio and historical load rate of the target chiller unit at a historical point in time.

[0112] S205, based on at least one second sample data pair, perform parameter training and adjustment on the target polynomial function module and the first correction network module in the target relation model to obtain the trained target relation model.

[0113] The training objective relationship model can be trained using a supervised training method, and there are no restrictions on the specific training process.

[0114] For example, using the historical load rate of the second sample data pair as training data and the historical energy efficiency ratio of the second sample data pair as the training objective, the parameters of the objective polynomial function module and the first correction network module are trained and adjusted to obtain the target relationship model. Specifically, the historical load rate of the target chiller unit can be used as training data, the historical energy efficiency ratio of the target chiller unit can be used as the training objective, and the parameters of the objective polynomial function module and the first correction network module in the target relationship model can be trained and adjusted in combination with the first loss function.

[0115] The first loss function can be set according to actual needs and is not restricted. For example, the first loss function is used to quantify at least the difference between the predicted energy efficiency ratio based on the historical load rate and the historical energy efficiency ratio corresponding to the historical load rate. That is, the first loss function includes at least a quantification term for quantifying the difference between the predicted energy efficiency ratio based on the historical load rate and the historical energy efficiency ratio corresponding to the historical load rate.

[0116] Based on the above, during the training of the target relationship model, the energy efficiency ratio difference between the predicted energy efficiency ratio generated by the current target relationship model based on the historical load rate and the historical energy efficiency ratio corresponding to the historical load rate can be continuously compared. Based on the energy efficiency ratio difference, the parameters to be trained and adjusted in the target polynomial function module and the first correction network module are adjusted to continuously narrow the energy efficiency ratio difference between the predicted energy efficiency ratio generated by the target relationship model and the corresponding historical energy efficiency ratio.

[0117] In the parameter training and adjustment of the target polynomial function module and the first correction network module, the polynomial coefficients in the target polynomial function corresponding to the target polynomial function module and the internal network parameters of the first correction network module can be adjusted.

[0118] In this application, after determining the target polynomial function module based on multiple reference polynomial function modules, the coefficient values ​​of each coefficient term in the target polynomial function corresponding to the target polynomial function module are not directly adopted. Instead, the coefficient values ​​of each coefficient term are gradually adjusted during the training process to avoid the risk of overfitting caused by directly transferring the coefficient terms of the reference polynomial function corresponding to the reference polynomial function module, while retaining the adaptive capability of the target polynomial function to the newly deployed target chiller unit.

[0119] As can be seen from the above, the objective polynomial function module in the objective relationship model is constructed and trained based on the historical operating data of the sample chiller units, enabling the objective polynomial function module to reflect the relationship between the load rate and the energy efficiency ratio of the target chiller units to a certain extent. However, considering that the curve relationship expressed by the objective polynomial function module still deviates somewhat from the actual relationship between the energy efficiency ratio and the load rate of the target chiller units, this application adds a first correction network module to the objective polynomial function module to correct this deviation. Furthermore, since the objective polynomial function module in the objective relationship model has been trained using the historical operating data of the sample chiller units, even if the historical load rate and historical energy efficiency ratio data of the target chiller units are relatively limited, training the objective relationship model based on a small amount of historical load rate and historical energy efficiency ratio data can reliably reflect the relationship between the load rate and the energy efficiency ratio of the target chiller units. This allows the trained objective relationship model to accurately predict the energy efficiency ratio of the target chiller units, thereby enabling reasonable control of their operation.

[0120] It is understandable that the training of the target relationship model in this application is actually divided into two stages. The first stage is to train the reference relationship model and then build the target relationship model based on the trained reference relationship model. The second stage is to train the target relationship model based on the operating data of the target chiller unit. Since the first stage does not depend on the operation of the target chiller unit and can be completed before the target chiller unit starts operating, the first stage can be considered the offline modeling stage. The second stage requires continuous optimization and adjustment of the target relationship model based on the operating data of the target chiller unit; therefore, the second stage is the online modeling stage.

[0121] In the first and second stages, to improve the accuracy of the trained reference or target relation model, after obtaining multiple data pairs of sample or target chiller units, this application requires data preprocessing to remove noise data. For ease of distinction, the initially obtained data pairs are referred to as candidate data pairs, which may include candidate load rates and candidate energy efficiency ratios.

[0122] For example, for a sample chiller unit, a candidate data pair can be a candidate sample data pair of the sample chiller unit, which includes the candidate sample load rate and the candidate sample energy efficiency ratio.

[0123] For the target chiller unit, the candidate data pair can be the candidate historical data pair of the target chiller unit, which includes the candidate historical load rate and the candidate historical energy efficiency ratio.

