Central air conditioning system control method and device, electronic equipment and storage medium

Abstract

The embodiment of the application provides a central air conditioning system control method and device, electronic equipment and a storage medium, and belongs to the field of multi-target control optimization. The method comprises the following steps: obtaining historical operation data and at least two task scenes of a central air conditioning system, then constructing sample data sets of each task scene according to preset performance indexes and the historical operation data under different task scenes, and inputting the sample data sets into a preset target system regulation and control prediction model to obtain candidate regulation and control prediction data through regulation and control prediction, optimizing the candidate regulation and control prediction data according to a preset optimization model to obtain target regulation and control prediction data, and controlling the central air conditioning system according to the target control prediction data. The embodiment of the application can screen out candidate regulation and control prediction data through a target system regulation and control prediction model for different task scenes and in combination with historical operation data, and then obtain target regulation and control prediction data through an optimization model, so as to control the central air conditioning system, so that the system is more energy-saving.

Description

Technical Field

[0001] This application relates to the field of multi-objective control optimization, and in particular to a control method and apparatus, electronic equipment and storage medium for a central air conditioning system. Background Technology

[0002] Currently, most intelligent control and energy-saving technologies for central air conditioning systems employ modular, single-objective control optimization methods. This approach may yield satisfactory results in some areas but unsatisfactory results in others. For example, while modeling and optimizing based on minimizing the main unit's energy consumption as the single objective often guarantees a certain level of accuracy, it suffers from poor versatility and overall system integration. Even with the lowest main unit power, increased power consumption by water pumps and cooling towers can lead to still significant system power consumption. Therefore, meeting the multi-objective optimization requirements for efficient linkage between various modules of the air conditioning system, and achieving both cooling capacity requirements and energy conservation, has become an urgent technical challenge. Summary of the Invention

[0003] The main objective of this application is to provide a control method and device, electronic device and storage medium for a central air conditioning system, which aims to reduce the system power consumption of the central air conditioning system while meeting the cooling demand.

[0004] To achieve the above objectives, a first aspect of this application provides a control method for a central air conditioning system, the method comprising:

[0005] Obtain historical operating data of the central air conditioning system and at least two task scenarios; wherein, different task scenarios focus on different preset performance indicators of the central air conditioning system;

[0006] A sample dataset for each task scenario is constructed based on the preset performance indicators and the historical running data;

[0007] The sample dataset is input into a preset target system regulation and prediction model; wherein, the target system regulation and prediction model includes: a gating network, a feature extraction network, and a regulation and prediction network;

[0008] The target weight information for each task scenario is obtained by weighting the sample dataset through the gating network.

[0009] The target feature vector is obtained by performing feature mapping on the sample dataset using the feature extraction network and the target weight information;

[0010] The target weight information and the target feature vector are controlled and predicted by the control and prediction network to obtain control and prediction data.

[0011] The control and prediction data are optimized according to a preset optimization model to obtain the target control data;

[0012] The central air conditioning system is controlled based on the target control data.

[0013] According to some embodiments of the present invention, before inputting the sample dataset into a preset multi-objective system regulation and prediction model, the method further includes:

[0014] Constructing the target system regulation and prediction model specifically includes:

[0015] Obtain the training dataset for each of the aforementioned task scenarios in the central air conditioning system; wherein, the training dataset includes: historical training dataset and validation dataset;

[0016] The historical training dataset is input into a preset original system regulation and prediction model; wherein, the original system regulation and prediction model includes: a gating network, a feature extraction network, and regulation and prediction networks for each sub-network;

[0017] The training dataset is weighted using the gating network to obtain training weight information for each task scenario;

[0018] The historical training dataset is mapped using the feature extraction network to obtain training feature vectors;

[0019] The training weight information and the training feature vector are adjusted and predicted by the prediction network through each sub-network to obtain the training adjustment and prediction data for each task scenario.

