A control method, apparatus, device, and storage medium

Abstract

The present disclosure provides a control method, device, equipment and storage medium, relates to the technical field of computers, in particular to the technical field of deep learning. The specific implementation scheme is: in response to receiving a parameter issuing instruction for an energy consumption system, obtaining specified data of the energy consumption system in a current time period as target data; obtaining typical day data; wherein the typical day data is specified data required by the energy consumption system in an environment similar to the environment represented by the environment data in the target data; based on the target data and the typical day data, determining candidate running parameters of the energy consumption system in a next time period of the current time period that meet a specified condition; based on the candidate running parameters, issuing a to-be-utilized running parameter to the energy consumption system, so that the energy consumption system adjusts the current running parameter to the to-be-utilized running parameter. It can be seen that, by the present scheme, the interpretability of the control method can be improved while ensuring energy saving effect.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and more particularly to the field of deep learning technology, specifically to a control method, apparatus, device, and storage medium. Background Technology

[0002] With the deepening of the concept of energy conservation and emission reduction, energy saving and consumption reduction of energy consumption systems have become particularly important.

[0003] In related technologies, machine learning models are mainly used to model energy consumption systems, and energy-saving control parameters are found through operations research optimization algorithms to achieve control of energy consumption systems. Summary of the Invention

[0004] This disclosure provides a control method, apparatus, device, and storage medium.

[0005] According to one aspect of this disclosure, a control method is provided, comprising:

[0006] In response to receiving a parameter command for the energy consumption system, the specified data of the energy consumption system within the current time period is obtained as target data; wherein, the specified data includes at least operating parameters and environmental data;

[0007] Acquire typical daily data; wherein, the typical daily data is: the specified data required by the energy consumption system under an environment similar to that represented by the environmental data in the target data;

[0008] Based on the target data and the typical daily data, candidate operating parameters of the energy consumption system that meet specified conditions are determined for the next time period of the current time period; wherein, the specified conditions are used to ensure that the power of the energy consumption system in the next time period is less than the power in the current time period.

[0009] Based on the candidate operating parameters, the energy consumption system is issued operating parameters to be utilized, so that the energy consumption system adjusts its current operating parameters to the operating parameters to be utilized.

[0010] According to another aspect of this disclosure, a control device is provided, comprising:

[0011] The first acquisition module is used to, in response to receiving a parameter distribution instruction for the energy consumption system, acquire specified data of the energy consumption system within the current time period as target data; wherein, the specified data includes at least operating parameters and environmental data;

[0012] The second acquisition module is used to acquire typical daily data; wherein, the typical daily data is: the specified data required by the energy consumption system under an environment similar to the environment represented by the environmental data in the target data;

[0013] The determination module is used to determine, based on the target data and the typical daily data, candidate operating parameters of the energy consumption system that meet specified conditions in the next time period of the current time period; wherein, the specified conditions are used to ensure that the power of the energy consumption system in the next time period is less than the power in the current time period.

[0014] The adjustment module is used to issue the operating parameters to be utilized to the energy consumption system based on the candidate operating parameters, so that the energy consumption system adjusts the current operating parameters to the operating parameters to be utilized.

[0015] According to another aspect of this disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform any of the control methods described above.

[0016] According to another aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are used to cause the computer to perform the control method according to any of the preceding claims.

[0017] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the control method according to any of the preceding claims.

[0018] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0019] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0020] Figure 1 This is a flowchart of a control method according to the present disclosure;

[0021] Figure 2 This is a flowchart of step S101 of this disclosure;

[0022] Figure 3 This is a flowchart of step S103 of this disclosure;

[0023] Figure 4 This is a schematic diagram of a message queue-based cloud-edge data link according to this disclosure;

[0024] Figure 5 This is a schematic diagram of the working principle of a control system based on typical day and machine learning according to the present disclosure;

[0025] Figure 6A This is a structural schematic diagram of a building refrigeration system according to the present disclosure;

[0026] Figure 6B It is a heatmap of a Pearson correlation coefficient based on this disclosure;

[0027] Figure 6C This is a schematic diagram illustrating a predicted effect according to this disclosure;

[0028] Figure 7 This is a schematic diagram of the structure of a control device according to the present disclosure;

[0029] Figure 8 This is a block diagram of an electronic device used to implement the control method of the embodiments of this disclosure. Detailed Implementation

[0030] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0031] With the continuous advancement of the goals of "carbon neutrality" and "carbon peaking," advanced technologies such as the Internet of Things (IoT), big data, and artificial intelligence are increasingly empowering traditional industries like real estate, accelerating their low-carbon transformation. The continuous development of IoT technology provides the prerequisites for energy consumption data collection, monitoring, and control in building refrigeration and boiler systems: in edge-cloud integrated IoT solutions, edge sensors can collect real-time field data and upload it to cloud databases. Simultaneously, cloud-based control devices, based on edge data and through cloud-based big data modeling, can predict future energy consumption needs and adjust edge control devices in advance, or schedule and configure relevant energy resources, such as the number of chillers required for refrigeration equipment.

[0032] In related technologies, machine learning models are mainly used to model building HVAC systems. Operations research algorithms are then used to iterate and optimize energy-saving control parameters to achieve control of refrigeration equipment. However, because machine learning models are data-based black-box models, they suffer from poor interpretability and difficulty in learning from boundary data.

[0033] Based on the above, in order to improve the interpretability of the control method while ensuring energy-saving effect, this disclosure provides a control method, apparatus, device, and storage medium.

[0034] The following section will first introduce a control method provided by an embodiment of this disclosure.

[0035] The control method provided in this disclosure can be applied to a control device that can communicate with an energy consumption system. In practical applications, the control device can be a server deployed in the cloud. The control device can receive data collected from sensors regarding the energy consumption system, process the collected data, and send operating parameters to the energy consumption system, etc.

[0036] Specifically, the entity executing this control method can be a control device. For example, when the control method is applied to a control device, the control device can be a computer program running on a server, which can be used to issue operating parameters to the energy consumption system.

[0037] One of the control methods provided in this disclosure may include the following steps:

[0038] In response to receiving a parameter command for the energy consumption system, the specified data of the energy consumption system within the current time period is obtained as target data; wherein, the specified data includes at least operating parameters and environmental data;

[0039] Acquire typical daily data; wherein, the typical daily data is: the specified data required by the energy consumption system under an environment similar to that represented by the environmental data in the target data;

[0040] Based on the target data and the typical daily data, candidate operating parameters of the energy consumption system that meet specified conditions are determined for the next time period of the current time period; wherein, the specified conditions are used to ensure that the power of the energy consumption system in the next time period is less than the power in the current time period.

[0041] Based on the candidate operating parameters, the energy consumption system is issued operating parameters to be utilized, so that the energy consumption system adjusts its current operating parameters to the operating parameters to be utilized.

[0042] In the scheme provided in this disclosure, since the candidate operating parameters are those that ensure the power of the energy-consuming system in the next time period is less than the power in the current time period, issuing the operating parameters to be utilized to the energy-consuming system allows it to adjust its current operating parameters to those parameters, thereby reducing energy consumption. Furthermore, the candidate operating parameters that meet the specified conditions are determined based on the target data and typical daily data; that is, the determination of the candidate operating parameters incorporates typical daily data. Typical daily data represents reasonable operating parameters adjusted to fit an environment similar to that represented by the environmental data in the target data. Therefore, by combining typical daily data to predict the operating parameters for the next time period, the interpretability of the model can be improved. Thus, this scheme can improve the interpretability of the control method while ensuring energy-saving effects.

[0043] The control method provided by the embodiments of this disclosure will now be described in conjunction with the accompanying drawings.

