Method and apparatus for predicting load information of building heating ventilation system

Abstract

This invention provides a method and apparatus for predicting load information of a building HVAC system, which can be applied to the fields of smart building and building energy efficiency control technology. The method includes: obtaining an initial prediction model for predicting load information based on multiple prior operating information of the building HVAC system; the initial prediction model having multiple initial regression coefficients indicating the importance of each of the prior operating information, the multiple initial regression coefficients being obtained based on a loss function including prediction error information and constraint information; responding to receiving current operating information, updating the multiple initial regression coefficients based on the current error between the current load information in the current operating information and the current predicted load information output by the initial prediction model, thus obtaining a prediction model; and updating the predicted load information output by the prediction model based on the prior errors and correction coefficients of multiple prior times in the prior period, thus obtaining updated load information, which serves as the load prediction result for the building HVAC system.

Description

Technical Field

[0001] This invention relates to the fields of smart building and building energy efficiency control technology, specifically to a method and apparatus for predicting load information of building heating, ventilation and air conditioning systems. Background Technology

[0002] For load forecasting of building energy efficiency equipment, the relevant technologies mainly predict the future load of building energy efficiency equipment by capturing long-term data trends. These include time series analysis methods (such as autoregressive integral moving average models and seasonal autoregressive integral moving average models) and traditional machine learning methods (such as support vector machines, random forests, and gradient boosting machines).

[0003] However, in new building projects, building energy efficiency equipment is not yet fully operational, and data acquisition equipment is not fully installed, resulting in limited historical data. Related technologies rely on a large amount of historical data, leading to data scarcity or cold start and generalization issues in new scenarios. This results in low efficiency and accuracy of energy efficiency management and equipment control, making them difficult to apply to actual engineering scenarios such as building renovations, new projects, and rapid adjustments to operating strategies. Summary of the Invention

[0004] In view of the above problems, the present invention provides a method and apparatus for predicting load information of building heating, ventilation and air conditioning systems.

[0005] According to a first aspect of the present invention, a method for predicting load information of a building heating, ventilation, and air conditioning (HVAC) system is provided, comprising: obtaining an initial prediction model for predicting load information based on multiple prior operating information of the building HVAC system, the initial prediction model having multiple initial regression coefficients for indicating the importance of each of the multiple prior operating information, the multiple initial regression coefficients being obtained based on a loss function including prediction error information and constraint information, the constraint information being used to adjust the constraint strength on the multiple initial regression coefficients according to the prediction information, the multiple prior operating information including prior water supply and return information and prior environmental information; in response to receiving current operating information, updating the multiple initial regression coefficients based on the current error between the current load information in the current operating information and the current predicted load information output by the initial prediction model, thereby obtaining a prediction model; and updating the predicted load information output by the prediction model based on the prior errors of multiple prior times in the prior period and the correction coefficients of the prior errors, thereby obtaining updated load information as the load prediction result of the building HVAC system.

[0006] A second aspect of the present invention provides a device for predicting load information of a building heating, ventilation, and air conditioning (HVAC) system, comprising: an information determination module, configured to obtain an initial prediction model for predicting load information based on multiple prior operating information of the building HVAC system, the initial prediction model having multiple initial regression coefficients for indicating the importance of each of the multiple prior operating information, the multiple initial regression coefficients being obtained based on a loss function including prediction error information and constraint information, the constraint information being used to adjust the constraint strength on the multiple initial regression coefficients according to the prediction information, the multiple prior operating information including prior water supply and return information and prior environmental information; a coefficient update module, configured to update the multiple initial regression coefficients based on the current error between the current load information in the current operating information and the current predicted load information output by the initial prediction model in response to receiving current operating information, thereby obtaining a prediction model; and an information update module, configured to update the predicted load information output by the prediction model based on the prior errors of multiple prior times in the prior time period and the correction coefficients of the prior errors, thereby obtaining updated load information as the load prediction result of the building HVAC system.

[0007] A third aspect of the present invention provides an electronic device comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0008] A fourth aspect of the present invention also provides a computer-readable storage medium having a computer program or instructions stored thereon, wherein the computer program or instructions, when executed by a processor, implement the steps of the above-described method.

[0009] A fifth aspect of the present invention also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method.

[0010] According to embodiments of the present invention, by constructing a loss function based on constraint information containing prediction error information and dynamically adjustable strength to solve multiple initial regression coefficients, overfitting can be effectively suppressed under limited data conditions, resulting in a robust initial prediction model. This allows the initial regression coefficients to be updated in real time based on the current error, enabling the prediction model to have online incremental learning capabilities. This allows the model to adapt to the nonlinear changes in the environment and load of newly constructed projects, continuously track the dynamic characteristics of the system, avoid dependence on large amounts of historical data, and then perform feedback calibration on the prediction results based on historical prior errors and their correction coefficients. This dynamically compensates for system deviations, improves the accuracy and time-series adaptability of load prediction results, and solves the technical challenge of adaptive load prediction for building HVAC systems under limited data conditions with low computational complexity. Attached Figure Description

[0011] The above-described features, other objects, and advantages of the present invention will become clearer from the following description of embodiments of the invention with reference to the accompanying drawings, in which:

[0012] Figure 1 An application scenario diagram of the method and apparatus for predicting load information of building heating, ventilation and air conditioning systems according to an embodiment of the present invention is shown;

[0013] Figure 2 A flowchart of a method for predicting load information of a building heating, ventilation, and air conditioning system according to an embodiment of the present invention is shown;

[0014] Figure 3 A flowchart of a method for predicting load information of a building HVAC system according to another embodiment of the present invention is shown;

[0015] Figure 4 A structural block diagram of a building HVAC system load information prediction device according to an embodiment of the present invention is shown;

[0016] Figure 5 A block diagram of an electronic device suitable for implementing a method for predicting load information of a building heating, ventilation, and air conditioning system according to an embodiment of the present invention is shown. Detailed Implementation

[0017] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the invention. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the invention for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0018] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0019] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0020] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0021] The relevant technologies rely on a large amount of historical data, which leads to data scarcity or cold start and generalization problems in new scenarios. The efficiency and accuracy of energy efficiency management and equipment control are low, making them difficult to apply to actual engineering scenarios such as building renovation, new projects, and rapid adjustment of operating strategies.

[0022] In some examples, autoregressive moving average models or seasonal autoregressive integral moving average models are used to extrapolate future load values ​​by analyzing the inherent trends, seasonality, and periodicity of historical data. However, the relationship between load changes and influencing factors is primarily non-linear, making it difficult for linear models to accurately characterize the relationship, thus limiting their effectiveness.

[0023] In some examples, traditional machine learning methods (such as support vector machines, decision trees, and random forests), which use historical load data and related features (weather, time, etc.) as input, learn a mapping function from features to load through algorithms, making them more suitable for handling nonlinear relationships. However, model performance largely depends on the selection of effective features; models such as support vector machines and random forests are not designed to handle strong time series dependencies, requiring the construction of additional lagged features to capture temporal dynamics, resulting in relatively complex prediction logic and significant consumption of hardware and software resources.