[0020] Based on the training and control prediction data for each task scenario, the validation dataset, and the training weight information, loss calculation is performed to obtain training loss data;

[0021] The parameters of the preset original system regulation and prediction model are adjusted based on the training loss data to obtain the target system regulation and prediction model.

[0022] According to some embodiments of the present invention, the feature extraction network includes: a shared underlying feature extraction sub-network, a task feature extraction sub-network for each task scenario, and a feature filtering sub-network. The step of performing feature mapping on the sample dataset using the feature extraction network and the target weight information to obtain a target feature vector includes:

[0023] The shared underlying feature extraction subnetwork is used to extract shared features from the sample dataset to obtain a shared underlying feature vector;

[0024] The task feature extraction sub-network is used to extract task features from the sample dataset to obtain the task-level sub-feature vector for each task scenario.

[0025] The shared underlying feature vector and the task underlying sub-feature vector are concatenated according to the target weight information to obtain the candidate feature vector;

[0026] The candidate feature vectors are filtered through the feature filtering sub-network to obtain the target feature vector.

[0027] According to some embodiments of the present invention, the step of concatenating the shared underlying feature vector and the task underlying sub-feature vector according to the target weight information to obtain a candidate feature vector includes:

[0028] Based on the target weight information, the weighted concatenation of all task-level sub-feature vectors for each task scenario is used to obtain the training-level feature vector for each task scenario.

[0029] The candidate feature vector is obtained by concatenating the training low-level feature vector and the shared low-level feature vector.

[0030] According to some embodiments of the present invention, the step of calculating the training loss data based on the training control prediction data for each task scenario, the validation dataset, and the training weight information includes:

[0031] The training loss function for each task scenario is constructed based on the validation dataset and the training and prediction data for each task scenario.

[0032] The training loss function for each task scenario is weighted and summed based on the training weight information to obtain the total training loss function.

[0033] The training loss data is obtained by converging the total training loss function.

[0034] According to some embodiments of the present invention, the target control prediction data includes state variable information; the step of optimizing the candidate control prediction data according to a preset optimization model to obtain the target control prediction data specifically includes:

[0035] The candidate regulation prediction data are input into the optimization model to obtain the objective function;

[0036] By adjusting the variables within the candidate regulation prediction data until the objective function meets the preset conditions, the corresponding candidate regulation prediction data is used as the target regulation prediction data.

[0037] According to some embodiments of the present invention, obtaining the training dataset of the central air conditioning system includes:

[0038] Obtain the historical training operation data of the central air conditioning system;

[0039] After filtering out abnormal data through data filtering preprocessing, the historical training data is used to obtain the initial selected historical data.

[0040] The initial historical data is sliced ​​to obtain training input variable data and training target variable data;

[0041] The training dataset and the validation dataset are constructed based on the training input variable data and the training target variable data.

[0042] To achieve the above objectives, a second aspect of this application provides a control device for a central air conditioning system, the device comprising:

[0043] The data acquisition module is used to acquire historical operating data of the central air conditioning system and at least two task scenarios; wherein, different task scenarios focus on different preset performance indicators of the central air conditioning system.

[0044] The sample construction module is used to construct a sample dataset for each of the task scenarios based on the preset performance indicators and the historical running data.

[0045] The sample input module is used to input the sample dataset into a preset target system regulation and prediction model; wherein, the target system regulation and prediction model includes: a gating network, a feature extraction network, and a regulation and prediction network;

[0046] The weight allocation module is used to allocate weights to the sample dataset through the gating network to obtain target weight information for each task scenario;

[0047] The feature extraction module is used to perform feature mapping on the sample dataset through the feature extraction network and the target weight information to obtain the target feature vector;

[0048] The regulation prediction module is used to perform regulation prediction on the target weight information and the target feature vector through the regulation prediction network to obtain candidate regulation prediction data.

[0049] The control optimization module is used to optimize the candidate regulation prediction data according to a preset optimization model to obtain the target regulation prediction data.