[0044] like Figure 1 As shown, the control method provided in this embodiment of the present disclosure, applied to a control device, may include steps S101-S104:

[0045] S101, in response to receiving a parameter distribution instruction for the energy consumption system, obtain specified data of the energy consumption system within the current time period as target data; wherein, the specified data includes at least operating parameters and environmental data;

[0046] In this embodiment, the control device can receive parameter issuance commands for the energy consumption system triggered by the user on the client through the communication interface, or it can periodically trigger parameter issuance commands for the energy consumption system, both of which are reasonable. For example, when periodically triggering parameter issuance commands for the energy consumption system, the triggering period can be 30 minutes, 1 hour, etc. The current time period can be the time period between the current moment and the previous command triggering moment. For example, if the triggering period is 30 minutes, after receiving a parameter issuance command for the energy consumption system, specified data within the range of the previous 30 minutes can be obtained as the target data.

[0047] For example, in practical applications, this energy-consuming system can be a building refrigeration system, or a boiler system, etc. For example, if the energy-consuming system is a building refrigeration system, which includes multiple devices such as cooling towers, cooling pumps, chillers, and refrigeration pumps, then the parameter commands issued to the building refrigeration system are parameter commands used to control the operating parameters of these multiple devices. For example, these operating parameters may include operating status, rated power, etc. The environmental data may include indoor temperature, indoor humidity, and outdoor temperature, etc.

[0048] Understandably, when it's necessary to predict the operating parameters of an energy consumption system, these parameters are influenced by environmental factors—that is, the operating parameters differ under different environments. Therefore, we can use the operating parameters of the energy consumption system within the current time period, along with environmental data, to predict its operating parameters for future moments. When the control device receives a parameter command, it can first acquire the specified data for the energy consumption system within the current time period. This specified data can then be used to predict the operating parameters of the energy consumption system, specifically the operating parameters of each device included in the system.

[0049] S102, Obtain typical daily data; wherein, the typical daily data is: the specified data required by the energy consumption system under an environment similar to the environment represented by the environmental data in the target data;

[0050] In this embodiment, the typical daily data refers to the specified data required by the energy consumption system under an environment similar to that represented by the environmental data in the target data. That is, the environment represented by the environmental data in the typical daily data is similar to the environment represented by the environmental data within the current time period. It can be understood that since the typical daily data is the specified data required by the energy consumption system under an environment similar to that represented by the environmental data in the target data, the operating parameters in the typical daily data are reasonable operating parameters adjusted for an environment similar to that within the current time period. Therefore, the typical daily data can be used to predict operating parameters, thereby learning the operating parameters required for energy consumption under similar conditions.

[0051] Optionally, in one implementation, obtaining typical daily data may include steps A1-A2:

[0052] A1, calculate the similarity between the environmental data in the target data and the environmental data in the historical specified data; wherein, the historical specified data is the specified data of the energy consumption system within the historical period;

[0053] In this implementation, the historical time period refers to the time period preceding the current time period, and this historical time period can have the same duration as the current time period. For example, if a parameter is issued at time T with a trigger period of 30 minutes, then the current time period is [T-30, T], and the historical time period can include [T-60, T-30], [T-90, T-60], and so on.

[0054] For example, the similarity between environmental data in the target data and environmental data in the specified historical data can be calculated as follows: construct the environmental data in the target data and the environmental data in the specified historical data into vectors respectively, and then calculate the Euclidean distance, or cosine distance, etc. between the environmental data in the target data and the environmental data in the specified historical data. The calculated distance value is the similarity.

[0055] Understandably, since Euclidean distance can better reflect the differences in values ​​between data points, the Euclidean distance between the environmental data in the target dataset and the environmental data in the specified historical dataset can be calculated using the following formula:

[0056]

[0057] Among them, similarity i,j This represents the similarity between vector i constructed from environmental data in the target data and vector j constructed from environmental data in the specified historical data; n represents the dimension of vector i, and x represents the similarity between vector i and vector j. i,k Let x represent the k-th dimension of vector i. j,k Let represent the k-th dimension of vector j.

[0058] A2, based on the calculated similarity, select typical daily data from the specified historical data.

[0059] Understandably, after calculating the similarity, designated data with high similarity can be selected from various historical data sets as typical daily data, thereby giving the operating parameters in the selected typical daily data sets better learning value. For example, the calculated similarities can be sorted in descending order, and a predetermined number of similarities from the top of the descending sequence can be selected. The historical data corresponding to the selected similarities can then be determined as the predetermined number of typical daily data sets.

[0060] S103, Based on the target data and the typical daily data, determine the candidate operating parameters of the energy consumption system that meet the specified conditions in the next time period of the current time period; wherein, the specified conditions are used to make the power of the energy consumption system in the next time period less than the power in the current time period.

[0061] In this embodiment, the typical daily data can be one or more. For example, based on the target data and the typical daily data, the operating parameters of the energy consumption system in the next time period can be predicted, and then candidate operating parameters that meet specified conditions can be determined from the predicted operating parameters. It is understood that since the specified conditions are used to ensure that the power of the energy consumption system in the next time period is less than the power in the current time period, determining the candidate operating parameters that meet the specified conditions ensures that when any candidate operating parameters are subsequently issued to the energy consumption system, the system will operate according to the issued operating parameters to achieve energy-saving effects.

[0062] It should be noted that if no candidate operating parameters that meet the specified conditions are found among the predicted operating parameters, the current operating parameters of the energy consumption system will be determined as candidate operating parameters. Furthermore, for clarity of the scheme, the implementation method for predicting the operating parameters of the energy consumption system in the next time period after the current time period, and for determining the candidate operating parameters that meet the specified conditions from the predicted operating parameters, will be described below; it will not be repeated here.

[0063] S104, based on the candidate operating parameters, issue the operating parameters to be utilized to the energy consumption system so that the energy consumption system adjusts its current operating parameters to the operating parameters to be utilized.

[0064] In this embodiment, if there is only one candidate operating parameter, it can be sent to the energy consumption system as the operating parameter to be used. If there are multiple candidate operating parameters, one parameter can be selected from the candidate operating parameters as the operating parameter to be used and sent to the energy consumption system. For example, the selection method can be based on preset boundary conditions, such as selecting candidate operating parameters that indicate that at least two chillers are turned on; or, the selection method can be sorted according to user needs, or random selection, etc.

[0065] It is understandable that, since the operating parameter to be utilized is one of the candidate operating parameters, and the candidate operating parameter is the operating parameter that makes the power of the energy consumption system in the next time period less than the power in the current time period, issuing the operating parameter to be utilized to the energy consumption system so that the energy consumption system can adjust its current operating parameter to the operating parameter to be utilized can reduce the energy consumption of the energy consumption system.

[0066] Optionally, in one implementation, the number of candidate runtime parameters is multiple;

[0067] Accordingly, in this implementation, issuing the operational parameters to be utilized to the energy consumption system based on the candidate operational parameters may include steps B1-B2:

[0068] B1. Sort the candidate operating parameters according to the preset sorting conditions; wherein, the sorting conditions include sorting index and sorting method, and the sorting method is such that when sorting according to the sorting index, the first digit of the resulting sequence is the optimal operating parameter.

[0069] B2 selects the first operating parameter from the sorted sequence and sends it to the energy consumption system as the operating parameter to be used.

[0070] In this implementation, the ranking metric can be the pattern similarity of the energy consumption system, the predicted power value, etc. It is understood that different ranking methods can be adopted for different ranking metrics, so that the first element of the resulting sequence is the optimal operating parameter. For example, if the ranking metric is the predicted power value, then when the candidate operating parameters are sorted in ascending order according to their corresponding predicted power values, the first element of the sequence is the operating parameter that minimizes power, i.e., the optimal operating parameter under this ranking metric. For example, if the ranking metric is the pattern similarity of the energy consumption system, i.e., the similarity between the system mode included in the operating parameter and the current mode of the system, then when the candidate operating parameters are sorted in descending order according to their corresponding pattern similarity, the first element of the sequence is the operating parameter that maximizes pattern similarity, i.e., the optimal operating parameter under this ranking metric.