[0024] In some examples, based on backpropagation neural networks or radial basis function networks, the relationship between features and loads is predicted by training to learn the complex nonlinear mapping relationship between features and loads. However, the training process requires a large amount of data and is prone to getting stuck in local optima or overfitting. The training process also has high requirements for parameter settings (such as learning rate and network structure).

[0025] In situations where data is scarce, traditional methods struggle to provide sufficiently accurate predictions, resulting in low precision and efficiency in energy management and equipment control, leading to wasted energy or failure to meet user needs in a timely manner.

[0026] To address the aforementioned technical problems, embodiments of the present invention provide a method and apparatus for predicting load information of a building heating, ventilation, and air conditioning (HVAC) system. The method includes: obtaining an initial prediction model for predicting load information based on multiple prior operational information of the building HVAC system. The initial prediction model has multiple initial regression coefficients indicating the relative importance of each of the multiple prior operational information items. The multiple initial regression coefficients are obtained based on a loss function containing prediction error information and constraint information. The constraint information is used to adjust the constraint strength on the multiple initial regression coefficients according to the prediction information. The multiple prior operational information items include prior water supply and return information and prior environmental information. In response to receiving current operational information, updating the multiple initial regression coefficients based on the current error between the current load information in the current operational information and the current predicted load information output by the initial prediction model, thereby obtaining a prediction model. Updating the predicted load information output by the prediction model based on the prior errors at multiple prior times in the prior time period and the correction coefficients of the prior errors, thereby obtaining updated load information, which serves as the load prediction result for the building HVAC system.

[0027] According to embodiments of the present invention, by constructing a loss function based on constraint information containing prediction error information and dynamically adjustable strength to solve multiple initial regression coefficients, overfitting can be effectively suppressed under limited data conditions, resulting in a robust initial prediction model. This allows the initial regression coefficients to be updated in real time based on the current error, enabling the prediction model to have online incremental learning capabilities. This allows it to adapt to nonlinear changes in the environment and load, continuously track the dynamic characteristics of the system, avoid dependence on large amounts of historical data, and then perform feedback calibration on the prediction results based on historical prior errors and their correction coefficients. This dynamically compensates for system deviations, improves the accuracy and time-series adaptability of load prediction results, and solves the technical challenge of achieving adaptive load prediction under limited data conditions with low computational complexity.

[0028] Figure 1 An application scenario diagram of the method and apparatus for predicting load information of building heating, ventilation, and air conditioning systems according to an embodiment of the present invention is shown.

[0029] like Figure 1 As shown, application scenario 100 according to this embodiment may include an information acquisition device 101, a network 102, and a server 103. The network 102 is used as a medium to provide a communication link between the information acquisition device 101 and the server 103. The network 102 may include various connection types, such as wired or wireless communication links or fiber optic cables, etc.

[0030] Users can use the information acquisition device 101 to interact with the server 103 via the network 102 to receive or send messages, etc. The information acquisition device 101 can be a device used to collect operating or status information of various equipment in a building's HVAC system, such as temperature sensors and flow meters installed on heating pipes, temperature and humidity sensors deployed inside and outside the building, solar radiation meters, and heat meters or power meters installed on heating stations or main pipelines. It should be noted that the embodiments of the present invention can be applied to scenarios such as building energy efficiency management, HVAC system control, and heat pump system optimization.

[0031] For example, the information collection device 101 continuously and periodically (e.g., every 15 minutes) collects a set of previously running information and uploads it to the server 103 via the network 102.

[0032] Server 103 may be a processor that carries and executes the calculation logic and algorithm (including model initialization, online updates, and error correction) of the load forecasting method in the embodiments of the present invention. For example, it may be an industrial control computer or server deployed in a building's computer room, running the forecasting algorithm software. Server 103 may receive and store time-series data sent by information acquisition device 101 in real time, forming a training dataset for building the initial forecasting model.

[0033] It should be noted that the method for predicting building HVAC system load information provided in this embodiment of the invention can generally be executed by server 103. Correspondingly, the device for predicting building HVAC system load information provided in this embodiment of the invention can generally be located in server 103. The method for predicting building HVAC system load information provided in this embodiment of the invention can also be executed by a server or server cluster that is different from server 103 and capable of communicating with information acquisition device 101 and / or server 103. Correspondingly, the device for predicting building HVAC system load information provided in this embodiment of the invention can also be located in a server or server cluster that is different from server 103 and capable of communicating with information acquisition device 101 and / or server 103.

[0034] It should be understood that Figure 1 The number of information collection devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of information collection devices, networks, and servers can be included.

[0035] Figure 2 A flowchart of a method for predicting load information of a building heating, ventilation, and air conditioning system according to an embodiment of the present invention is shown.

[0036] like Figure 2 As shown, the method for predicting the load information of a building heating, ventilation, and air conditioning system in this embodiment includes operations S210 to S230.

[0037] In operation S210, an initial prediction model for predicting load information is obtained based on multiple prior operational information of the building's HVAC system. The initial prediction model has multiple initial regression coefficients that indicate the relative importance of each of the multiple prior operational information. These multiple initial regression coefficients are obtained based on a loss function that includes prediction error information and constraint information. The constraint information is used to adjust the constraint strength on the multiple initial regression coefficients according to the prediction information. The multiple prior operational information includes prior water supply and return information and prior environmental information.

[0038] In operation S220, in response to receiving the current operating information, based on the current error between the current load information in the current operating information and the current predicted load information output by the initial prediction model, multiple initial regression coefficients are updated to obtain the prediction model.

[0039] In operation S230, based on the previous errors and correction coefficients of multiple previous times in the previous period, the predicted load information output by the prediction model is updated to obtain the updated load information, which serves as the load prediction result for the building HVAC system.

[0040] In embodiments of the present invention, the prior operational information can be a series of historical information reflecting the historical state of the system and the external environment collected from the building's HVAC system before the prediction time. This prior operational information, after processing, can be used as input features of the model. Multiple initial regression coefficients can be used to indicate the importance of the model's input features to the load prediction results. The loss function can be used to measure the quality of the load prediction results output by the prediction model. Prediction error information is used to fit the model to the data, ensuring the accuracy of the load prediction results output by the model. Constraint information can be used to control model complexity, prevent overfitting with limited data, and the constraint strength can be dynamically adjusted based on the prediction results.

[0041] The preceding supply and return water information may include the preceding supply and return water temperature difference and flow rate; the preceding environmental information may include outdoor temperature, outdoor humidity, and solar radiation information. Current load information may be the system operating data collected by the acquisition equipment at the current moment and the actual load value. The preceding error may be a historical prediction error sequence within a historical period. The correction coefficient may be a dynamically calculated weighting factor used to correct the current prediction, ensuring that the correction is appropriate to the current error magnitude and the building's HVAC system operating information.

[0042] Taking the cloud-edge collaborative central air conditioning system cooling load prediction system based on newly built factory buildings as an example, after receiving a small amount of historical data from the first week of operation of the central air conditioning system, a loss function (including the sum of squared errors of prediction error information and dynamic constraint information) is constructed, and an initial prediction model containing the initial regression coefficients of each parameter is trained. At this time, the coefficients of outdoor temperature and chilled water flow rate in the model prediction results are relatively large, which are the main influencing factors affecting the prediction results.