[0050] The control module is used to control the central air conditioning system based on the target regulation prediction data.

[0051] To achieve the above objectives, a third aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect.

[0052] To achieve the above objectives, a fourth aspect of the present application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect.

[0053] The control method, device, electronic equipment, and storage medium for a central air conditioning system proposed in this application acquire historical operating data and at least two task scenarios of the central air conditioning system. Different task scenarios focus on different preset performance indicators in the central air conditioning system. A sample dataset for each task scenario is constructed based on the preset performance indicators and historical operating data. The sample dataset is input into a preset target system regulation prediction model, which includes a gating network, a feature extraction network, and a regulation prediction network. The gating network assigns weights to the sample dataset to obtain target weight information for each task scenario. The feature extraction network and the target weight information are used to perform feature mapping on the sample dataset to obtain a target feature vector. The regulation prediction network regulates the target weight information and the target feature vector to obtain candidate regulation prediction data. The candidate regulation prediction data is optimized according to a preset optimization model to obtain the target regulation prediction data. Therefore, the embodiments of this application can initially screen candidate control prediction data through the target system control prediction model for different task scenarios and in combination with historical operating data, and then further optimize the candidate control prediction data through the optimization model to obtain target control prediction data, and control the central air conditioning system according to the target control prediction data to achieve the lowest total power consumption of the central air conditioning system. Attached Figure Description

[0054] Figure 1 This is a flowchart of a control method for a central air conditioning system provided in an embodiment of this application;

[0055] Figure 2 This is a flowchart of the steps involved in constructing a target system regulation and prediction model;

[0056] Figure 3 yes Figure 1 The flowchart of step S105 in the process;

[0057] Figure 4 yes Figure 3 The flowchart of step S303 in the process;

[0058] Figure 5 yes Figure 2The flowchart of step S205 in the document;

[0059] Figure 6 yes Figure 1 The flowchart of step S107 in the process;

[0060] Figure 7 yes Figure 2 The flowchart of step S201 in the text;

[0061] Figure 8 This is a schematic diagram of the structure of the control device of a central air conditioning system provided in another embodiment of this application;

[0062] Figure 9 This is a schematic diagram of the hardware structure of an electronic device provided in another embodiment of this application. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0064] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0065] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0066] First, let's analyze some of the terms used in this application:

[0067] Bidirectional Long Short-Term Memory (BiLSTM) recurrent neural network models are suitable for natural language processing (NLP) tasks. Compared to traditional unidirectional LSTM models, BiLSTM processes both forward and backward input sequences simultaneously at each time step, thus better capturing contextual information.

[0068] Deep learning models (Convolutional Neural Networks, CNNs) are suitable for processing image and spatial data. Their core idea is to extract local features from the input data through convolution operations and reduce the size of the feature maps through pooling operations, thereby achieving abstraction and compression of the input data. Transformers are neural network models based on self-attention mechanisms, primarily used for Natural Language Processing (NLP) tasks. They achieve interaction and information transfer between different positions through self-attention, solving some problems of traditional sequence models.

[0069] In building energy consumption, lighting accounts for approximately 20-30% of total energy consumption, drainage and transportation account for about 20%, while central air conditioning, as the most important component, accounts for 40%-60% of total building energy consumption, and this trend is increasing further. Therefore, energy conservation in air conditioning systems is crucial for building carbon neutrality. Common central air conditioning systems generally include several major energy-consuming components such as chillers, cooling towers, chilled water pumps, and cooling water pumps. Typically, central air conditioning systems are designed for the most unfavorable operating conditions, generally considering the maximum load. However, in actual operation, the air conditioning load is a dynamic process, with full load or even high load accounting for only a small proportion. Therefore, adopting reasonable energy-saving control strategies can significantly tap into the energy-saving potential of central air conditioning systems. This involves considering advanced technological solutions during the initial design phase and adjusting the cooling capacity supply in a timely manner during operation management to fully exploit the energy-saving potential under low cooling load conditions. The main challenge in solving the current energy-saving problem of air conditioning systems has shifted to the operational coordination between various sub-equipment and efficient linkage with other systems. With the rapid advancement of "smart city" construction, the realization of intelligent control and energy saving of central air conditioning has also been put on the agenda.