[0071] It is understandable that the preset sorting conditions can be set according to different user needs, so that the operating parameters selected according to the preset sorting conditions can not only save energy, but also meet different user needs. For example, the selected operating parameters have the lowest energy consumption in the next time period, or the system mode contained in the selected operating parameters is closest to the current mode.

[0072] In the scheme provided in this disclosure, since the candidate operating parameters are those that ensure the power of the energy-consuming system in the next time period is less than the power in the current time period, issuing the operating parameters to be utilized to the energy-consuming system allows it to adjust its current operating parameters to those parameters, thereby reducing energy consumption. Furthermore, the candidate operating parameters that meet the specified conditions are determined based on the target data and typical daily data; that is, the determination of the candidate operating parameters incorporates typical daily data. Typical daily data represents reasonable operating parameters adjusted to fit an environment similar to that represented by the environmental data in the target data. Therefore, by combining typical daily data to predict the operating parameters for the next time period, the interpretability of the model can be improved. Thus, this scheme can improve the interpretability of the control method while ensuring energy-saving effects.

[0073] Alternatively, in another embodiment of this disclosure, such as Figure 2 As shown, obtaining specified data of the energy consumption system within the current time period as target data in step S101 above may include steps S1011-S1012:

[0074] S1011, according to the preset sampling interval, acquire multiple initial specified data of the energy consumption system within the current time period;

[0075] For example, the preset sampling interval could be 2 minutes, 5 minutes, etc. Each device in the energy consumption system can report operating parameters to the control device according to the preset sampling interval, and sensors used to collect environmental data can report the collected environmental data to the control device according to the preset sampling interval. Thus, multiple initial specified data points for the energy consumption system within the current time period can be obtained according to the preset sampling interval.

[0076] S1012, Perform specified preprocessing on the multiple initial specified data to obtain specified data of the energy consumption system within the current time period, which is used as target data; wherein, the specified preprocessing includes data aggregation processing on the multiple specified data.

[0077] Understandably, after acquiring multiple initial specified data points according to a preset sampling interval, to reduce the complexity of subsequent data processing, these initial specified data points can be aggregated to obtain a specified data point corresponding to the current time period. For example, this data aggregation process could involve taking the average or extreme values ​​of the multiple initial specified data points. For instance, for parameters such as frequency and flow rate in the specified data, the average value can be taken; for discrete data such as equipment status and valve on / off states in the specified data, the value representing the highest corresponding energy consumption can be taken.

[0078] Additionally, it is understandable that for data with multiple backup measurement points, such as indoor temperature data, since multiple indoor temperatures are collected each time, there may be erroneous data among these multiple indoor temperatures. Therefore, before aggregating the indoor temperatures, for each sampling time, the percentage error between the multiple indoor temperatures collected at that sampling time can be calculated first, so as to remove indoor temperatures with large errors and then take the average value, thereby obtaining more accurate temperature data.

[0079] Optionally, in one implementation, before performing specified preprocessing on the multiple initial specified data to obtain specified data of the energy consumption system within the current time period, and using this as the target data, steps C1-C2 may be included:

[0080] C1, determine the device mode sequence corresponding to multiple initial specified data; wherein, the device mode sequence corresponding to the specified data is the device sequence of the specified device represented by the specified data that is in the on state;

[0081] In this implementation, the device mode sequence corresponding to the initial specified data can be determined by using the data representing the operating status of the specified device in the initial specified data. For example, if the energy consumption system is a building cooling system, then the specified device is a chiller, and the device mode sequence corresponding to the initial specified data is the sequence of devices in the specified data that represent the chiller's operating status as "on". For instance, if chillers 1, 2, and 3 are on in the building cooling system, then the device mode sequence is "123".

[0082] C2, remove specified data corresponding to a specified pattern sequence from the multiple initial specified data; wherein, the specified pattern sequence is the sequence other than the device pattern sequence with the longest runtime among the determined device pattern sequences.

[0083] In this implementation, by removing specified data corresponding to sequences other than the device mode sequence with the longest runtime, the initial specified data corresponding to the device mode sequence with the longest runtime can be retained. It can be understood that in practical applications, when acquiring initial specified data at a preset sampling interval, the device mode sequence with the longest runtime is the specified data whose corresponding device mode sequence has the highest frequency among the multiple initial specified data.

[0084] It is understandable that if the energy consumption system has undergone a mode switch during the current time period, the transitional mode during the switch will affect the initial specified data, making subsequent predictions based on that initial specified data inaccurate. Therefore, to reduce the impact of mode switching on the initial specified data, the specified data corresponding to sequences other than the device mode sequence with the longest runtime can be removed from the multiple initial specified data sets.

[0085] As can be seen, this solution can reduce the complexity of data processing.

[0086] Alternatively, in another embodiment of this disclosure, the number of typical daily data is multiple;

[0087] like Figure 3 As shown, in step S103 above, based on the target data and the typical daily data, determining the candidate operating parameters of the energy consumption system that meet the specified conditions in the next time period of the current time period may include steps S301-S302:

[0088] S301, Based on the target data and the multiple typical daily data, predict the target operating parameters of the energy consumption system in the next time period of the current time period; wherein, the number of the target operating parameters is multiple;

[0089] In this embodiment, to improve the accuracy of parameter prediction, predictions can be made based on target data and multiple typical daily data. For example, in practical applications, target data and multiple typical daily data can be used as input data, and a pre-trained machine learning model, or a mechanistic model of the energy consumption system, combined with the pre-trained machine learning model, can be used to predict the target operating parameters of the energy consumption system for the next time period within the current time period.

[0090] Optionally, in one implementation, predicting the target operating parameters of the energy system for the next time period within the current time period, based on the target data and the multiple typical daily data, may include steps D1-D2:

[0091] D1, determine multiple input data corresponding to the specified inference model; wherein, the input data includes the target data and a typical daily data obtained; wherein, the specified inference model is used to predict the operating parameters of the energy consumption system in the next time period;

[0092] D2, based on the determined input data, uses the specified inference model to determine the target operating parameters of the energy consumption system for the next time period.

[0093] In this implementation, the designated inference model is a model used to predict the operating parameters of the energy consumption system in the next time period. For example, the designated inference model can be a pre-trained machine learning model, or it can include a mechanistic model of the energy consumption system along with the pre-trained machine learning model; both are reasonable. It is understood that when making predictions using target data and multiple typical daily data, each typical daily data point and the target data constitute one input data point, thereby determining multiple input data points. Inputting these determined multiple input data points into the designated inference model yields multiple output results from the designated inference model, namely, multiple target operating parameters of the energy consumption system in the next time period.

[0094] For example, in one specific implementation, the operating parameters of the energy consumption system include a first type of parameters and a second type of parameters; the specified inference model includes: a mechanism model and a pre-trained target machine learning model; wherein, the mechanism model is a model for predicting the first type of parameters, and the target machine learning model is a model for predicting the second type of parameters; wherein, the first type of parameters are parameters that conform to a linear law, and the second type of parameters are parameters other than the first type of parameters;

[0095] In this implementation, the specified inference model includes a mechanistic model and a pre-trained target machine learning model. It is understood that the operating parameters of the energy consumption system include a first type of parameter that follows a linear law, and a second type of parameter other than the first type. For example, if the rated flow rate parameter of a chilled pump follows a linear law with the chilled pump frequency and the total flow rate of the chilled pump, then the rated flow rate parameter of the chilled pump is a first type of parameter. Since the first type of parameter follows a linear law, a mechanistic model for this first type of parameter can be determined using historical data; this mechanistic model is a linear model used to predict the first type of parameter.