[0043] The system receives a set of current operating information at preset time intervals, predicts the cooling load, compares the current prediction with the measured value to obtain the current error, and updates the regression coefficients using a gradient descent method with a preset step size. For example, when entering the transitional season, the impact of outdoor humidity increases, and the model can automatically increase the regression coefficient corresponding to "outdoor humidity" through online learning.

[0044] Before outputting the load forecast result for the next moment, the weighted average of the forecast error in the past preset period is used as a correction term. Combined with the correction coefficient, the output result of the forecast model is calibrated to obtain the final updated load information (i.e., the cooling load forecast value), which is then sent to the control system of the plant building to adjust the operation strategy of the chiller and water pump in advance.

[0045] For example, the Least Absolute Shrinkage and Selection Operator (LASSO) can be used as the initial prediction model to select or optimize the input features, thereby capturing the relationship between operational information (such as supply and return water temperature difference, supply and return water flow rate, outdoor temperature, outdoor humidity, and solar radiation) and load prediction results when the amount of data is small.

[0046] For example, the minimum absolute shrinkage and selection operator regression algorithm can iteratively update the regression coefficients using coordinate descent or minimum angular regression. In each iteration, each regression coefficient is updated while keeping other coefficients fixed, and the magnitude of the coefficients is constrained by a regularization term.

[0047] It is understood that in this embodiment of the invention, multiple prior operational information are used to solve for multiple initial regression coefficients with feature selection capabilities through a loss function that includes prediction error information and constraint information, forming an initial prediction model. When the current operational information arrives, the current error triggers online updates of the model coefficients, keeping the model adaptive; thus, based on historical prior errors and dynamic correction coefficients, the prediction results output by the model are calibrated in real time, generating the final updated load information. Throughout the process, both constraint information and correction coefficients have dynamic adjustment capabilities, linked with the prediction error, realizing a two-layer prediction accuracy and efficiency guarantee mechanism.

[0048] According to embodiments of the present invention, by constructing a loss function based on constraint information containing prediction error information and dynamically adjustable strength to solve multiple initial regression coefficients, overfitting can be effectively suppressed under limited data conditions, resulting in a robust initial prediction model. This allows the initial regression coefficients to be updated in real time based on the current error, enabling the prediction model to have online incremental learning capabilities. This allows it to adapt to nonlinear changes in the environment and load, continuously track the dynamic characteristics of the system, avoid dependence on large amounts of historical data, and then perform feedback calibration on the prediction results based on historical prior errors and their correction coefficients. This dynamically compensates for system deviations, improves the accuracy and time-series adaptability of load prediction results, and solves the technical challenge of achieving adaptive load prediction under limited data conditions with low computational complexity.

[0049] As you can understand, the above text has already explained how to obtain the load prediction results of the building HVAC system from the overall technical solution. The following text will explain in detail how to obtain the initial prediction model.

[0050] According to an embodiment of the present invention, the upstream supply and return water information includes the supply and return water temperature difference and flow rate information, and the upstream environmental information includes outdoor temperature information, solar radiation information and outdoor humidity information; based on multiple upstream operating information of the building HVAC system, an initial prediction model for predicting load information is obtained, including: combining the supply and return water temperature difference, flow rate information, outdoor temperature information, solar radiation information and outdoor humidity information and their respective initial regression coefficients to obtain the initial prediction model.

[0051] In embodiments of the present invention, the supply and return water temperature difference can refer to the temperature difference between the fluid in the supply pipe and the return pipe in the water circulation loop of a building HVAC system. Flow rate information can refer to the volume or mass of fluid flowing through a pipe per unit time in the HVAC system water circulation loop. Initial regression coefficients can include the regression coefficients corresponding to the supply and return water temperature difference, flow rate information, outdoor temperature information, solar radiation information, and outdoor humidity information, respectively.

[0052] For example, by integrating the supply and return water temperature difference, flow rate information, outdoor temperature information, solar radiation information, and outdoor humidity information, along with their respective initial regression coefficients, an initial prediction model can be obtained, as shown in the following formula (1):

[0053] (1);

[0054] in, It could be the temperature difference between the supply and return water at time t. It could be the flow rate at time t. It could be the outdoor temperature at time t. It could be the solar radiation at time t. It could be the humidity at time t. These can be multiple initial regression coefficients calculated using minimum absolute shrinkage and selection operators. It can be the predicted load information output by the initial prediction model at time t.

[0055] According to an embodiment of the present invention, the method for predicting load information of a building heating, ventilation, and air conditioning system further includes: taking the average prediction error of the initial prediction model in the current time window as the first error and the cumulative average prediction error of the previous operating information as the second error, and determining the relative ratio between the first error and the second error; and obtaining constraint information based on the relative ratio, the number of previous operating information, and the initial constraint information obtained by cross-validation based on the previous operating information.

[0056] In embodiments of the present invention, the first error may refer to the average difference between the predicted load and the actual load of the initial prediction model within the most recent sliding time window (e.g., the past 24 hours), for example, the average prediction error of the current time window. The second error may refer to the global average prediction error obtained when training and evaluating the model using all available prior operational information (historical data), for example, the cumulative average prediction error of prior operational information. The relative ratio can be used to measure the degree of deviation of the model's current short-term predicted performance from its long-term average performance.

[0057] The number of prior running information can be the total number of data points or sample sizes of prior running information that have been collected and used for modeling. Initial constraint information can refer to a basic regularization strength parameter value determined from prior running information through cross-validation during the model initialization phase. This parameter serves as initial information to balance model fit and complexity under the current data conditions.

[0058] Taking a central air conditioning system in a commercial building as an example, a cloud-edge collaborative load forecasting system can be deployed. During system initialization, cross-validation is performed using the previous two weeks' operational data (supply and return water temperatures, flow rates, and outdoor meteorological parameters) to determine initial constraints. During continuous operation, the first error (the average forecast error over the past 24 hours) is calculated every 24 hours, and the second error (the cumulative average error since system commissioning) is updated. The relative ratio between the first and second errors is then calculated, and combined with the number of accumulated operational data points (total data points), the constraint information for the next time window is dynamically calculated using a preset exponential decay formula. For example, if continuous high temperatures cause the first error to increase and the relative ratio to be much greater than 1, the constraint information value can be automatically increased, making the forecasting model focus more on core characteristics under abnormal weather conditions and avoiding interference from abnormal fluctuations. When the temperature returns to normal, the relative ratio is reduced, thus obtaining the constraint information.

[0059] To ensure compatibility between the constraint information and the prediction results of the initial prediction model, the initial constraint information can be dynamically optimized or adjusted. For example, traditional constraint information can be transformed from fixed values ​​into dynamic constraint information that changes over time. The initial constraint information can be initial regularization parameters obtained through cross-validation with a small amount of historical data, and can be denoted as... The constraint information is determined using the following formula (2):

[0060] (2);

[0061] in, It can be the constraint information at time t. It can be initial constraint information, which can be obtained through cross-validation using a small amount of historical data. It could be the first error, which is the average error of the model prediction within the current sliding window. This can be a second error, such as the historical cumulative average error. It can be a sensitivity factor (positive number) used to control the intensity of the impact of error fluctuations on constraint information. This could be the number of prior learning data points, i.e., the total number of samples learned at time t. As the sample size increases, the model confidence improves, which can be achieved through... The penalty intensity should be appropriately reduced, allowing for the introduction of more features.