[0124] Data preprocessing for candidate data pairs can include, but is not limited to, the following:

[0125] Candidate data pairs that do not meet the set physical specifications are removed. These specifications may include, but are not limited to, the following: candidate load rate being between 0% and 100%; candidate energy efficiency ratio being a positive number not exceeding 100%; the time point (timestamp) at which the candidate data pair was obtained being a valid time point; and the key fields associated with the candidate data pair being complete. These key fields may include flow rate and inlet / outlet temperature, etc.

[0126] Filter out the candidate data pairs with the highest corresponding density values ​​and the highest proportion of the number of the preceding targets;

[0127] Based on the interquartile range, candidate data pairs that are extreme outliers are removed;

[0128] Remove isolated noise points from multiple candidate sample data pairs that cannot be clustered into clusters;

[0129] Candidate data pairs that belong to local outliers are removed based on Local Outlier Factor (LOF) detection.

[0130] To facilitate understanding, the following example illustrates one method for preprocessing multiple candidate data pairs. Figure 3 This diagram illustrates one implementation flow for preprocessing candidate data pairs, which may include:

[0131] S301, obtain multiple candidate data pairs.

[0132] As mentioned earlier, during the training phase of the reference relation model, the candidate data pairs are candidate sample data pairs for the sample chiller units. During the training phase of the target relation model, the candidate data pairs are candidate historical data pairs for the target chiller units.

[0133] S302, Remove candidate data pairs from multiple candidate data pairs that do not conform to the set physical specifications.

[0134] As mentioned earlier, if the candidate load rate in a candidate data pair exceeds [0%, 100%], the candidate energy efficiency ratio is negative, the candidate energy efficiency ratio is a positive number exceeding 100%, the time point of the candidate data pair is not a valid time point (e.g., the timestamp format of the time point is incorrect, or it significantly exceeds the time range of the candidate data pair collection), or the key fields associated with the candidate data pair are missing, then the candidate data pair can be considered to be a candidate data pair that does not conform to the physical specifications, and this candidate data pair needs to be removed.

[0135] S303 uses Gaussian kernel density estimation to determine the density value of each candidate data pair, and retains the candidate data pair with the highest density value and the highest proportion of the number of previous targets among the remaining candidate data pairs.

[0136] Among them, the density value (also known as the probability density value) of the candidate data pair reflects the frequency of occurrence of the candidate data pair (the working condition represented by the candidate data pair). The higher the density value, the more common the candidate data pair is; the lower the density value, the rarer the candidate data pair is (e.g., it may be a data pair under abnormal, noisy, or transient working conditions).

[0137] The target quantity ratio can be set according to actual needs.

[0138] For example, if the target number ratio can be 75%, then after estimating the density value of each candidate data pair based on Gaussian kernel density estimation (KDE), they can be sorted in descending order of density value to determine the top 75% of candidate data pairs, while removing the bottom 25% of candidate data pairs.

[0139] S304, based on the interquartile range (IQR), identifies extreme outliers among the remaining candidate data pairs and removes them.

[0140] The interquartile range (IQR) refers to the difference between the third quartile (Q3) and the first quartile (Q1) of multiple candidate data pairs. Based on the IQR, the normal range of values ​​that belong to the normal range can be determined, while candidate data pairs that do not belong to the normal range are extreme outliers.

[0141] For example, the two boundary values ​​of the normal range are as follows:

[0142] Lower boundary: Q1−S×IQR;

[0143] Upper boundary: Q3 + S × IQR;

[0144] Where S is the set adjustment coefficient, which can be set as needed; for example, S can be 3. Correspondingly, if a candidate data pair exceeds any boundary of the normal value range, then that candidate data pair is an extreme outlier. Since the candidate data pair includes candidate load factor and candidate energy efficiency ratio, the upper boundary includes the upper boundary of the candidate load factor and the upper boundary of the candidate energy efficiency ratio, and the lower boundary also includes the lower boundary of the candidate load factor and the lower boundary of the candidate energy efficiency ratio.

[0145] S305, determine whether the number of remaining candidate data pairs exceeds the set number. If yes, proceed to step S306; otherwise, determine the remaining candidate data pairs as preprocessed data pairs.

[0146] If the number of remaining candidate data pairs is less than the set number, the effect of clustering and local anomaly detection will be relatively poor due to the relatively small number of candidate data pairs. Therefore, the remaining candidate data pairs can be directly used as the first sample data pairs for training the reference relation model or the second sample data pairs for training the target relation model.

[0147] S306 uses the density-based clustering algorithm DBSCAN to cluster the remaining candidate data pairs and remove isolated noise points identified in the clustering.

[0148] Among them, the density-based spatial clustering of applications with noise (DBSCAN) algorithm can identify isolated noise points (i.e., isolated candidate data pairs) that cannot be clustered with other candidate data pairs during the clustering process. While clustering at least one cluster, it can also remove isolated noise points.