[0070] Currently, most intelligent control and energy-saving technologies for central air conditioning systems employ modular, single-objective control optimization methods. This approach may yield satisfactory results in some areas but unsatisfactory results in others. For example, while modeling and optimizing based on minimizing the main unit's energy consumption as the single objective often guarantees a certain level of accuracy, it suffers from poor versatility and overall system integration. Even with the lowest main unit power, increased power consumption by water pumps and cooling towers can lead to still significant system power consumption. Therefore, meeting the multi-objective optimization requirements for efficient linkage between various modules of the air conditioning system, and achieving both cooling capacity requirements and energy conservation, has become an urgent technical challenge.

[0071] Based on this, embodiments of this application provide a control method and apparatus, electronic device and storage medium for a central air conditioning system, which aims to reduce the system power consumption of the central air conditioning system while meeting the cooling demand.

[0072] The control method, apparatus, electronic device, and storage medium for the central air conditioning system provided in this application are specifically described through the following embodiments. First, the control method for the central air conditioning system in this application embodiment is described.

[0073] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0074] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0075] The central air conditioning system control method provided in this application relates to the field of artificial intelligence technology. The central air conditioning system control method provided in this application can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, etc.; the server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application implementing the central air conditioning system control method, but is not limited to the above forms.

[0076] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0077] Figure 1 This is an optional flowchart of the control method for a central air conditioning system provided in the embodiments of this application. Figure 1 The method may include, but is not limited to, steps S101 to S108.

[0078] Step S101: Obtain historical operating data of the central air conditioning system and at least two task scenarios, wherein different task scenarios focus on different preset performance indicators in the central air conditioning system.

[0079] Step S102: Construct a sample dataset for each task scenario based on preset performance indicators and historical running data;

[0080] Step S103: Input the sample dataset into the preset target system regulation prediction model, wherein the target system regulation prediction model includes: a gating network, a feature extraction network, and a regulation prediction network;

[0081] Step S104: Weights are assigned to the sample dataset through a gating network to obtain the target weight information for each task scenario.

[0082] Step S105: The sample dataset is feature-mapped using a feature extraction network and target weight information to obtain the target feature vector;

[0083] Step S106: The target weight information and target feature vector are controlled and predicted by the control prediction network to obtain candidate control prediction data;

[0084] Step S107: Optimize the candidate regulation prediction data according to the preset optimization model to obtain the target regulation prediction data;

[0085] Step S108: Control the central air conditioning system according to the target control prediction data.

[0086] Steps S101 to S108 of this application embodiment involve acquiring historical operating data of the central air conditioning system and at least two task scenarios. Different task scenarios focus on different preset performance indicators in the central air conditioning system. A sample dataset for each task scenario is then constructed based on the preset performance indicators and historical operating data. This sample dataset is input into a preset target system control prediction model. The target system control prediction model includes a gating network, a feature extraction network, and a control prediction network. The gating network assigns weights to the sample dataset to obtain target weight information for each task scenario. The feature extraction network and the target weight information are used to perform feature mapping on the sample dataset to obtain a target feature vector. The control prediction network then performs control prediction on the target weight information and the target feature vector to obtain candidate control prediction data that meets the basic power consumption and energy-saving requirements of practical applications. Finally, the candidate control prediction data is optimized using a preset optimization model to obtain target control prediction data. The central air conditioning system is then controlled based on the target control prediction data to achieve the lowest possible total power consumption.