[0062] The first error RMSE can be calculated using the following formula (3):

[0063] (3);

[0064] in, It can be the total number of sample data. It could be the actual load value of the i'th sample; It can be the predicted load value of the i'th sample.

[0065] In traditional regression methods, constraint information is usually determined once before training through cross-validation and remains fixed throughout the model's lifecycle. This has serious limitations for building HVAC systems that are dynamically affected by multiple factors such as seasons, weather, equipment status, and usage patterns: the pre-set constraint information is difficult to adapt to the slow time-varying characteristics of the system (such as seasonal changes) or rapid disturbances (such as extreme weather), which can easily lead to overfitting (too few constraints) or underfitting (too many constraints) of the prediction model after changes.

[0066] The embodiments of the present invention introduce a dynamic calculation mechanism based on a first error, a second error, and their relative ratio, which enables the constraint information to respond to the adaptive changes in the real-time performance of the model. When the model performance fluctuates, the constraint strength can be automatically adjusted, thereby achieving a dynamic balance between preventing overfitting and ensuring fitting ability, and improving the environmental adaptability and prediction robustness of the model in long-term operation.

[0067] According to an embodiment of the present invention, the method for predicting load information of a building heating, ventilation, and air conditioning system further includes: combining the difference between the current load information and the current predicted load information, constraint information, and multiple initial regression coefficients to obtain a loss function; inputting the previous supply and return water information and the previous environmental information into the initial prediction model; and determining multiple regression coefficients of the initial prediction model as multiple initial regression coefficients when the function value of the loss function indicates that the error of the predicted load information output by the initial prediction model is less than the error threshold.

[0068] In embodiments of the present invention, the supply and return water temperature difference in the preceding supply and return water information can be the temperature difference between the current moment and the previous moment (at the previous moment). Constraint information can be used to minimize the loss function. The initial regression coefficients of the initial prediction model are optimized as shown in the following formula (4):

[0069] (4);

[0070] in, It can be the actual load information of the building's HVAC system detected at time t, where T is the total number of times. It can be the predicted load information output by the initial prediction model at time t. It can be constraint information. It can be the i-th regression coefficient selected from multiple initial regression coefficients.

[0071] Taking the initial prediction model of a building cluster cooling load prediction system as an example, the system can obtain the preceding operating information of the building cluster's cooling load equipment during the first week of operation, including the preceding supply and return water temperature difference, preceding supply and return water flow rates, outdoor temperature and humidity, solar radiation intensity, and the corresponding total cooling load. An error threshold based on the root mean square error is set, a set of regression coefficients is initialized, and initial values ​​for constraint information are set. Historical data is input into the initial prediction model, and the difference between the predicted value output by the initial prediction model and the actual load value is calculated. The current loss function value is calculated in conjunction with the constraint information, and the regression coefficients are iteratively updated using the coordinate descent method to minimize the loss function. After each iteration, the prediction error of the model on the entire training set can be evaluated. If the error is less than the error threshold, the iteration stops, and multiple regression coefficients of the current initial prediction model are determined as multiple initial regression coefficients.

[0072] When building a predictive model, the selection of input variables or features is crucial. Traditional methods often rely on domain knowledge for screening or involve tedious stepwise regression experiments, which are subjective and inefficient. The embodiments of this invention optimize the loss function to determine the initial regression coefficients, while automatically reducing the coefficients of unimportant features to zero. This achieves end-to-end automated feature selection, reduces reliance on expert experience, and improves modeling efficiency and objectivity.

[0073] According to an embodiment of the present invention, when the error of the predicted load information output by the initial prediction model, as indicated by the function value of the loss function, is less than an error threshold, multiple regression coefficients of the initial prediction model are determined as multiple initial regression coefficients, including: keeping the current values ​​of the remaining coefficients in the initial coefficient vector, excluding the target coefficient, unchanged; determining the target solution of the target coefficient when the function value is at a valley value, wherein the initial coefficient vector is obtained by initializing multiple initial regression coefficients to zero; updating the initial value of the target coefficient using the target solution to obtain the updated coefficient vector; and using the updated coefficient vector as multiple initial regression coefficients when the difference between the function value at the current time and the function value at the previous time is less than a difference threshold.

[0074] In embodiments of the present invention, the initial coefficient vector can be a vector composed of multiple initial regression coefficients arranged in order, and all coefficients can be initialized to zero. The target coefficient can be a specific regression coefficient that is currently selected and optimized in a single iteration. In the same round of iteration, the other coefficients in the vector, except for the target coefficient, are considered as fixed values.

[0075] A function value at its trough refers to the minimum (valley) value of the loss function when the target coefficients are used as variables and the values ​​of the remaining coefficients in the initial coefficient vector are fixed. The target solution can be the specific numerical solution of the target coefficients calculated when the function value is at its trough. The updated coefficient vector can be obtained by replacing the corresponding values ​​in the initial coefficient vector (or the vector from the previous iteration) with the target solution after optimizing the current target coefficients, thus obtaining a coefficient vector with partially updated coefficients. It can be understood that a difference value less than the difference threshold is a sufficient condition or indicator signal for determining that the model's prediction accuracy meets the standard (error less than the error threshold).

[0076] Taking the integration of a lightweight prediction module into the edge smart gateway of the heating system in an industrial plant as an example, the prediction module is used to execute the method for predicting the load information of the building HVAC system in this invention. A training set and loss function can be constructed using the operating data from the first three days of the initial run. The prediction module on the embedded processor executes the coordinate descent algorithm. Starting with an initial coefficient vector of all zero values, in a resource-constrained environment, a sequential iterative approach is used to select target coefficients. For each coefficient, the algorithm calculates the target solution that reduces the objective function based on fixed other coefficients and the current data, and updates the coefficient vector. Due to the small amount of data, the change in the loss function value is calculated after each iteration. When the change is consistently less than the difference threshold set according to hardware accuracy, the model training is considered complete, and the obtained coefficient vector is stored as fixed parameters, i.e., the initial regression coefficients for that specific site, for subsequent local real-time load prediction and control.

[0077] Traditional gradient (derivative)-based optimization algorithms (such as standard gradient descent) are difficult to apply directly to loss functions that contain absolute value terms but are not differentiable at zero, such as the above formula (4). If a complex quadratic programming solver is used, the computational complexity becomes too high when the data dimension is slightly high or online updates are required, making it unsuitable for embedded or real-time scenarios.