[0096] Furthermore, since the second type of parameter is a parameter other than the first type of parameter, it does not conform to a certain linear law. Therefore, the prediction of the second type of parameter can be carried out by using a machine learning model to explore the various influencing factors related to the second type of parameter, thereby determining the target machine learning model for predicting the second type of parameter.

[0097] Accordingly, in this implementation, step D2, based on the determined input data and using the specified inference model, determines the target operating parameters of the energy consumption system for the next time period, which may include:

[0098] Given the determined input data, based on the first data in the input data, the mechanism model is used to predict the first type of parameters of the energy consumption system in the next time period, and based on the second data in the input data, the target machine learning model is used to predict the second type of parameters of the energy consumption system in the next time period, thereby obtaining the target operating parameters of the energy consumption system in the next time period.

[0099] The first data is the pre-set input parameter data belonging to the mechanism model from the target data and / or typical daily data included in the input data; the second data is the pre-set input parameter data belonging to the target machine learning model from the target data and / or typical daily data included in the input data.

[0100] It is understandable that when making predictions using a specified inference model, since the specified inference model includes a mechanistic model and a target machine learning model, after the input data is input into the specified inference model, the mechanistic model in the specified inference model can make predictions using the first data of the input parameters of the mechanistic model that are pre-set in the input data, and the target machine learning model in the specified inference model can make predictions using the second data of the input parameters of the target machine learning model that are pre-set in the input data.

[0101] It is understandable that, since when making a prediction for a certain first-type parameter or second-type parameter, it may not only use the content of the input data, but also the content of other first-type data or second-type data that have already been predicted, the first and second data can be data from the target data and / or typical daily data.

[0102] Furthermore, since there are first-class parameters among the parameters to be predicted that can be predicted using the mechanistic model, and the mechanistic model is a simple linear model, combining the mechanistic model and the target machine learning model for prediction can reduce the consumption of computational resources in the prediction process and further improve the interpretability of the specified inference model.

[0103] S302, From the predicted target operating parameters, select the target operating parameters that meet the specified conditions as candidate operating parameters.

[0104] In this embodiment, since there are multiple typical daily data points and multiple target operating parameters to be predicted, target operating parameters that meet the specified conditions can be selected from these multiple target operating parameters as candidate operating parameters. It is understood that when using a single typical daily data point and target data for prediction, the prediction accuracy is greatly affected by the reliability of the data. Therefore, using multiple typical daily data points and target data to predict the target operating parameters of the energy consumption system in the next time period can solve the problem of inaccurate prediction results caused by anomalies in a single typical daily data point, thereby improving the accuracy of the prediction.

[0105] It is evident that this approach can improve the accuracy of parameter prediction.

[0106] Optionally, in another embodiment of this disclosure, the target machine learning model is a machine learning model that meets the preset model accuracy requirements after training multiple machine learning models using sample data, and the input parameters of different machine learning models are of different categories.

[0107] In this embodiment, by training multiple machine learning models with different input parameters using sample data, multiple machine learning models with different prediction accuracies can be obtained. Therefore, a machine learning model that meets the preset model accuracy requirement can be selected from these multiple models as the target machine learning model. It is understood that since the target machine learning model is the selected model that meets the preset model accuracy requirement, using the target machine learning model to predict the second type of parameter can ensure the accuracy of the prediction results.

[0108] For example, in practical applications, machine learning models with different input parameters for different categories can be set up, meaning that the input features of these multiple machine learning models are different. After all these machine learning models have been trained, they can be evaluated using evaluation metrics to select the model with high prediction accuracy. Then, in subsequent predictions, this high-accuracy model can be used to predict the second type of parameter to be predicted. For example, the evaluation metrics could be mean squared error, mean absolute error, or r... 2 Machine learning regression metrics such as (coefficient of determination).

[0109] Additionally, it should be noted that this machine learning model can be a neural network model, an XGBoost model, or something similar. Since the XGBoost model excels at capturing dependencies between complex data, it can be chosen as the machine learning model for this purpose.

[0110] Optionally, in one implementation, the method for determining the category of the input parameters corresponding to the machine learning model may include steps E1-E2:

[0111] E1 calculates the correlation coefficients between each data category of the specified historical data and the output parameters of the machine learning model; where the specified historical data refers to the specified data of the energy consumption system within the historical period.

[0112] E2: Select the data category whose corresponding correlation coefficient meets the preset threshold from each data category, and use it as the category of the input parameter corresponding to the machine learning model.

[0113] In this implementation, the data category can be categories such as operating status or rated power, corresponding to different operating parameters. For example, the Pearson correlation coefficient formula can be used to calculate the correlation coefficient between each data category of the specified historical data and the output parameters of the machine learning model. It can be understood that by calculating the correlation coefficient between each data category of the specified historical data and the output parameters of the machine learning model, and then selecting data categories whose correlation coefficients meet a preset threshold, the data categories with a high correlation to the output parameters of the machine learning model can be determined as the input parameter categories.

[0114] It is understandable that, since the categories of input parameters and output parameters of a machine learning model are highly correlated, the data corresponding to the categories of the selected input parameters in the historical specified data can be used as input data to train the machine learning model, thereby obtaining a machine learning model with a certain level of accuracy.

[0115] It should be noted that, in addition to determining the categories of the input parameters corresponding to the machine learning model through steps E1-E2 described above, the categories of the input parameters can also be set based on the experience of relevant technical personnel. Furthermore, if the machine learning model is an XGBoost model, since the XGBoost model includes a built-in function for calculating the importance of input features, it is reasonable to determine the categories of the input parameters not only through steps E1-E2 or the experience of relevant technical personnel, but also by using this built-in function.

[0116] Optionally, in one implementation, the training method of the target machine learning model may include steps F1-F4:

[0117] F1, acquire multiple sample data; wherein, the sample data includes sample target data and a sample typical day data, the sample target data is the specified data of the energy consumption system within the sample time period, the sample typical day data is the specified data required by the energy consumption system under an environment similar to the environment represented by the environmental data in the sample target data, and the sample data has a specified label, the specified label representing the true value of the operating parameters of the energy consumption system in the next time period of the sample time period;

[0118] In this implementation, the method for obtaining the target data and typical daily data of the sample can be the same as the method for obtaining the target data and typical daily data in the above steps, and will not be repeated here.

[0119] F2, based on the sample data and the initial target machine learning model, predicts the operating parameters of the energy system in the next time period of the sample time period, and obtains the prediction results;

[0120] In this implementation, sample data can be input into the initial target machine learning model to obtain the model output, which is then used as the prediction result.

[0121] F3, based on the specified label and the prediction result, determines the model loss value of the target machine learning model;

[0122] In this implementation, since the specified label represents the true value of the operating parameters of the energy consumption system in the next time period after the sample time period, the model loss value of the target machine learning model can be determined by calculating the difference between the specified label and the prediction result.

[0123] F4, based on the model loss value, adjusts the model parameters of the target machine learning model and returns to the step of obtaining multiple sample data until the target machine learning model converges.

[0124] In this implementation, after determining the model loss value, the parameters of the target machine learning model are adjusted using this loss value until the target machine learning model converges, training ends, and the trained target machine learning model is obtained. For example, in practical applications, the model parameters of the target machine learning model can be adjusted through backpropagation by minimizing the model loss value.

[0125] It is evident that this approach can further improve the accuracy of parameter prediction.