[0087] In step S101 of some embodiments, historical operating logs of the central air conditioning system can be obtained to collect historical operating data and at least two task scenarios, with a collection frequency f on the order of minutes. Simultaneously, data analysis is performed according to different task scenarios. For example, variables related to host energy consumption, chilled water pump energy consumption, cooling pump energy consumption, and cooling tower energy consumption are analyzed and collected, such as the number of host units in operation, chilled water inlet temperature, chilled water outlet temperature, chilled water outlet-return temperature difference, chilled water pump frequency, chilled water pump flow rate, pressure difference, head, host-side cooling water inlet temperature, cooling pump frequency, wet-bulb temperature, humidity, and cooling tower-side cooling water outlet temperature, etc. (m features in total), and their corresponding target values ​​at any given time, such as host power, chilled water pump power, cooling pump power, cooling tower power, and cooling tower-side cooling water outlet temperature, collecting a total of n historical data samples.

[0088] In step S102 of some embodiments, a sample dataset for each task scenario is constructed based on historical data samples corresponding to each task scenario, thereby enabling more accurate power consumption prediction and control prediction.

[0089] In some embodiments, prior to step S102, the control method for a central air conditioning system further includes: constructing a target system regulation prediction model. This is because the control method for a central air conditioning system requires pre-constructing and training the target system regulation prediction model to perform preliminary regulation predictions on the sample dataset before solving for the target regulation prediction data.

[0090] Please see Figure 2The construction of the target system regulation and prediction model may include, but is not limited to, steps S201 to S207:

[0091] Step S201: Obtain the training dataset for each task scenario in the central air conditioning system, wherein the training dataset includes: historical training dataset and validation dataset;

[0092] Step S202: Input the historical training dataset into the preset original system regulation prediction model, wherein the original system regulation prediction model includes: a gating network, a feature extraction network, and regulation prediction networks of each sub-network;

[0093] Step S203: The training dataset is weighted using a gating network to obtain the training weight information for each task scenario.

[0094] Step S204: The historical training dataset is mapped using a feature extraction network to obtain training feature vectors;

[0095] Step S205: The training weight information and training feature vector are adjusted and predicted by each sub-network to obtain the training adjustment and prediction data for each task scenario.

[0096] Step S206: Calculate the loss based on the training control prediction data, validation dataset, and training weight information for each task scenario to obtain training loss data.

[0097] Step S207: Adjust the parameters of the preset original system regulation prediction model based on the training loss data to obtain the target system regulation prediction model.

[0098] In step S201 of some embodiments, it should be noted that the training dataset can be divided into historical training dataset and validation dataset by random allocation. Other methods can also be used to divide the training dataset, and this is not limited to these.

[0099] In step S202 of some embodiments, the historical training dataset is input into a preset original system regulation prediction model to make a preliminary regulation prediction on the historical training dataset.

[0100] In step S203 of some embodiments, the number of weights output by the gating network depends on the number of task scenarios involved in the historical training dataset, and a weight shared by a lower-level feature extraction network is added. Therefore, the first layer has a total of N+1 gating networks, where N is the number of task scenarios. Each task has its own weights. Learning the weights of different lower-level feature networks is essentially learning the weights of different sample tasks. Each task obtains the weights of each lower-level feature network through the gating network. During gradient backpropagation, the parameter update calculation of different lower-level feature extraction networks is directly related to the weights, which is equivalent to using different weights for samples of different tasks in the lower-level feature extraction layer.

[0101] It should be further explained that the gating network consists of a fully connected layer and a softmax output layer. The fully connected layer is used to compute a linear transformation of the input features, mapping the features from the historical training sample set to a new representation space. The parameters of the fully connected layer are learned during pre-training. Next, the softmax output layer is used to compute the output weights of the gating units corresponding to each task. The softmax function transforms the input values ​​into a probability distribution such that the sum of the output weights of the gating units corresponding to each task equals 1. In this way, the output weights of each gating unit can be used to control the importance of the input features, thereby determining which features are more useful for the current task.