[0078] To address this technical problem, embodiments of the present invention employ a strategy of fixing other coefficients and optimizing only one variable at a time. This transforms a non-differentiable multivariate problem into a series of analytically solvable (through soft thresholding) univariate problems, improving the reliability of sparse modeling with limited data. Simultaneously, the soft thresholding operator automatically sets coefficients with absolute values ​​less than the threshold to zero. Therefore, during iteration, the coefficients of unimportant features are dynamically and precisely set to zero. This characteristic of simultaneous iteration and feature selection makes the entire optimization process consistent with the goal of building a sparse model, resulting in higher efficiency compared to the traditional two-stage method of first training the model and then pruning.

[0079] According to an embodiment of the present invention, when the error of the predicted load information output by the initial prediction model, as indicated by the function value of the loss function, is less than an error threshold, multiple regression coefficients of the initial prediction model are determined as multiple initial regression coefficients, including: selecting multiple target active information from the input information based on the correlation between the initial residual and the input information composed of prior water supply and return information and prior environmental information, wherein the initial residual is obtained by assigning the values ​​of multiple regression coefficients and the predicted load information to zero; determining an update direction with the same correlation between the multiple target active information and the initial residual, and updating multiple regression coefficients corresponding to the multiple target active information along the update direction; when the sum of the absolute values ​​of the updated multiple regression coefficients is greater than a preset threshold, the updated multiple regression coefficients are determined as multiple initial regression coefficients of the initial prediction model.

[0080] In embodiments of the present invention, all regression coefficients and the initial values ​​of the predicted load information can be set to zero. At this point, the model prediction value is zero, and the initial residual can be the difference vector between the actual load information (i.e., the target value vector in the previous running information) and this zero-predicted value, which is equal to the actual load value vector itself at the initial moment. Relevance can refer to the correlation between each input feature (each column of the input information) and the current residual vector (usually measured using cosine similarity or inner product). The absolute value of the correlation can characterize the potential contribution of that feature to explaining the current residual. Target active information can be a subset of features selected from all input information (features) based on their correlation with the current initial residual (or the updated residual). The features in the feature subset are those currently considered to contribute the most to explaining the residual and have the same correlation (i.e., consistent direction).

[0081] The update direction can be calculated after determining the target active information set, so that the residual correlation of all active features remains equal and decreases synchronously. It can be a vector direction in the feature space. Updating the coefficients along this direction can efficiently and reasonably reduce the correlation between all active features and residuals.

[0082] For example, in the intelligent controller of a variable air volume (VAV) terminal unit, an initial predictive model can be established to predict the relationship between room load and damper opening. Using the operational data from the first five days after deployment, the algorithm is initialized by setting a small preset threshold to control model complexity, with the residuals representing the actual damper opening values. In each iteration, the correlation between each input feature (such as the deviation between the setpoint and measured values ​​of room temperature and humidity, and supply air temperature) and the current residual is calculated, and the most relevant feature is selected as the target active information. The update direction is then calculated, and the corresponding coefficients are fine-tuned to ensure that the correlation between these features and the residual decreases equally. At each step, the sum of the absolute values ​​of the coefficients is checked to ensure it exceeds a preset threshold. The iteration terminates if the sum of the absolute values ​​exceeds the preset threshold. The resulting sparse coefficient vector is then written into the controller's fixed memory as the personalized initial regression coefficients for the terminal unit.

[0083] Traditional direct optimization methods, such as coordinate descent, can only yield one model solution for a given set of constraints. Selecting the optimal constraints requires multiple trials and cross-validation, resulting in high computational costs. The embodiments of this invention calculate the update direction, ensuring that the correlation between all active target information and the residuals decreases at an equal rate. This allows the residual vector to move at equal angles towards the bisectors of the angles between each active feature within the feature-spanned space. This ensures that the sequence of feature introduction and the path of coefficient updates are continuous and stable, reducing fluctuations during iteration and making the model construction process more rational and interpretable.

[0084] According to an embodiment of the present invention, based on the current error between the current load information in the current operating information and the current predicted load information output by the initial prediction model, multiple initial regression coefficients are updated to obtain a prediction model, including: the sum of the product of the current error and the current water supply and return information and the current environmental information and the multiple regression coefficients at the previous time point as multiple updated regression coefficients, and the initial prediction model is updated using the multiple updated regression coefficients to obtain a prediction model.

[0085] In an embodiment of the present invention, the multiple regression coefficients at the previous time step can refer to the initial coefficient vector of the model before the current update.

[0086] For example, incremental learning can be used to optimize or update the initial prediction model. When inputting the current time-to-time information into the prediction model, the regression coefficients from previous time steps can be fine-tuned, allowing the prediction model to gradually adapt to the latest acquired data. The incremental learning update method is shown in the following formula (5):

[0087] (5);

[0088] in, It could be the regression coefficient updated at time t. It can be the actual load information of the building's HVAC system detected at time t. It can be the predicted load information output by the initial prediction model at time t. It can be the value of the input feature (such as the temperature difference between supply and return water, flow rate, etc.) at time t. It refers to the learning rate, which controls the magnitude of each update. Incremental learning, by continuously refining the regression coefficients, enables the predictive model to quickly adjust after receiving the latest data, providing more accurate load forecasts.

[0089] For example, a central air conditioning management system deployed in the cloud maintains an independent predictive model for each chiller unit. During online operation, the edge controller of each unit collects current supply and return water information (temperature difference, flow rate) and current environmental information (outdoor temperature, radiation, outdoor humidity) every minute and uploads them to the cloud. The cloud server performs online updates, using the model to calculate the current predicted load information and comparing it with the measured current load information to obtain the current error; it then multiplies the current error with the current feature vector, multiplies the product by a fixed step size factor, and sums it with multiple regression coefficients of the unit stored in the database from previous time steps; the summation result is written back to the database as multiple updated regression coefficients for the unit, completing the model fine-tuning and obtaining the predictive model.

[0090] Traditional models, once trained and deployed, have fixed parameters. However, building HVAC systems are affected by seasonal changes, equipment aging, and changes in usage patterns, causing their thermodynamic characteristics to drift conceptually. This leads to static models gradually deviating from reality, resulting in increasingly larger prediction errors. The embodiments of this invention utilize a simplified update rule composed of product and summation to make minor, continuous, and online adjustments to the model using the latest acquired data. This allows the model parameters to gradually drift along with changes in system characteristics, achieving real-time optimization of predictive capabilities. This solves the performance degradation of static models caused by time-varying environments and improves the accuracy and adaptability of long-term predictions.

[0091] According to an embodiment of the present invention, based on the preceding errors and correction coefficients of the preceding errors at multiple preceding times in the preceding period, the predicted load information output by the prediction model is updated to obtain updated load information, including: using preset weights to perform a weighted average of the preceding errors at multiple preceding times to obtain an average error; and combining the predicted load information, the average error, and the correction coefficient to obtain updated load information.

[0092] In embodiments of the present invention, preset weights can be used to define the proportion of previous errors at different historical times when calculating the average error. The average error can be the error value obtained by weighting the previous errors at multiple previous times in the previous time period using preset weights. Updating load information can be the optimized load forecast value obtained by combining the predicted load information with the average error scaled by the correction factor.

[0093] For example, a bias update method based on the error feedback mechanism of the previous time step (historical time step) can improve the accuracy of the prediction model in the future time step (later time step) by updating the historical prediction error with weight.