[0149] S307. The Local Outlier Factor (LOF) detection is used to remove candidate data pairs that belong to local outliers from multiple candidate data pairs, and the remaining candidate data pairs are determined as the data pairs after data preprocessing.

[0150] The Local Access Count (LOF) can be used to calculate the local reachability density of candidate data with respect to its target number of nearest neighbors. Based on the local reachability density of each candidate data pair, candidate data pairs belonging to local outliers can be identified. For example, based on the local reachability density of a candidate data pair and the local reachability densities of its neighboring candidate data pairs, the LOF value of the candidate data pair can be calculated. If the LOF value is greater than 1, it indicates that the density of the candidate data pair is lower than the average density of its neighbors. Therefore, the candidate data pair is a local outlier and needs to be removed.

[0151] Understandably, by preprocessing the initial candidate data pairs of chiller units (such as sample chiller units or target chiller units), various noise data can be removed, thereby reducing the distortion of the relationship curve between energy efficiency ratio and load rate caused by noise data. Naturally, this can also reduce the distortion of the curve expressed by the target relationship model or reference relationship model trained later.

[0152] To better understand the benefits of data preprocessing, the following explanation uses two sets of chiller units as examples. Figure 4 This is an example diagram showing the distribution of candidate data pairs for chiller unit A and chiller unit B, respectively. Figure 5 This is an example diagram showing the distribution of data pairs after data preprocessing for chiller unit A and chiller unit B.

[0153] exist Figure 4 and Figure 5 In the example diagram of the distribution of data pairs corresponding to each chiller unit, each point represents a data pair, and the horizontal axis of each distribution example diagram represents the load rate, and the vertical axis represents the energy efficiency ratio.

[0154] Depend on Figure 4It can be seen that due to sensor malfunctions, transient start-up and shutdown of chiller units, or severe load fluctuations, the distribution of the initially obtained data pairs of chiller unit A and chiller unit B is relatively discrete, and the fitted relationship curve cannot truly reflect the relationship between load rate and energy efficiency ratio.

[0155] Through this pair Figure 4 After data preprocessing, noise can be removed from the data pairs, making the data pairs consisting of the chiller unit's load rate and energy efficiency ratio relatively concentrated, such as... Figure 5 As shown. Based on this, Figure 5 The relationship curve constructed from the data of the intermediate chiller unit can more accurately reflect the relationship between the energy efficiency ratio and the load rate of the chiller unit.

[0156] In any of the above embodiments of this application, in order to enable the target polynomial function module, constructed based on the reference polynomial function module in the reference relationship model, to more accurately reflect the relationship between the energy efficiency ratio and load rate in the target chiller unit, the composition structure of the reference relationship model can be similar to that of the target relationship model. Based on this, in this application, the reference relationship model may include a reference polynomial function module and a second correction network module. The second correction network module is a network module used to correct the deviation of the reference polynomial function module. For example, the second correction network module can be a neural network model or other network models, without limitation.

[0157] In this model, the polynomial functions included in the reference polynomial function module of the reference relation model can be of the same type as the polynomial functions included in the target polynomial function module of the target relation model, and the first and second correction network modules can be network modules of the same structure and type. For example, the polynomial functions included in both the reference and target polynomial function modules are quadratic polynomial functions, while the first and second correction network modules are both residual network modules. The specific form of the residual network module can also be varied; for example, the residual network module can be a feedforward neural network model.

[0158] It is understandable that the reference relationship model is trained using the sample energy efficiency ratio and sample load rate of sample chiller units with a large amount of historical operating data (such as chiller units of the same model as the target chiller unit). When the reference relationship model and the target relationship model have the same composition structure, that is, both include a polynomial function module and a correction network module, the polynomial function corresponding to the target polynomial function module constructed based on the reference polynomial function module in at least one trained reference relationship model can more accurately reflect the relationship between the energy efficiency ratio and load rate of the target chiller unit.

[0159] Based on this, after constructing a target relationship model including the objective polynomial function module and another correction network module (i.e., the first correction network module), this application uses the historical load rate and historical energy efficiency ratio of the target chiller unit at at least one historical time point to train the target relationship model. This can more efficiently complete the parameter adjustment of the objective polynomial function module and the first correction network module in the target relationship model, and can make the trained target relationship model more accurately reflect the relationship between the load rate and energy efficiency ratio of the target chiller unit.

[0160] It is understood that in the above embodiments of this application, the relationship curve between the load rate and energy efficiency ratio of the chiller unit should be a relatively smooth curve, without sudden increases or decreases. Therefore, in order to ensure that the curve expressed by the trained target relationship model has good smoothness and conforms to the continuous change physical characteristics of a thermodynamic system, in this application, the first loss function may include a first regularization term, which includes the third derivative of the target relationship function corresponding to the target relationship model. Here, the target relationship function is the functional expression of the target relationship model.