[0126] To better understand this solution, the method provided in this disclosure embodiment will be illustrated below with a specific example.

[0127] Currently, most building cooling systems can be manually controlled based on mechanistic parameters (such as primary side pressure difference and primary side temperature difference). However, manual control is expensive and cannot effectively utilize the data resources accumulated by enterprises during production, resulting in a lack of universality in the modeling methods. Physical mechanism-based building energy consumption system simulation software, such as EnergyPlus and DeST, requires extensive expert experience for simulation modeling, leading to high application costs. Data-driven modeling methods, while requiring no expert experience and possessing some universality, rely heavily on data quality—specifically, a continuous and stable data flow. Abnormal data and missing data significantly impact the model's prediction accuracy and generalization performance. More importantly, these methods are extremely sensitive to data distribution. When the actual distribution of the data to be predicted changes due to variations in equipment and piping characteristics or uncontrollable factors such as equipment aging, prediction accuracy drops drastically. Furthermore, even with a reliable equipment and piping model, traversing and optimizing using operations research methods can result in the "curse of dimensionality" problem in the solution space due to the large number of equipment and piping components, consuming significant computational resources and failing to guarantee real-time result delivery. In high-dimensional solution spaces, reinforcement learning models require more computational resources due to the large amount of training data they need. Furthermore, reinforcement learning models may exhibit exploratory behavior, posing a significant risk to production and operational safety.

[0128] Based on the above, this example discloses a control scheme for a building cooling system, which mainly includes three parts: a cloud-edge data link based on message queues; a control system based on typical daily data and machine learning; and a prediction model based on XGBoost.

[0129] I. Cloud-Edge Data Link Based on Message Queues

[0130] In enterprise production processes, each production line uses numerous sensors to collect data in a time-sharing manner. If each sensor reports data to the control system (corresponding to the control equipment mentioned above), a denial-of-service (DDoS) problem can occur when the control system lacks high performance and receives a large influx of data. This poses a significant challenge to monitoring and early warning systems. Therefore, gateway devices, such as industrial control computers or edge gateways, can be used to manage sensor metadata and process reported data. In the data link between the gateway device and the control system, if communication is via HTTP (Hypertext Transfer Protocol) or TCP (Transmission Control Protocol), data loss may occur if the communication link fails to be established. Therefore, middleware message queues, such as RabbitMQ (a message-oriented middleware), can be used to ensure message reliability.

[0131] like Figure 4 As shown, the gateway device uses a high-performance non-blocking communication framework, such as Netty, to receive data reported by sensors via HTTP. For the gateway device's control of the sensors, TCP (Transmission Control Protocol) Socket communication is used. This is a real-time, compact communication method that ensures the sensors receive commands and execute the operations indicated by those commands. The message queue uses a pub / sub (publish / subscribe) delivery model. When the gateway device reports data, it acts as the sender, packaging the data into a fixed format and pushing it to the message queue. The control system acts as the consumer, consuming data from the message queue and writing it to the database. When the control system issues control commands, the gateway device acts as the consumer, retrieving the control commands from the message queue, while the control system acts as the sender, sending the control commands to the message queue.

[0132] Understandably, by employing a message queue-based cloud-edge data link, compared to real-time data push, this link can choose to cache and push data in batches to address the high-performance requirements of the control system. Batch pushing shifts the data load to the gateway device, ensuring the real-time performance and availability of the control system. Furthermore, by partitioning resources through message queues, the mapping relationship between gateway devices and message queues can be planned, allowing for rapid problem localization.

[0133] II. Control System Based on Typical Daily Data and Machine Learning

[0134] The working principle of a control system based on typical day and machine learning is as follows: Figure 5As shown. The strategy calculation task (corresponding to the control method described above) is triggered at a certain frequency f using conditional triggering or timed triggering methods, receiving trigger conditions through a RESTful API (an application programming interface that accesses or uses data via HTTP requests). The control system can concurrently calculate the control parameters (corresponding to the operating parameters described above) of each building's cooling system and distribute them from the cloud where the control system resides. The specific calculation process for each strategy calculation task is as follows:

[0135] 1. Real-time data processing

[0136] (1) Obtain real-time data from the device (corresponding to the initial specified data within the current time period mentioned above): Based on the timestamp of the triggered task, advance one trigger cycle and read the real-time data of the cooling system for the time period corresponding to the previous trigger cycle (corresponding to the current time period mentioned above) from the database. If T is the trigger time and f is the policy issuance frequency, such as 30 minutes / time, then the time filtering range of the data is T–30 to T. Figure 6A As shown, the building's refrigeration system includes equipment such as cooling towers, cooling pumps, chillers, and refrigeration pumps. The real-time data includes cooling-side (cooling tower, cooling pump) data, chiller data, refrigeration-side (refrigeration pump) data, and secondary-side (outdoor temperature, indoor temperature, humidity) data, etc.

[0137] (2) Data preprocessing for real-time data: To avoid the impact of mode switching on real-time data, the real-time data is first sorted according to the runtime of different equipment mode sequences within the time period corresponding to the previous trigger cycle. The equipment mode sequence in the building refrigeration system is defined as the sequence of chiller activation. For example, if chillers 1, 2, and 3 are activated, the mode is "123". Real-time data under the longest equipment mode sequence is retained, and abnormal data under other modes are removed. Then, through data aggregation, the maximum value is taken for discrete variables such as equipment status and valve opening / closing, and the average value is taken for equipment parameters such as frequency and flow rate. The real-time data is then resampled and smoothed. In addition, for data with multiple backup measuring points, such as indoor temperature, the percentage error of the indoor temperature at these multiple backup measuring points is calculated. Measuring points with a percentage error greater than 5% are removed, and then the average value is taken. The percentage error calculation formula is as follows:

[0138]

[0139] Among them, APE i x represents the percentage error at the i-th measurement point. mean x represents the average indoor temperature at multiple measuring points. i Let be the indoor temperature at the i-th measuring point.

[0140] 2. Calculate similar days

[0141] (1) Obtaining Historical Specified Data: Based on the timestamp of the triggered task, advance several trigger cycles and read the environmental parameters (corresponding to the environmental data in the above text) and equipment energy consumption or power data within the time period corresponding to the previous several trigger cycles (corresponding to the historical time period in the above text) from the time series database as the initial historical specified data. The environmental parameters may include outdoor temperature, wet-bulb temperature, etc. Specified data with outliers are removed from the obtained initial historical specified data. Then, data aggregation processing is performed on the historical specified data to calculate the average environmental parameters and historical energy consumption of multiple initial historical specified data, which are then used as the historical specified data.

[0142] (2) Calculate similarity

[0143] The similarity between environmental parameters in historical and real-time data is obtained by calculating the Euclidean distance between the two sets of data. Euclidean distance better reflects the numerical differences between the data, with a similarity range of (0,1]. The calculation formula is as follows:

[0144]

[0145] Among them, similarity i,j This represents the similarity between vector i constructed from environmental parameters in real-time data and vector j constructed from environmental parameters in specified historical data; n represents the dimension of vector i, and x represents the similarity between vector i and vector j. i,k Let x represent the k-th dimension of vector i. j,k Let $k$ represent the k-th dimension of vector $j$. By removing historical data with similarity less than a certain threshold and sorting them in descending order of similarity, the $K$ most similar data points from the remaining historical data are selected as typical daily data.

[0146] 3. Building Refrigeration System Model Reasoning

[0147] (1) Cooling load forecasting: Based on real-time data, the hourly cooling capacity required by the building's cooling system is predicted. The formula for calculating the cooling load Q is as follows:

[0148] Q = 1.163 × ΔT × G

[0149] Where G is the chilled water main flow rate, ΔT is the temperature difference between the return water and the supply water in the chilled water main, i.e., the difference between the outlet water temperature and the inlet water temperature on the chilled side, and 1.163 is the specific heat capacity of water.