[0094] The prediction error at time t can be calculated. As shown in the following formula (6):

[0095] (6);

[0096] in, It can be the actual load information of the building's HVAC system detected at time t. It can be the predicted load information output by the initial prediction model at time t.

[0097] For example, a weighted average of the errors over the past week can be used to obtain the average error of the prediction error at multiple points in time. As shown in the following formula (7):

[0098] (7);

[0099] in, This is the preset weight for the prediction error at time t, which can be set based on time decay, where T is the total number of time points. Then, based on the weighted average error... The forecast results are then corrected or updated to obtain updated load information. As shown in the following formula (8):

[0100] (8);

[0101] in, It is the correction factor for the average error, which controls the magnitude of the correction. It can be the average error.

[0102] For example, in a scenario where multiple heat exchange stations in a regional cooling system are used for load forecasting, the forecasting model deployed at each heat exchange station (site) can output a predicted load information every minute. A leading error queue of 30 samples (i.e., the most recent 30 minutes) can be maintained for each station. Using a preset weighting of exponential decay, the errors in the queue are weighted and averaged to calculate the average error, ensuring that errors from the most recent minutes have a greater impact on the average. Based on the stability of the error sequence for each station, a correction coefficient is dynamically set (for stations with large error fluctuations, the coefficient can be set smaller; for stations with small fluctuations and stable deviations, the coefficient can be set larger). The original predicted load information for a station can be added to the product of the average error and the correction coefficient to obtain the updated load information for that station.

[0103] Related technologies employ complex online model retraining, which incurs high computational costs. While effective, retraining is inefficient for concept drift caused by slowly changing system characteristics (e.g., seasonal changes, equipment performance degradation), making it difficult to meet real-time engineering requirements. The embodiments of this invention, by adding a lightweight feedback correction loop, compensate for biases by calibrating the model output without altering the model's internal parameters. This achieves real-time tracking and compensation for systematic, slowly time-varying biases in model predictions, improving the overall accuracy of the prediction results.

[0104] According to an embodiment of the present invention, the method for predicting load information of a building HVAC system further includes: updating the correction coefficient by the ratio between the current prediction error and the fusion error information, and using the updated correction coefficient as the correction coefficient at a later time to update the predicted load information output by the prediction model, wherein the fusion error information is obtained by combining the absolute value of the current load information and the current prediction error.

[0105] In embodiments of the present invention, the fused error information can be a comprehensive error metric obtained by combining the absolute value of the current load information with the current prediction error. For example, the absolute value of the current prediction error can be added to the absolute value of the current load information, or some form of fusion (such as a weighted average) can be performed to construct a scale that can simultaneously reflect the absolute magnitude and relative importance of the error. The correction coefficient at a later time can be used in the prediction correction process at the next time (later time).

[0106] Correction coefficient for the average error at time t The error can be dynamically adjusted based on the actual error and the predicted error, as shown in the following formula (9):

[0107] (9);

[0108] in, It could be the prediction error at time t.

[0109] Figure 3 A flowchart of a method for predicting load information of a building heating, ventilation, and air conditioning system according to another embodiment of the present invention is shown.

[0110] like Figure 3 As shown, the method for predicting the load information of a building HVAC system includes operations S310 to S350.

[0111] In operation S310, input data and feature extraction are performed.

[0112] Receive and process multiple upstream operational information from the building's HVAC system. Extract features from these upstream operational information to obtain upstream supply and return water information reflecting the system's internal operating status (such as supply and return water temperature difference, flow rate, etc.) and upstream environmental information reflecting external influencing factors (such as outdoor temperature, solar radiation, outdoor humidity, etc.).

[0113] In operation S320, an initial forecasting model is constructed to predict load information.

[0114] An initial prediction model is constructed by minimizing a loss function using extracted features and historical load data. This loss function includes not only prediction error information, which measures prediction accuracy, but also constraint information.

[0115] In operation S330, based on the current error between the current load information and the current predicted load information in the current operating information, multiple initial regression coefficients are updated to obtain the prediction model.

[0116] For example, when receiving current operating information (including current water supply and return information, current environmental information, and actual current load information), the current predicted load information is first calculated by the initial prediction model (or the prediction model updated in the previous round); then the current error between the predicted value and the actual value is calculated, the update amount is calculated based on the current error and the current input information, and multiple initial regression coefficients (or regression coefficients in the previous round) are iteratively updated to obtain an updated prediction model that can track the time-varying characteristics of the system.

[0117] In operation S340, based on the previous errors and correction coefficients of the previous time steps, the predicted load information output by the prediction model is updated to obtain the updated load information.

[0118] For example, by using preset weights to calculate the weighted average of a recent set of previous errors, an average error that can reflect recent typical deviations can be calculated. This average error is then multiplied by a correction coefficient to obtain the final correction amount. The predicted load information directly output by the prediction model is then combined with this correction amount to output calibrated updated load information.

[0119] When operating S350, output the load prediction results of the building's HVAC system.

[0120] Updated load information can be used as load forecast results and output to downstream energy management systems, optimization control platforms, or equipment controllers.

[0121] It should be noted that the method for predicting the load information of building HVAC systems in the embodiments of the present invention has strong versatility and is applicable to various fields that require load prediction, including but not limited to building energy efficiency management, HVAC system control, heat pump system optimization, smart grid management, etc.

[0122] Based on the above-mentioned method for predicting building HVAC system load information, this invention also provides a device for predicting building HVAC system load information. The following will be combined with... Figure 4 The device is described in detail.

[0123] Figure 4 A structural block diagram of a device for predicting load information of a building heating, ventilation, and air conditioning system according to an embodiment of the present invention is shown.

[0124] like Figure 4 As shown, the building HVAC system load information prediction device 400 of this embodiment includes an information determination module 410, a coefficient update module 420 and an information update module 430.

[0125] The information determination module 410 is used to obtain an initial prediction model for predicting load information based on multiple prior operational information of the building's HVAC system. The initial prediction model has multiple initial regression coefficients indicating the importance of each of the multiple prior operational information items. These initial regression coefficients are obtained based on a loss function that includes prediction error information and constraint information. The constraint information is used to adjust the constraint strength on the multiple initial regression coefficients according to the prediction information. The multiple prior operational information items include prior water supply and return information and prior environmental information. In one embodiment, the information determination module 410 can be used to perform the operation S210 described above, which will not be repeated here.

[0126] The coefficient update module 420 is used to update multiple initial regression coefficients in response to receiving current operating information, based on the current error between the current load information in the current operating information and the current predicted load information output by the initial prediction model, to obtain the prediction model. In one embodiment, the coefficient update module 420 can be used to perform the operation S220 described above, which will not be repeated here.

[0127] The information update module 430 is used to update the predicted load information output by the prediction model based on the previous errors and correction coefficients of multiple previous times in the previous period, thereby obtaining updated load information as the load prediction result of the building HVAC system. In one embodiment, the information update module 430 can be used to perform the operation S230 described above, which will not be repeated here.