[0161] For example, one form of expression for the first loss function It can be shown in Formula 2 below:

[0162] (Formula 2);

[0163] Where N represents the number of the second sample data pairs, This represents the target relation function corresponding to the target relation model. For example, Formula 1 above is one expression of this target relation function. Of course, this target relation function can also have other forms, and there are no restrictions on this. Representing the Historical load rate in a second sample data pair The target relationship model is based on historical load rate. The predicted energy efficiency ratio, Indicates the first The historical energy efficiency ratio in the second sample data pair, i.e., the ratio to the historical load rate. The actual corresponding historical energy efficiency ratio. The set regularization weight can be set according to actual needs.

[0164] As can be seen, the first term in Formula 2 (the part before the plus sign) is a quantification term used to quantify the difference between the predicted energy efficiency ratio and the actual historical energy efficiency ratio. The second term in Formula 2 (the part after the plus sign) is the first regularization term, which is used to characterize the third derivative of the target relation function corresponding to the target relation model. By introducing the third derivative of the target relation function as a regularization term into the first loss function, the smoothness of the curve expressed by the target relation function corresponding to the trained target relation model can be improved.

[0165] In another possible implementation, considering that the number of historical load rates and historical energy efficiency ratios of the target chiller unit is relatively small, in order to improve the accuracy of the target polynomial function in the target polynomial function module through training, a second regularization term can be introduced into the first loss function in this application. The second regularization term includes the difference between the first coefficient value of the coefficient term in the target polynomial function and the mean value of the coefficient corresponding to the coefficient term.

[0166] It is understandable that the target polynomial function and the reference polynomial function corresponding to the reference polynomial function module are the same type of polynomial function. For example, both the target polynomial function and the reference polynomial function are cubic polynomial functions or quadratic polynomial functions. Based on this, the target polynomial function and the reference polynomial function have the same type and number of coefficient terms, only the coefficient values ​​of the same coefficient term differ between the two functions. For example, if both the target polynomial function and the reference polynomial function are quadratic polynomials, then the coefficient terms of the target polynomial function and the reference polynomial function include quadratic coefficients and linear coefficients. However, the coefficient values ​​of the quadratic coefficients in these two functions may differ, and the coefficients of the linear coefficients may also differ.

[0167] For any given coefficient term, the mean of the coefficients corresponding to that coefficient term is the average of the second coefficient values ​​of that coefficient term in at least one reference polynomial function.

[0168] For example, another expression of the first loss function It can be shown in Formula 3 below:

[0169] (Formula 3);

[0170] The meanings of the first term (the part before the plus sign) and each parameter in Formula 3 can be found in the relevant introduction to Formula 2, and will not be repeated here.

[0171] The prior regularization weight is set, and its specific value can be set according to actual needs. For the objective polynomial function module, the objective polynomial function is in the objective polynomial function. The coefficient value of the secondary term coefficient; for The mean of the coefficients corresponding to the coefficients of the secondary terms; The value of M is a natural number from 0 to M, where M represents the order of the objective polynomial function. For example, if the objective polynomial function is a quadratic polynomial, then the order is 2, and the value of M is 2.

[0172] In Formula 3, the second term (the part after the plus sign) represents the sum of squares of the differences between the first coefficient value of each coefficient term in the objective polynomial function and the mean value of the corresponding coefficient term. This second term can characterize the difference between the first coefficient value of each coefficient term in the objective polynomial function and the mean value of the corresponding coefficient term.

[0173] Understandably, since the second regularization term is used to characterize the difference between the first coefficient value of the coefficient term in the objective polynomial function and the mean value of the corresponding coefficient term, by introducing this second regularization term into the first loss function, the deviation between the first coefficient value of the coefficient term and the prior mean value of the coefficient term can be penalized during the training of the objective relation model. This allows the values ​​of the coefficient terms of the objective polynomial function to be gradually adjusted under prior constraints, thus enabling more reasonable control over the adjustment of the coefficient values ​​during training. This avoids overfitting caused by directly transferring the values ​​of the coefficient terms and ensures that the objective polynomial function is adaptive to the characteristics of the newly deployed target chiller unit.

[0174] Of course, in practical applications, in order to ensure that the trained target relationship model can accurately represent the relationship between the energy efficiency ratio and the load rate in the target chiller unit, and that the curve expressed by the target relationship model has good smoothness, the first loss function can also introduce a first regularization term and a second regularization term. Thus, another expression of this first loss function... It can be represented by Formula 4 below:

[0175] (Formula 4);

[0176] The meanings of the relevant parameters in Formula 4 can be found in the previous explanations of Formulas 2 and 3, and will not be repeated here.