[0150] (2) Total flow rate on the chilled side: The frequency of a single variable frequency pump is directly proportional to the flow rate. The mechanism model for predicting the total flow rate of the chilled pipe based on the chilled pump frequency parameters in typical daily data is as follows:

[0151]

[0152] Where f is the frequency of the variable frequency water pump, and G e This is the rated flow rate of the variable frequency water pump. Model learning parameter G e This allows you to obtain the flow rate model for a single water pump. The model can be used to model a single water pump, or it can be based on a mechanistic model to model the total flow rate of multiple water pumps.

[0153] (3) Chilled water supply temperature: Since the chilled water supply temperature can be obtained by adjusting the chiller outlet water temperature, the chilled water supply temperature in the typical daily data is used.

[0154] (4) Chilled Water Return Temperature: The chilled water return temperature is predicted based on real-time environmental data, chilled water supply temperature, predicted cooling load, and predicted total chilled water flow. The chilled water temperature difference can be calculated using the predicted chilled water return and supply temperatures. The control objectives of the strategy can be consistent with expert experience. For example, if the chilled water temperature difference is greater than a certain threshold, the strategy will be pruned if it is lower than the threshold, and no further calculations will be performed.

[0155] (5) Evaporator inlet water temperature: Based on real-time environmental parameters, predicted chilled side flow rate, predicted cooling load value, chilled side supply water temperature, predicted return water temperature, chiller start-up and shutdown parameters, and based on the evaporator outlet water temperature (i.e. chiller outlet water temperature) in typical daily data, the evaporator inlet water temperature of each chiller that is turned on is predicted, which is also the chiller chilled side inlet water temperature.

[0156] (6) Total flow rate on the cooling side: The principle is the same as that on the freezing side. The total flow rate of the cooling pipe can be predicted based on the cooling pump control frequency parameters in the typical daily data.

[0157] (7) Cooling side inlet and outlet water temperatures: The cooling side inlet and outlet water temperatures can be predicted based on the environmental parameters in the real-time data, the predicted total flow rate of the cooling side, the predicted parameters of the refrigeration side, and the parameters of the refrigeration side when the chiller is turned on.

[0158] (8) Power of chilled water pump and cooling pump: The power of a single variable frequency water pump is cubically related to the frequency. Therefore, based on the mechanism formula and historical data, a power model for chilled water pump and cooling pump can be established. The mechanism model is as follows:

[0159]

[0160] Where p is the power of a single variable frequency water pump, f is the frequency of the variable frequency water pump, and p e This is the rated power of the variable frequency water pump. Mechanism model learning parameter p eThis allows you to obtain the power model of a single water pump. You can model a single water pump, or you can model the total power of multiple water pumps based on a mechanistic model.

[0161] (9) Chiller power: Based on the environmental parameters, chiller start-up and shutdown parameters, flow rates on the refrigeration side and cooling side, chiller evaporator outlet water temperature, inlet water temperature, and cooling side main pipe inlet and outlet water temperature data in the real-time data, the refrigeration power of each chiller can be predicted.

[0162] (10) Cooling tower power: The mechanism model of cooling tower power is the same as that of variable frequency water pump power model, and the power model of variable frequency water pump can be used as a reference for modeling. Alternatively, the cooling tower power can be predicted based on environmental parameters in real-time data, predicted cooling side flow rate, inlet and outlet water temperature, and cooling tower frequency and start-up / shutdown parameters on a typical day.

[0163] 4. Boundary condition filtering

[0164] Candidate solutions are derived based on K typical daily data and real-time data. The total cooling power corresponding to each candidate solution can be calculated using the inference model. Candidate solutions that do not meet the pre-set boundary conditions, such as requiring at least two chillers to be turned on, are filtered out.

[0165] 5. Sort candidate solutions

[0166] Based on real-time data, the similarity between the current device mode sequence and the device mode sequences corresponding to the K candidate solutions is calculated using the following formula:

[0167]

[0168] Where n represents the number of chillers in the building's refrigeration system, and the is_same_mode function represents cd i,k ,cd j,k The value is 1 if the k-th chiller is turned on simultaneously at times i and j, and 0 if they are not turned on simultaneously. i,j The similarity is the similarity of the device pattern sequences, and the similarity range is [0,1].

[0169] 6. Output of the optimal solution

[0170] Candidate solutions with predicted power less than current power are selected (if none are found, the current control parameters are directly recommended). The optimal solution can be selected by combining the similarity of the device mode sequence, the similarity of environmental parameters, and the magnitude of predicted power in descending, descending, and ascending order, respectively.

[0171] 7. Strategy Recommendations

[0172] The selected control parameters are written into a time series database, and the recommended values ​​displayed on the front-end page are refreshed periodically using long polling.

[0173] III. XGBoost-based Prediction Model

[0174] In the reasoning process of the above building refrigeration system model, besides the parameters that can be inferred using existing linear formulas (corresponding to the mechanism model mentioned above), there are other parameters that do not conform to linear laws. Predicting these non-linear parameters requires establishing numerous prediction models. This solution uses the XGBoost model to perform regression modeling on the parameters that need to be predicted. Taking the evaporator inlet water temperature of the above chiller as an example, the modeling process is as follows:

[0175] (1) Data preprocessing: Before modeling, it is necessary to remove abnormal data that exists during the operation of the building's cooling system, that is, to remove abnormal data including the following three situations:

[0176] a. When the total chilled water flow rate is less than or equal to 0, the refrigeration system is not providing cooling;

[0177] b. If the hourly cooling capacity is less than 0, the refrigeration system is not providing cooling.

[0178] c. Data without environmental variables is difficult to model;

[0179] 80% of the processed historical data was divided into a training set and 20% was used as a validation set for model training and parameter tuning.

[0180] (2) Input feature selection (corresponding to the determination of the category of input parameters mentioned above): Based on the mechanism of the refrigeration system, calculate the Pearson correlation coefficient of the feature data related to the water inlet temperature of the evaporator of the chiller. The Pearson correlation coefficient represents the linear correlation between two variables. Figure 6B A heatmap showing the Pearson correlation coefficients of feature data related to the evaporator inlet water temperature of the chiller is presented. The horizontal and vertical axes represent the feature data to be selected, and the values ​​in different squares indicate the magnitude of the correlation coefficient. Based on the correlation, the following features are selected as model input features: Taking the evaporator inlet water temperature of chiller unit 5 as an example, the input features may include: chilled water supply main temperature, chilled water return main temperature, hourly cooling capacity, chilled water temperature difference, total chilled water flow rate, wet-bulb temperature, ambient temperature, start / stop hours, and the status of chiller units 1, 2, 3, 4, and 5.

[0181] (3) Model building: Mean squared error, mean absolute error, and r can be used. 2 Multiple machine learning regression metrics, such as the coefficient of determination, are used as evaluation metrics. From XGBoost models with different input features, the XGBoost model with the highest model accuracy after training is selected as the prediction model to be used (corresponding to the target machine learning model mentioned above).

[0182] Figure 6C The prediction results of the evaporator inlet water temperature of chiller No. 3 on the validation set are shown. The horizontal axis of the figure is time and the vertical axis is water temperature. The prediction results in this figure fully verify the feasibility of learning the characteristics of refrigeration equipment by introducing relevant variables.

[0183] Model training.

[0184] (4) Model iteration: During the operation of the control system, the prediction model can be updated through regular training to make full use of online data and form a positive cycle iteration.

[0185] As can be seen, this solution, by combining mechanistic and machine learning models and utilizing typical daily and real-time data for model inference, can improve the interpretability of the control method and reduce computational resource consumption during the prediction process while ensuring energy-saving effects.