[0128] According to an embodiment of the present invention, the information determination module 410, coefficient update module 420, and information update module 430 in the prediction device 400 based on building HVAC system load information solve multiple initial regression coefficients by constructing a loss function based on constraint information containing prediction error information and dynamically adjustable intensity. This effectively suppresses overfitting under limited data conditions, resulting in a robust initial prediction model. The initial regression coefficients are then updated in real time based on the current error, enabling the prediction model to have online incremental learning capabilities. This allows it to adapt to nonlinear changes in the environment and load, continuously track the dynamic characteristics of the system, avoid dependence on large amounts of historical data, and further calibrate the prediction results based on historical prior errors and their correction coefficients. This dynamically compensates for system deviations, improves the accuracy and time series adaptability of the load prediction results, and solves the technical problem of achieving adaptive load prediction under limited data conditions with low computational complexity.

[0129] According to an embodiment of the present invention, the above-described apparatus further includes a ratio determination module and a verification module. The ratio determination module is used to determine the relative ratio between the first error and the second error by taking the average prediction error of the initial prediction model in the current time window as the first error and the cumulative average prediction error of the preceding running information as the second error. The verification module is used to obtain constraint information based on the relative ratio, the number of preceding running information, and the initial constraint information obtained by cross-validation based on the preceding running information.

[0130] According to an embodiment of the present invention, the apparatus further includes a combination module and an input module. The combination module is used to combine the difference between the current load information and the current predicted load information, constraint information, and multiple initial regression coefficients to obtain a loss function; the input module is used to input the previous water supply and return information and the previous environmental information into the initial prediction model, and determine the multiple regression coefficients of the initial prediction model as multiple initial regression coefficients when the function value of the loss function indicates that the error of the predicted load information output by the initial prediction model is less than an error threshold.

[0131] According to an embodiment of the present invention, the input module includes: a target solution determination submodule, an update submodule, and an as-used submodule. The target solution determination submodule is used to determine the target solution for the target coefficients when the function value is at a valley value, while keeping the current values ​​of the remaining coefficients in the initial coefficient vector unchanged (excluding the target coefficient). The initial coefficient vector is obtained by initializing multiple initial regression coefficients to zero. The update submodule is used to update the initial values ​​of the target coefficients using the target solution, obtaining the updated coefficient vector. The as-used submodule is used to use the updated coefficient vector as multiple initial regression coefficients when the difference between the function value at the current time and the function value at the previous time is less than a difference threshold.

[0132] According to an embodiment of the present invention, the input module includes: a selection submodule, a direction determination submodule, and a coefficient determination submodule. The selection submodule is used to select multiple target active information from the input information based on the correlation between the initial residual and the input information composed of prior water supply and return information and prior environmental information, wherein the initial residual is obtained by assigning multiple regression coefficients and predicted load information to zero; the direction determination submodule is used to determine the update direction in which the multiple target active information have the same correlation with the initial residual, and update the multiple regression coefficients corresponding to the multiple target active information along the update direction; the coefficient determination submodule is used to determine the updated multiple regression coefficients as the multiple initial regression coefficients of the initial prediction model if the sum of the absolute values ​​of the updated multiple regression coefficients is greater than a preset threshold.

[0133] According to an embodiment of the present invention, the upstream supply and return water information includes the supply and return water temperature difference and flow rate information, and the upstream environmental information includes outdoor temperature information, solar radiation information and outdoor humidity information; the information determination module 410 includes: a combination submodule, used to combine the supply and return water temperature difference, flow rate information, outdoor temperature information, solar radiation information and outdoor humidity information and their respective initial regression coefficients to obtain an initial prediction model.

[0134] According to an embodiment of the present invention, the coefficient update module 420 includes: a summation submodule, which is used to sum the product of the current error and the current water supply and return information and the current environmental information with the multiple regression coefficients at the previous time as multiple updated regression coefficients, and use the multiple updated regression coefficients to update the initial prediction model to obtain the prediction model.

[0135] According to an embodiment of the present invention, the information update module 430 includes a weighting submodule and an information combination submodule. The weighting submodule is used to perform a weighted average of the previous errors at multiple previous times using preset weights to obtain the average error; the information combination submodule is used to combine the predicted load information, the average error, and the correction coefficient to obtain the updated load information.

[0136] According to an embodiment of the present invention, the above-mentioned device further includes: a coefficient update module, used to update the correction coefficient by the ratio between the current prediction error and the fusion error information, and use the updated correction coefficient as the correction coefficient at a later time to update the prediction load information output by the prediction model, wherein the fusion error information is obtained by combining the absolute value of the current load information and the current prediction error.

[0137] According to embodiments of the present invention, any plurality of modules among the information determination module 410, coefficient update module 420, and information update module 430 may be combined into one module, or any one of these modules may be split into multiple modules. Alternatively, at least a portion of the functionality of one or more of these modules may be combined with at least a portion of the functionality of other modules and implemented in one module. According to embodiments of the present invention, at least one of the information determination module 410, coefficient update module 420, and information update module 430 may be at least partially implemented as a hardware circuit, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging the circuit, or implemented in software, hardware, or firmware, or in any appropriate combination of any of these three implementation methods. Alternatively, at least one of the information determination module 410, coefficient update module 420, and information update module 430 may be at least partially implemented as a computer program module, which, when run, can perform corresponding functions.

[0138] Figure 5 A block diagram of an electronic device suitable for implementing a method for predicting load information of a building heating, ventilation, and air conditioning system according to an embodiment of the present invention is shown.

[0139] like Figure 5 As shown, an electronic device 500 according to an embodiment of the present invention includes a processor 501, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 502 or a program loaded from a storage portion 508 into a random access memory (RAM) 503. The processor 501 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 501 may also include onboard memory for caching purposes. The processor 501 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present invention.

[0140] RAM 503 stores various programs and data required for the operation of electronic device 500. Processor 501, ROM 502, and RAM 503 are interconnected via bus 504. Processor 501 executes various operations of the method flow according to embodiments of the present invention by executing programs in ROM 502 and / or RAM 503. It should be noted that the programs may also be stored in one or more memories other than ROM 502 and RAM 503. Processor 501 may also execute various operations of the method flow according to embodiments of the present invention by executing programs stored in said one or more memories.

[0141] According to an embodiment of the present invention, the electronic device 500 may further include an input / output (I / O) interface 505, which is also connected to a bus 504. The electronic device 500 may also include one or more of the following components connected to the input / output (I / O) interface 505: an input section 506 including a keyboard, mouse, etc.; an output section 507 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 508 including a hard disk, etc.; and a communication section 509 including a network interface card such as a LAN card, modem, etc. The communication section 509 performs communication processing via a network such as the Internet. A drive 510 is also connected to the input / output (I / O) interface 505 as needed. A removable medium 511, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 510 as needed so that computer programs read from it can be installed into the storage section 508 as needed.

[0142] The present invention also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of the present invention.

[0143] According to embodiments of the present invention, the computer-readable storage medium may be a non-volatile computer-readable storage medium, such as including, but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In the present invention, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of the present invention, the computer-readable storage medium may include ROM 502 and / or RAM 503 and / or one or more memories other than ROM 502 and RAM 503 described above.