[0177] In this application, to ensure the curves represented by the trained reference relationship model have good smoothness, and to enable the target polynomial function module constructed based on the reference polynomial function module in the reference relationship model to be better suited to the target chiller unit, the second loss function used to train the reference relationship model includes a third regularization term. This third regularization term includes the third derivative of the reference relationship function corresponding to the reference relationship model.

[0178] To facilitate understanding, the model training method of this application will be introduced below using one possible implementation as an example. For example... Figure 6 This illustrates another flowchart of the model training method provided in this application. The method in this embodiment may include:

[0179] S601, obtain at least one first sample data pair corresponding to each of the at least one sample chiller unit.

[0180] In one possible implementation, this application may first obtain at least one candidate sample data pair of the sample chiller unit, and then perform data preprocessing on the at least one candidate sample data pair in accordance with the previous data preprocessing method to obtain at least one first sample data pair, the details of which will not be elaborated further.

[0181] S602, for each sample chiller unit, the sample load rate in the first sample data pair corresponding to the chiller sample group is used as training data, and the sample energy efficiency ratio in the first sample data pair corresponding to the chiller sample group is used as the training target. The reference relationship model is trained in combination with the second loss function to obtain at least one trained reference relationship model corresponding to each sample chiller unit.

[0182] The second loss function includes a third regularization term, which includes the third derivative of the reference relation function corresponding to the reference relation model.

[0183] The functional expression of the reference relation model is the reference relation function.

[0184] In this application, the reference relation model may include only the reference polynomial function module, in which case the reference relation model only includes the reference polynomial function corresponding to the reference polynomial function module.

[0185] Furthermore, the reference relation model may also include a second correction network module. In this case, the reference relation function may include a reference polynomial function and the function corresponding to the second correction network module. For example, when the reference relation model includes a quadratic polynomial function module and a second correction network module, the reference relation function corresponding to this model can have a similar function structure to that in Equation 1, except that the coefficients of the quadratic polynomial function and the function corresponding to the second correction network module will be different.

[0186] The third regularization term is similar to the first regularization term in Formula 2, and will not be elaborated further.

[0187] Understandably, introducing a third-order loss function into the second loss function used to train the reference relation model can make the output relation curve of the reference relation model smoother. For easier understanding, please refer to [reference needed]. Figure 7 Please provide an explanation.

[0188] Figure 7 An example graph showing the relationship between energy efficiency ratio and load rate as represented by the reference relationship model trained using the scheme of this application is shown.

[0189] Figure 7 Example graphs are shown, illustrating the relationship curves output by the reference relationship models for chiller unit A and chiller unit B, respectively. Figure 7 The dashed lines represent the relationship curves output by the reference relationship model, while the solid broken lines represent the relationship broken lines constructed by combining the distribution of the second sample data pairs. The two endpoints of each vertical line on these broken lines are the lines connecting the set upper and lower quantiles corresponding to each energy efficiency ratio at the same load rate. The relationship broken lines are composed of the lines connecting the centers of the set upper and lower quantiles corresponding to each load rate.

[0190] contrast Figure 7 As can be seen from the broken line relationship and the relationship curve represented by the dashed line corresponding to chiller unit A, and the broken line relationship and the relationship curve represented by the dashed line corresponding to chiller unit B, the relationship curves and broken lines represented by the reference relationship model trained by this application are similar in distribution and smoother, better reflecting the physical change law between load rate and energy efficiency ratio, and more in line with the continuous change physical characteristics of thermodynamic system.

[0191] S603, a reference polynomial function module based on a reference relation model, and a target polynomial function module.

[0192] For example, if the reference polynomial function module includes quadratic polynomial functions, then the target polynomial function corresponding to the target polynomial function module is also a quadratic polynomial function. In this case, the coefficient of the quadratic term of the target polynomial function is the average of the coefficients of the quadratic terms in the quadratic polynomial functions of each reference polynomial function module, while the coefficient of the linear term of the target polynomial function is the average of the coefficients of the linear terms corresponding to each reference polynomial function module, and the constant term of the target polynomial function is the average of the constant terms corresponding to each reference polynomial function module.

[0193] S604, Construct the target relationship model to be trained.

[0194] The target relation model includes: a target polynomial function module to be trained and a first correction network module to be trained.

[0195] For example, the target relation function corresponding to the target relation model can be as shown in Formula 1 above. Formula 1 takes the polynomial function in the target polynomial function module of the target relation model as a quadratic polynomial function as an example. If the polynomial function in the target polynomial function module is another case, you only need to replace the quadratic polynomial function in Formula 1 with other polynomial functions. The details will not be elaborated further.

[0196] In this application, the first correction network module in the constructed target relation model does not directly adopt the second correction network module in the reference relation model. The purpose is to enable the first correction network module to effectively capture the unique nonlinear behavior of the target chiller unit after training.