[0186] Corresponding to the embodiments of the above methods, this disclosure also provides a control device, such as... Figure 7 As shown, the device includes:

[0187] The first acquisition module 710 is used to, in response to receiving a parameter distribution instruction for the energy consumption system, acquire specified data of the energy consumption system within the current time period as target data; wherein, the specified data includes at least operating parameters and environmental data;

[0188] The second acquisition module 720 is used to acquire typical daily data; wherein, the typical daily data is: the specified data required by the energy consumption system under an environment similar to the environment represented by the environmental data in the target data;

[0189] The determining module 730 is used to determine, based on the target data and the typical daily data, candidate operating parameters of the energy consumption system that meet specified conditions in the next time period of the current time period; wherein, the specified conditions are used to ensure that the power of the energy consumption system in the next time period is less than the power in the current time period.

[0190] The adjustment module 740 is used to send the operating parameters to be utilized to the energy consumption system based on the candidate operating parameters, so that the energy consumption system adjusts the current operating parameters to the operating parameters to be utilized.

[0191] Optionally, the second acquisition module includes:

[0192] The calculation submodule is used to calculate the similarity between environmental data in the target data and environmental data in historical specified data; wherein, the historical specified data is specified data of the energy consumption system within a historical period.

[0193] A selection submodule is used to select typical daily data from the specified historical data based on the calculated similarity.

[0194] Optionally, the number of typical daily data points may be multiple;

[0195] The step of determining candidate operating parameters for the energy consumption system that meet specified conditions in the next time period based on the target data and the typical daily data includes:

[0196] Based on the target data and the multiple typical daily data, predict the target operating parameters of the energy consumption system in the next time period of the current time period; wherein, the number of target operating parameters is multiple;

[0197] From the predicted target operating parameters, target operating parameters that meet the specified conditions are selected as candidate operating parameters.

[0198] Optionally, predicting the target operating parameters of the energy consumption system for the next time period based on the target data and the multiple typical daily data includes:

[0199] Multiple input data corresponding to a specified inference model are determined; wherein the input data includes the target data and a typical daily data obtained; wherein the specified inference model is used to predict the operating parameters of the energy consumption system in the next time period;

[0200] Based on the determined input data, the target operating parameters of the energy consumption system for the next time period are determined using the specified inference model.

[0201] Optionally, the operating parameters of the energy consumption system include a first type of parameters and a second type of parameters; the specified inference model includes a mechanism model and a pre-trained target machine learning model; wherein, the mechanism model is a model used to predict the first type of parameters, and the target machine learning model is a model used to predict the second type of parameters; wherein, the first type of parameters are parameters that conform to a linear law, and the second type of parameters are parameters other than the first type of parameters;

[0202] The step of determining the target operating parameters of the energy consumption system for the next time period based on the determined input data and using the specified inference model includes:

[0203] For the determined input data, based on the first data in the input data, the mechanism model is used to predict the first type of parameters of the energy consumption system in the next time period, and based on the second data in the input data, the target machine learning model is used to predict the second type of parameters of the energy consumption system in the next time period, so as to obtain the target operating parameters of the energy consumption system in the next time period.

[0204] Wherein, the first data is the pre-set input parameter data belonging to the mechanism model from the target data and / or typical daily data included in the input data;

[0205] The second data is the pre-set input parameter data belonging to the target machine learning model from the target data and / or typical daily data included in the input data.

[0206] Optionally, the target machine learning model is a machine learning model that meets the preset model accuracy requirements after training multiple machine learning models using sample data, and the input parameters of different machine learning models are of different categories.

[0207] Optionally, the method for determining the category of the input parameters corresponding to the machine learning model includes:

[0208] Calculate the correlation coefficients between each data category of the specified historical data and the output parameters of the machine learning model; wherein, the specified historical data refers to the specified data of the energy consumption system within a historical period;

[0209] From the various data categories, select the data categories whose corresponding correlation coefficients meet the preset threshold as the categories of input parameters for the machine learning model.

[0210] Optionally, the training method of the target machine learning model includes:

[0211] Multiple sample data are acquired; wherein, the sample data includes sample target data and a sample typical day data, the sample target data is the specified data of the energy consumption system within the sample time period, the sample typical day data is the specified data required by the energy consumption system under an environment similar to the environment represented by the environmental data in the sample target data, and the sample data has a specified label, the specified label representing the true value of the operating parameters of the energy consumption system in the next time period of the sample time period;

[0212] Based on the sample data and the initial target machine learning model, the operating parameters of the energy consumption system in the next time period of the sample time period are predicted to obtain the prediction results.

[0213] Based on the specified label and the prediction result, determine the model loss value of the target machine learning model;

[0214] Based on the model loss value, adjust the model parameters of the target machine learning model, and return to the step of obtaining multiple sample data until the target machine learning model converges.

[0215] Optionally, obtaining specified data from the energy consumption system within the current time period as target data includes:

[0216] According to the preset sampling interval, acquire multiple initial specified data of the energy consumption system within the current time period;

[0217] The specified preprocessing is performed on the plurality of initial specified data to obtain specified data of the energy consumption system within the current time period, which is used as target data; wherein, the specified preprocessing includes data aggregation processing of the plurality of specified data.

[0218] Optionally, before performing specified preprocessing on the plurality of initial specified data to obtain specified data of the energy consumption system within the current time period as target data, the method further includes:

[0219] Determine a sequence of device modes corresponding to multiple initial specified data; wherein the sequence of device modes corresponding to the specified data is the sequence of devices of the specified device represented by the specified data that is in the on state;

[0220] From the plurality of initial specified data, remove specified data corresponding to the specified pattern sequence;

[0221] The specified mode sequence is the sequence other than the device mode sequence with the longest runtime among the determined device mode sequences.

[0222] Optionally, the number of candidate operating parameters may be multiple;

[0223] The step of issuing the operational parameters to be utilized to the energy consumption system based on the candidate operational parameters includes:

[0224] The candidate operating parameters are sorted according to preset sorting conditions; wherein, the sorting conditions include sorting index and sorting method, and the sorting method is such that when sorted according to the sorting index, the first digit of the resulting sequence is the optimal operating parameter.

[0225] The first operating parameter is selected from the sorted sequence and sent to the energy consumption system as the operating parameter to be used.

[0226] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0227] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0228] An electronic device provided in this disclosure may include:

[0229] At least one processor; and

[0230] A memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform any of the aforementioned control methods.

[0231] This disclosure provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of any of the above-described control methods.

[0232] The present disclosure provides a computer program product containing instructions that, when run on a computer, causes the computer to perform the steps of any of the control methods described in the above embodiments.

[0233] Figure 8 A schematic block diagram of an example electronic device 800 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0234] like Figure 8 As shown, device 800 includes a computing unit 801, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 802 or a computer program loaded from storage unit 808 into random access memory (RAM) 803. RAM 803 may also store various programs and data required for the operation of device 800. The computing unit 801, ROM 802, and RAM 803 are interconnected via bus 804. Input / output (I / O) interface 805 is also connected to bus 804.

[0235] Multiple components in device 800 are connected to I / O interface 805, including: input unit 806, such as keyboard, mouse, etc.; output unit 807, such as various types of monitors, speakers, etc.; storage unit 808, such as disk, optical disk, etc.; and communication unit 809, such as network card, modem, wireless transceiver, etc. Communication unit 809 allows device 800 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0236] The computing unit 801 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 801 performs the various methods and processes described above, such as control methods. For example, in some embodiments, the control method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 808. In some embodiments, part or all of the computer program may be loaded and / or installed on device 800 via ROM 802 and / or communication unit 809. When the computer program is loaded into RAM 803 and executed by the computing unit 801, one or more steps of the control method described above may be performed. Alternatively, in other embodiments, the computing unit 801 may be configured to perform control methods by any other suitable means (e.g., by means of firmware).