[0144] Embodiments of the present invention also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to enable the computer system to implement the method for predicting building HVAC system load information provided in the embodiments of the present invention.

[0145] When the computer program is executed by the processor 501, it performs the functions defined in the system / apparatus of this invention. According to embodiments of the invention, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0146] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and may be downloaded and installed via the communication section 509, and / or installed from a removable medium 511. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0147] In such an embodiment, the computer program can be downloaded and installed from a network via communication section 509, and / or installed from removable medium 511. When the computer program is executed by processor 501, it performs the functions defined in the system of this embodiment of the invention. According to embodiments of the invention, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0148] According to embodiments of the present invention, program code for executing the computer programs provided in the embodiments of the present invention can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0149] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, can be implemented using a combination of dedicated hardware and computer instructions.

[0150] Those skilled in the art will understand that the features described in the various embodiments of the present invention can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in the present invention. In particular, the features described in the various embodiments of the present invention can be combined and / or combined in various ways without departing from the spirit and teachings of the present invention. All such combinations and / or combinations fall within the scope of the present invention.

[0151] The embodiments of the present invention have been described above. However, these embodiments are merely illustrative and not intended to limit the scope of the invention. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of the invention, and all such substitutions and modifications should fall within the scope of the invention.

Claims

1. A method for predicting load information of a building heating, ventilation, and air conditioning system, characterized in that, The method includes: Based on multiple prior operational information of the building HVAC system, an initial prediction model for predicting load information is obtained. The initial prediction model has multiple initial regression coefficients for indicating the importance of each of the multiple prior operational information. The multiple initial regression coefficients are obtained based on a loss function that includes prediction error information and constraint information. The constraint information is used to adjust the constraint strength on the multiple initial regression coefficients according to the prediction information. The multiple prior operational information includes prior water supply and return information and prior environmental information. In response to receiving current operating information, based on the current error between the current load information in the current operating information and the current predicted load information output by the initial prediction model, the plurality of initial regression coefficients are updated to obtain the prediction model; Based on the preceding errors and correction coefficients of multiple preceding moments in the preceding period, the predicted load information output by the prediction model is updated to obtain the updated load information, which serves as the load prediction result for the building's HVAC system.

2. The method according to claim 1, characterized in that, The method further includes: The average prediction error of the initial prediction model in the current time window is taken as the first error, the cumulative average prediction error of the previous running information is taken as the second error, and the relative ratio between the first error and the second error is determined. The constraint information is obtained based on the relative ratio, the number of preceding running information, and the initial constraint information obtained by cross-validation based on the preceding running information.

3. The method according to claim 1, characterized in that, The method further includes: By combining the difference between the current load information and the current predicted load information, the constraint information, and the multiple initial regression coefficients, a loss function is obtained. The preceding water supply and return information and the preceding environmental information are input into the initial prediction model. If the function value of the loss function indicates that the error of the predicted load information output by the initial prediction model is less than the error threshold, the multiple regression coefficients of the initial prediction model are determined as the multiple initial regression coefficients.

4. The method according to claim 3, characterized in that, When the function value of the loss function indicates that the error of the predicted load information output by the initial prediction model is less than an error threshold, multiple regression coefficients of the initial prediction model are determined as the multiple initial regression coefficients, including: Keeping the current values ​​of the remaining coefficients in the initial coefficient vector (excluding the target coefficient) unchanged, the target solution of the target coefficient is determined when the function value is at its valley value, wherein the initial coefficient vector is obtained by initializing the plurality of initial regression coefficients to zero; The initial values ​​of the target coefficients are updated using the target solution to obtain the updated coefficient vector; If the difference between the function value at the current time and the function value at the previous time is less than the difference threshold, the updated coefficient vector is used as the plurality of initial regression coefficients.

5. The method according to claim 3, characterized in that, When the function value of the loss function indicates that the error of the predicted load information output by the initial prediction model is less than an error threshold, multiple regression coefficients of the initial prediction model are determined as the multiple initial regression coefficients, including: Based on the correlation between the initial residual and the input information composed of the prior water supply and return information and the prior environmental information, multiple target active information is selected from the input information, wherein the initial residual is obtained by assigning the multiple regression coefficients and the predicted load information to zero; Determine update directions with the same correlation between multiple target activity information and the initial residual, and update multiple regression coefficients corresponding to the multiple target activity information along the update directions; If the sum of the absolute values ​​of the updated regression coefficients is greater than a preset threshold, the updated regression coefficients are determined as the initial regression coefficients of the initial prediction model.

6. The method according to claim 1, characterized in that, The preceding water supply and return information includes the temperature difference and flow rate of the supply and return water, and the preceding environmental information includes outdoor temperature information, solar radiation information, and outdoor humidity information; Based on multiple prior operational information from the building's HVAC system, an initial prediction model for predicting load information is obtained, including: The initial prediction model is obtained by combining the supply and return water temperature difference, the flow rate information, the outdoor temperature information, the solar radiation information, and the outdoor humidity information, along with their respective initial regression coefficients.

7. The method according to any one of claims 1 to 6, characterized in that, Based on the current error between the current load information in the current operating information and the current predicted load information output by the initial prediction model, the plurality of initial regression coefficients are updated to obtain the prediction model, including: The product of the current error and the current water supply and return information and the current environmental information is added to the sum of the regression coefficients among the multiple regression coefficients at the previous time step, and these sums are used as multiple updated regression coefficients. The initial prediction model is then updated using these multiple updated regression coefficients to obtain the prediction model.

8. The method according to claim 1, characterized in that, Based on the preceding errors and correction coefficients of multiple preceding times in the preceding period, the predicted load information output by the prediction model is updated to obtain updated load information, including: The average error is obtained by weighting the previous errors of the multiple previous time points using preset weights. The updated load information is obtained by combining the predicted load information, the average error, and the correction coefficient.

9. The method according to claim 8, characterized in that, The method further includes: The ratio between the current prediction error and the fusion error information is used to update the correction coefficient, and the updated correction coefficient is used as the correction coefficient at a later time to update the predicted load information output by the prediction model. The fusion error information is obtained by adding the absolute value of the current load information to the absolute value of the current prediction error.

10. A device for predicting load information of a building heating, ventilation, and air conditioning system, characterized in that, The device includes: An information determination module is used to obtain an initial prediction model for predicting load information based on multiple prior operational information of a building HVAC system. The initial prediction model has multiple initial regression coefficients for indicating the importance of each of the multiple prior operational information. The multiple initial regression coefficients are obtained based on a loss function that includes prediction error information and constraint information. The constraint information is used to adjust the constraint strength on the multiple initial regression coefficients according to the prediction information. The multiple prior operational information includes prior water supply and return information and prior environmental information. The coefficient update module is used to update the plurality of initial regression coefficients in response to receiving current operating information, based on the current error between the current load information in the current operating information and the current predicted load information output by the initial prediction model, to obtain the prediction model; The information update module is used to update the predicted load information output by the prediction model based on the previous errors and correction coefficients of multiple previous times in the previous period, so as to obtain the updated load information as the load prediction result of the building HVAC system.

Images