[0197] S605, obtain at least one second sample data pair of the target chiller unit in the most recent time period.

[0198] The second sample data includes the historical energy efficiency ratio and historical load rate of the target chiller unit at a historical point in time.

[0199] S606, using the historical load rate of the target chiller unit as training data and the historical energy efficiency ratio of the target chiller unit as the training objective, the target polynomial function module and the first correction network module in the target relationship model are trained and adjusted in combination with the first loss function until the training objective is met, and the trained target relationship model is obtained.

[0200] The first loss function includes at least one of a first regularization term and a second regularization term.

[0201] The first regularization term includes the third derivative of the target relation function corresponding to the target relation model.

[0202] The second regularization term includes the difference between the first coefficient value of the coefficient term in the target polynomial function and the mean coefficient value corresponding to that coefficient term. The target polynomial function is the polynomial function corresponding to the target polynomial function module. For any coefficient term in the target polynomial function, the mean coefficient value corresponding to that coefficient term is the average value corresponding to the second coefficient value of that coefficient term in at least one reference polynomial function. The reference polynomial function is the polynomial function corresponding to the reference polynomial function module.

[0203] Of course, the first loss function also includes a quantification term used to quantify the difference between the predicted energy efficiency ratio predicted by the target relationship model based on the historical load rate and the historical energy efficiency ratio corresponding to the historical load rate. For details, please refer to the introduction of the quantification term and the relevant introduction of Formulas 3 and 4 above, which will not be repeated here.

[0204] The determination of whether the training objective is met can be the convergence of the loss function value of the first loss function or the completion of a set number of training iterations, without any specific restrictions.

[0205] It is understandable that after training the target relationship model and controlling the operation of the chiller unit based on the target relationship model, as the target chiller unit continues to operate, the operating data of the target chiller unit increases, and the relationship between the load rate and energy efficiency ratio of the target chiller unit may also be updated and changed. Therefore, in order to make the target relationship model more accurately reflect the relationship between the load rate and energy efficiency ratio, in any of the above embodiments of this application, the target relationship model can be continuously updated. For example... Figure 8 This diagram illustrates one implementation flow of updating the target relation model in this application, which may include:

[0206] S801 determines the current actual energy efficiency ratio and actual load rate of the target chiller unit.

[0207] Among them, the actual energy efficiency ratio is the energy efficiency ratio of the target chiller unit at the current moment during actual operation; the actual load rate is the load rate of the target chiller unit at the current moment during actual operation.

[0208] S802 stores the actual energy efficiency ratio and actual load rate as the historical energy efficiency ratio and historical load rate of the target chiller unit, respectively, to obtain the second sample data pair corresponding to the current time point.

[0209] It is understandable that steps S801 and S802 can be repeated multiple times to obtain second sample data pairs of the target chiller unit at multiple different time points. The time point corresponding to each second sample data pair is the historical time point at which the second sample data pair was obtained, so that the second sample data pair of the target chiller unit can be obtained at least one historical time point in each time period.

[0210] S803, in response to satisfying the model update condition, update the target relation model based on at least one second sample data pair obtained in the most recent time period.

[0211] The model update condition can be set as needed. For example, the model update condition can include at least one of the following:

[0212] A model update command was detected;

[0213] Determine the current time when the model needs to be updated based on the time period;

[0214] The number of second sample data pairs in the most recent time period exceeds the set number, and the model update time is obtained at the current time.

[0215] The specific implementation of updating the target relation model based on the second sample data obtained in the most recent time period can be similar to the process of training the target relation model mentioned above, and will not be repeated here.

[0216] In this application, considering the limited historical operating data of newly deployed or shortly deployed target chillers, constructing a curve relating load rate and energy efficiency ratio solely based on this data would inevitably fail to accurately reflect the relationship between load rate and energy efficiency ratio within the target chiller. However, this application utilizes the sample energy efficiency ratio and sample load rate of at least one sample chiller of the same model as the target chiller to train at least one reference relationship model. A target relationship model is then constructed based on the reference polynomial functions in each reference relationship model. This target relationship model includes a target polynomial function constructed from the reference polynomial functions in each reference relationship model, and a first correction network module to correct any deviations in the target polynomial function. Furthermore, since the target polynomial function in the target relationship model can reflect the relationship between the energy efficiency ratio and load rate of the target chiller to a certain extent, training the target relationship model using only a limited amount of operating data (historical load rate and historical energy efficiency) is sufficient to ensure that the target relationship model accurately reflects the relationship between load rate and energy efficiency ratio of the target chiller.

[0217] To facilitate understanding of the benefits of this application, we will take the target energy efficiency ratio as the optimal energy efficiency ratio as an example, combined with... Figure 9 Please provide an explanation. Figure 9 An example diagram is shown showing the optimal energy efficiency ratio and the corresponding target load rate determined based on the target relationship model trained in this application.