[0237] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0238] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0239] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0240] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0241] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0242] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0243] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0244] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A control method, comprising: In response to receiving a parameter command for the energy consumption system, the specified data of the energy consumption system within the current time period is obtained as target data; wherein, the specified data includes at least operating parameters and environmental data; Acquire typical daily data; wherein, the typical daily data is: designated data selected from the historical designated data based on the similarity between the environmental data in the target data and the environmental data in the historical designated data; the historical designated data is the designated data of the energy consumption system within a historical period, and the environment represented by the environmental data in the selected designated data is similar to the environment represented by the environmental data in the target data; Based on the target data and the typical daily data, the target operating parameters of the energy consumption system in the next time period of the current time period are predicted. From the predicted target operating parameters, target operating parameters that meet specified conditions are selected. If a selected target operating parameter is found, it is determined as a candidate operating parameter. If no selected target operating parameter is found, the operating parameters in the target data are determined as candidate operating parameters. The specified conditions are used to ensure that the power of the energy consumption system in the next time period is less than the power in the current time period. Based on the candidate operating parameters, the energy consumption system is issued operating parameters to be utilized, so that the energy consumption system adjusts its current operating parameters to the operating parameters to be utilized.

2. The method according to claim 1, wherein, The acquisition of typical daily data includes: Calculate the similarity between environmental data in the target data and environmental data in historical specified data; wherein, the historical specified data is specified data of the energy consumption system within a historical period; Based on the calculated similarity, typical daily data are selected from the specified historical data.

3. The method according to claim 1 or 2, wherein, The number of typical daily data points is multiple; The step of predicting the target operating parameters of the energy consumption system for the next time period based on the target data and the typical daily data includes: Based on the target data and the multiple typical daily data, predict the target operating parameters of the energy consumption system for the next time period in the current time period; wherein, the number of target operating parameters is multiple.

4. The method according to claim 3, wherein, The step of predicting the target operating parameters of the energy consumption system for the next time period based on the target data and the multiple typical daily data includes: Multiple input data corresponding to a specified inference model are determined; wherein the input data includes the target data and a typical daily data obtained; wherein the specified inference model is used to predict the operating parameters of the energy consumption system in the next time period; Based on the determined input data, the target operating parameters of the energy consumption system for the next time period are determined using the specified inference model.

5. The method according to claim 4, wherein, The operating parameters of the energy consumption system include a first type of parameter and a second type of parameter; the specified inference model includes a mechanism model and a pre-trained target machine learning model; wherein, the mechanism model is a model used to predict the first type of parameter, and the target machine learning model is a model used to predict the second type of parameter; wherein, the first type of parameter is a parameter that conforms to a linear law, and the second type of parameter is a parameter other than the first type of parameter; The step of determining the target operating parameters of the energy consumption system for the next time period based on the determined input data and using the specified inference model includes: For the determined input data, based on the first data in the input data, the mechanism model is used to predict the first type of parameters of the energy consumption system in the next time period, and based on the second data in the input data, the target machine learning model is used to predict the second type of parameters of the energy consumption system in the next time period, so as to obtain the target operating parameters of the energy consumption system in the next time period. Wherein, the first data is the pre-set input parameter data belonging to the mechanism model from the target data and / or typical daily data included in the input data; The second data is the pre-set input parameter data belonging to the target machine learning model from the target data and / or typical daily data included in the input data.

6. The method according to claim 5, wherein, The target machine learning model is a machine learning model that meets the preset model accuracy requirements after training multiple machine learning models using sample data. Different machine learning models correspond to different categories of input parameters.

7. The method according to claim 6, wherein, The methods for determining the categories of input parameters corresponding to the machine learning model include: Calculate the correlation coefficients between each data category of the specified historical data and the output parameters of the machine learning model; wherein, the specified historical data refers to the specified data of the energy consumption system within a historical period; From the various data categories, select the data categories whose corresponding correlation coefficients meet the preset threshold as the categories of input parameters for the machine learning model.

8. The method according to claim 5, wherein, The training methods for the target machine learning model include: Multiple sample data are acquired; wherein, the sample data includes sample target data and a sample typical day data, the sample target data is the specified data of the energy consumption system within the sample time period, the sample typical day data is the specified data required by the energy consumption system under an environment similar to the environment represented by the environmental data in the sample target data, and the sample data has a specified label, the specified label representing the true value of the operating parameters of the energy consumption system in the next time period of the sample time period; Based on the sample data and the initial target machine learning model, the operating parameters of the energy consumption system in the next time period of the sample time period are predicted to obtain the prediction results. Based on the specified label and the prediction result, determine the model loss value of the target machine learning model; Based on the model loss value, adjust the model parameters of the target machine learning model, and return to the step of obtaining multiple sample data until the target machine learning model converges.

9. The method according to claim 1 or 2, wherein, The step of acquiring specified data from the energy consumption system within the current time period as target data includes: According to the preset sampling interval, acquire multiple initial specified data of the energy consumption system within the current time period; The specified preprocessing is performed on the plurality of initial specified data to obtain specified data of the energy consumption system within the current time period, which is used as target data; wherein, the specified preprocessing includes data aggregation processing of the plurality of specified data.

10. The method according to claim 9, wherein, Before performing specified preprocessing on the plurality of initial specified data to obtain specified data of the energy consumption system within the current time period, which is then used as the target data, the process further includes: Determine a sequence of device modes corresponding to multiple initial specified data; wherein the sequence of device modes corresponding to the specified data is the sequence of devices of the specified device represented by the specified data that is in the on state; From the plurality of initial specified data, remove specified data corresponding to the specified pattern sequence; The specified mode sequence is the sequence other than the device mode sequence with the longest runtime among the determined device mode sequences.

11. The method according to claim 1 or 2, wherein, The number of candidate operating parameters is multiple; The step of issuing the operational parameters to be utilized to the energy consumption system based on the candidate operational parameters includes: The candidate operating parameters are sorted according to preset sorting conditions; wherein, the sorting conditions include sorting index and sorting method, and the sorting method is such that when sorted according to the sorting index, the first digit of the resulting sequence is the optimal operating parameter. The first operating parameter is selected from the sorted sequence and sent to the energy consumption system as the operating parameter to be used.

12. A control device, comprising: The first acquisition module is used to, in response to receiving a parameter distribution instruction for the energy consumption system, acquire specified data of the energy consumption system within the current time period as target data; wherein, the specified data includes at least operating parameters and environmental data; The second acquisition module is used to acquire typical daily data; wherein, the typical daily data is: designated data selected from the historical designated data based on the similarity between the environmental data in the target data and the environmental data in the historical designated data; the historical designated data is the designated data of the energy consumption system within a historical period, and the environment represented by the environmental data in the selected designated data is similar to the environment represented by the environmental data in the target data; The determination module is used to predict the target operating parameters of the energy consumption system in the next time period of the current time period based on the target data and the typical daily data, and to filter the target operating parameters that meet the specified conditions from the predicted target operating parameters. If the selected target operating parameters are found, they are determined as candidate operating parameters. If the selected parameters are not found, the operating parameters in the target data are determined as candidate operating parameters. The specified conditions are used to ensure that the power of the energy consumption system in the next time period is less than the power in the current time period. The adjustment module is used to issue the operating parameters to be utilized to the energy consumption system based on the candidate operating parameters, so that the energy consumption system adjusts the current operating parameters to the operating parameters to be utilized.

13. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-11.

14. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-11.

15. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-11.

Images