[0102] In step S204 of some embodiments, the feature extraction network includes: a task-specific low-level feature extraction sub-network for each task scenario and a task-shared low-level feature extraction sub-network for each task scenario. The shared low-level feature extraction sub-network and the task-specific low-level feature extraction sub-network for each task scenario are used to perform feature mapping on the historical training dataset to obtain a training feature vector. This means that the common feature information and the unique feature information for each task scenario are considered in the subsequent preliminary regulation prediction, thereby improving the accuracy of the subsequent preliminary regulation prediction.

[0103] In step S205 of some embodiments, each sub-network regulation prediction network is a prediction network unique to each sub-task built on the fully connected layer of the gated network. It can input the weighted and filtered feature vector corresponding to each task scenario into its respective prediction network for regulation prediction according to the training weight information, and output the training regulation prediction data of each task scenario to achieve preliminary regulation prediction.

[0104] In step S206 of some embodiments, loss calculation is performed based on the training adjustment prediction data, validation dataset, and training weight information for each task scenario to obtain training loss data. The training loss data includes the predicted training loss rate for each task scenario and the maximum influential variable for each task scenario. The training loss data is obtained by preliminary adjustment prediction for subsequent preliminary model adjustment.

[0105] In step S207 of some embodiments, the parameters of the preset original system regulation prediction model are adjusted according to the maximum influence variable of each task scenario in the training loss data to obtain a target system regulation prediction model that is more accurate in regulation prediction than the original system regulation prediction model.

[0106] Steps S201 to S207 as shown in the embodiments of this application involve obtaining the historical training dataset of the central air conditioning system, using different weights calculated by the gating network according to different task scenarios, and the pre-trained original regulation prediction model to perform preliminary regulation prediction on the historical training dataset, and calculating the loss based on the obtained training regulation prediction data to obtain training loss data, so as to adjust the original system regulation prediction model according to the training loss data to obtain a more accurate target system regulation prediction model.

[0107] In step S103 of some embodiments, the sample dataset is input into a pre-trained target system regulation prediction model to make more accurate regulation predictions on the sample dataset.

[0108] In step S104 of some embodiments, the sample dataset is weighted by a gating network to obtain the target weight information of the task in each task scenario. For example, a matrix x is generated based on the relevant data corresponding to one of the task scenarios k in the sample dataset and input into the gating network, and the output is the normalized weight w. k (x).

[0109] Please see Figure 3 In some embodiments, step S105 includes: a shared underlying feature extraction subnetwork, a task feature extraction subnetwork for each task scenario, and a feature filtering subnetwork. Step S105 may include, but is not limited to, steps S301 to S304.

[0110] Step S301: Extract shared features from the sample dataset using a shared underlying feature extraction sub-network to obtain a shared underlying feature vector;

[0111] Step S302: Extract task features from the sample dataset through the task feature extraction sub-network to obtain the underlying feature vector of each task scenario.

[0112] Step S303: Concatenate the shared underlying feature vector and the task underlying feature vector according to the target weight information to obtain the candidate feature vector;

[0113] Step S304: The candidate feature vectors are filtered through the feature filtering sub-network to obtain the target feature vector.

[0114] In step S301 of some embodiments, the shared underlying feature extraction subnetwork analyzes and extracts common feature information in the sample dataset, and then vectorizes the common feature information to obtain the shared underlying feature vector.

[0115] In step S302 of some embodiments, the task feature extraction subnetwork obtains the underlying feature vector of each task scenario by analyzing the task feature information corresponding to each task scenario in the extracted sample dataset.

[0116] It should be noted that both the underlying feature extraction network and the shared underlying feature extraction sub-network are composed of three task feature extraction sub-networks: BiLSTM, CNN, and Transformer.

[0117] In step S303 of some embodiments, it should be noted that each task scenario has at least one unique underlying feature vector. The shared underlying feature vector and the task underlying feature vector are concatenated according to the target weight information to obtain a candidate feature vector that can reflect the feature information of each task scenario.