[0218] exist Figure 9 Each point in the table represents a data pair consisting of the energy efficiency ratio and load factor at a specific point in time. Figure 9 It can be seen that, due to the relatively short deployment time of the target chilled water sample group, almost no data pairs with a load rate between 60% and 80% were collected for the target chilled water units. However, the relationship curves corresponding to the target relationship model trained by the scheme in this application (such as...) Figure 9 (As shown by the dashed line in the middle), it can also reflect the energy efficiency ratio of the target chiller unit when the load rate is between 60% and 80%.

[0219] Specifically, by differentiating the objective relation function of the objective relation model, the optimal energy efficiency ratio (EER) that the target chiller unit can achieve can be determined. The optimal EER corresponds to the pentagram position in the diagram. Correspondingly, the target load rate required for the optimal EER can be determined based on the objective relation model. This allows for reasonable control of the target chiller unit's operation based on the target load rate, enabling the target chiller unit to achieve that optimal EER.

[0220] This application also provides an electronic device in its embodiments. For example... Figure 10 As shown, it illustrates a schematic diagram of the composition structure of the electronic device, which includes at least a processor 1001 and a memory 1002;

[0221] The memory 1002 is used to store computer programs.

[0222] The processor 1001 is used to execute the computer program to perform the following steps:

[0223] Obtain the target relationship model between the energy efficiency ratio and load rate of the target chiller unit;

[0224] Based on this target relationship model, the target load rate required for the target chiller unit to achieve the target energy efficiency ratio is determined;

[0225] Based on the target load rate, adjust the operating parameters of the target chiller unit;

[0226] The objective relational model includes: an objective polynomial function module and a first correction network module;

[0227] The objective polynomial function module is constructed based on the reference polynomial function module in at least one reference relation model; the reference relation model is trained based on at least one first sample data pair, which includes: the sample energy efficiency ratio and sample load rate of the sample chiller unit;

[0228] The target relationship model is obtained by training the target polynomial function and the first correction network module based on the second sample data pair of the target chiller unit at at least one historical time point. The second sample data pair includes the historical energy efficiency ratio and historical load rate of the target chiller unit at that historical time point.

[0229] For details regarding the specific implementation of the steps executed by the processor, please refer to the relevant descriptions of the corresponding embodiments of the chiller unit operation control method and model training method above, which will not be repeated here.

[0230] It is understood that the electronic device may also include a display unit 1003 and an input unit 1004.

[0231] Of course, the electronic device can also have more than Figure 10 There are no restrictions on the number of components, whether more or fewer.

[0232] This application also provides a computer program product, including computer-readable instructions, which, when executed on an electronic device, cause the electronic device to implement any of the chiller unit operation control methods or model training methods provided in this application.

[0233] This application also provides a computer-readable storage medium that carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement any of the chiller unit operation control methods or model training methods provided in this application.

[0234] It should also be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. In addition, in the device embodiment drawings provided in this application, the connection relationship between modules indicates that they have a communication connection, which can be implemented as one or more communication buses or signal lines.

[0235] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware, or it can be implemented by special-purpose hardware including application-specific integrated circuits, special-purpose CPUs, special-purpose memory, special-purpose components, etc. Generally, any function performed by a computer program can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be diverse, such as analog circuits, digital circuits, or special-purpose circuits. However, for this application, software program implementation is more often the preferred implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a readable storage medium, such as a computer floppy disk, USB flash drive, mobile hard disk, ROM, RAM, magnetic disk, or optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, training equipment, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0236] In the above embodiments, the implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, in the form of a computer program product.

[0237] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, training device, or data center to another website, computer, training device, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a training device or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

Claims

1. A method for controlling the operation of a chiller unit, comprising: Obtain the target relationship model between the energy efficiency ratio and load rate of the target chiller unit; Based on the target relationship model, the target load rate required for the target chiller unit to achieve the target energy efficiency ratio is determined; Based on the target load rate, adjust the operating parameters of the target chiller unit; The target relation model includes: a target polynomial function module and a first correction network module; The objective polynomial function module is constructed based on the reference polynomial function module in at least one reference relation model; the reference relation model is trained based on at least one first sample data pair, the first sample data pair including: the sample energy efficiency ratio and sample load rate of the sample chiller unit; The target relationship model is obtained by training the target polynomial function and the first correction network module with parameters based on the second sample data pair of the target chiller unit at at least one historical time point. The second sample data pair includes the historical energy efficiency ratio and historical load rate of the target chiller unit at the historical time point.

2. In the chiller unit operation control method according to claim 1, the polynomial functions corresponding to the target polynomial function module and the reference polynomial function module are both quadratic polynomial functions; The first correction network module is a residual network module.

3. The chiller unit operation control method according to claim 1, wherein the first loss function used to train the target relationship model includes: The first regularization term includes the third derivative of the target relation function corresponding to the target relation model.

4. The chiller unit operation control method according to claim 3, wherein the first loss function further includes: The second regularization term; The second regularization term includes: the difference between the first coefficient value of the coefficient term in the objective polynomial function and the mean value of the coefficients corresponding to the coefficient term; Wherein, the target polynomial function is the polynomial function corresponding to the target polynomial function module; The mean value of the coefficients corresponding to the coefficient terms is the average value of the second coefficient values ​​of the coefficient terms in at least one reference polynomial function; the reference polynomial function is the polynomial function corresponding to the reference polynomial function module.

5. The chiller unit operation control method according to claim 1, wherein the reference relationship model includes the reference polynomial function module and the second correction network module; in, The second loss function used to train the reference relation model includes a third regularization term, which includes the third derivative of the reference relation function corresponding to the reference relation model.

6. The chiller unit operation control method according to any one of claims 1 to 4 further includes: Determine the current actual energy efficiency ratio and actual load rate of the target chiller unit; The actual energy efficiency ratio and the actual load rate are stored as the historical energy efficiency ratio and historical load rate of the target chiller unit, respectively, to obtain the second sample data pair corresponding to the current time point; In response to the satisfaction of the model update conditions, the target relationship model is updated based on at least one second sample data pair obtained in the most recent time period.

7. A model training method, comprising: At least one trained reference relation model is obtained, the reference relation model comprising: a reference polynomial function module, the reference relation model being trained based on at least one first sample data pair, the first sample data pair comprising: the sample energy efficiency ratio and sample load rate of the sample chiller unit; Based on the reference polynomial function module of the aforementioned reference relationship model, the target polynomial function module is determined. Construct a target relation model to be trained, the target relation model including: the target polynomial function module to be trained and the first correction network module to be trained; Obtain at least one second sample data pair of the target chiller unit in the most recent time period, the second sample data pair including: the historical energy efficiency ratio and historical load rate of the target chiller unit at a historical time point; Based on the at least one second sample data pair, the parameters of the target polynomial function module and the first correction network module in the target relation model are trained and adjusted to obtain the trained target relation model.

8. The model training method according to claim 7, wherein the step of training and adjusting the parameters of the target polynomial function module and the first correction network module in the target relation model based on the at least one second sample data pair includes: Using the historical load rate of the target chiller unit as training data and the historical energy efficiency ratio of the target chiller unit as the training objective, the parameters of the target polynomial function module and the first correction network module in the target relationship model are trained and adjusted in combination with the first loss function. The first loss function includes at least one of a first regularization term and a second regularization term; Wherein, the first regularization term includes: the third derivative of the target relation function corresponding to the target relation model; The second regularization term includes: the difference between the first coefficient value of the coefficient term in the target polynomial function and the mean value of the coefficient corresponding to the coefficient term, wherein the target polynomial function is the polynomial function corresponding to the target polynomial function module; the mean value of the coefficient corresponding to the coefficient term is the average value corresponding to the second coefficient value of the coefficient term in at least one reference polynomial function, wherein the reference polynomial function is the polynomial function corresponding to the reference polynomial function module.

9. The model training method according to claim 7, wherein obtaining at least one trained reference relation model comprises: Obtain at least one first sample data pair corresponding to each of the sample chiller units; Using the sample load rate corresponding to the chilled water sample group as training data and the sample energy efficiency ratio corresponding to the chilled water sample group as the training objective, a reference relationship model is trained by combining the second loss function to obtain a trained reference relationship model corresponding to at least one sample chiller unit. The second loss function includes a third regularization term, which includes the third derivative of the reference relation function corresponding to the reference relation model.

10. An electronic device, comprising: Memory and processor; The memory is used to store computer programs; The processor is configured to execute the computer program to perform the following steps: Obtain the target relationship model between the energy efficiency ratio and load rate of the target chiller unit; Based on the target relationship model, the target load rate required for the target chiller unit to achieve the target energy efficiency ratio is determined; Based on the target load rate, adjust the operating parameters of the target chiller unit; The target relation model includes: a target polynomial function module and a first correction network module; The objective polynomial function module is constructed based on the reference polynomial function module in at least one reference relation model; the reference relation model is trained based on at least one first sample data pair, the first sample data pair including: the sample energy efficiency ratio and sample load rate of the sample chiller unit; The target relationship model is obtained by training the target polynomial function and the first correction network module with parameters based on the second sample data pair of the target chiller unit at at least one historical time point. The second sample data pair includes the historical energy efficiency ratio and historical load rate of the target chiller unit at the historical time point.

Images