[0118] In step S304 of some embodiments, the feature filtering subnetwork filters candidate feature vectors to obtain the target feature vector. For example, when initially performing feature extraction on the sample dataset, the feature filtering subnetwork multiplies the weights calculated by the N+1 gating networks for each task scenario, obtained from the first-layer gating network, with the underlying feature vectors and shared underlying feature vectors of the N+1 subtasks respectively to obtain the input of the next feature extraction layer. For example, for the host power task k, the target feature vector g is obtained after filtering by the gating network. k (x), expressed as follows:

[0119]

[0120] g k (x)=w k (x)F k (x)

[0121] in, The feature vector extracted by the underlying feature extraction network for task k. To share the feature vectors extracted by the underlying feature subnetwork, F k(x) is the 2-row matrix obtained by concatenating two vectors, w k (x) represents the weights calculated by the gating network.

[0122] Steps S301 to S304, as illustrated in this embodiment, involve extracting shared features from the training dataset using a shared low-level feature extraction sub-network to obtain a shared low-level feature vector. Then, a task feature extraction sub-network is used to extract task features from the training dataset to obtain low-level feature vectors for each task scenario. Next, the shared low-level feature vector and the task low-level feature vector are concatenated according to target weight information to obtain candidate feature vectors that reflect the feature information of each task scenario. Finally, a feature filtering sub-network performs layer-by-layer filtering on the candidate feature vectors to obtain the target feature vector that best reflects the feature information of each task scenario.

[0123] Please see Figure 4 In some embodiments, step S303 may include, but is not limited to, steps S401 to S402:

[0124] Step S401: Based on the target weight information, the weighted sub-feature vectors of all tasks in each task scenario are concatenated to obtain the training sub-feature vector of each task scenario.

[0125] Step S402: The training low-level feature vector and the shared low-level feature vector are concatenated to obtain the candidate feature vector.

[0126] In step S401 of some embodiments, the target weight information is learned by the gating network for each task scenario. Based on the target weight information, the weighted concatenation of all task-level sub-feature vectors of each task scenario is used to obtain the training-level feature information of each task scenario. Then, the training-level feature vector is obtained by vectorization.

[0127] In step S402 of some embodiments, candidate feature vectors that can reflect both common features and unique features are obtained by concatenating the trained low-level feature vectors and the shared low-level feature vectors.

[0128] Steps S401 to S402 as shown in the embodiments of this application involve weighted concatenation of all task-level sub-feature vectors of each task scenario according to the target weight information to obtain the training-level feature vector of each task scenario, and then concatenating the training-level feature vector and the shared-level feature vector to obtain a candidate feature vector that can reflect both common features and unique features.

[0129] Please see Figure 5 In some embodiments, step S205 may also include, but is not limited to, steps S501 to S503:

[0130] Step S501: Construct the training loss function for each task scenario based on the validation dataset and the training and prediction data for each task scenario.

[0131] Step S502: The training loss function for each task scenario is weighted and summed according to the training weight information to obtain the total training loss function.

[0132] Step S503: Converge the total training loss function to obtain the training loss data.

[0133] In step S501 of some embodiments, the verification dataset includes real labeled data. The training control prediction data for each task scenario is subtracted from the real labeled data for each task scenario to obtain the training loss value. The training loss function for each task scenario is then constructed based on the training loss value.

[0134] In step S502 of some embodiments, the training loss function for each task scenario is weighted and summed according to the training weight information to obtain the total training loss function. Since it is a multi-task model, different results are output under different task scenarios. Therefore, a joint training method is adopted. For example, if all N tasks are regression prediction tasks, then the total loss function L is the weighted sum of the training loss functions for each task scenario, and the expression is as follows:

[0135]

[0136] Where K is the number of subtasks, α k y represents the weights of the loss function for each subtask. ′k For each task scenario, the training and prediction data are adjusted, y k For each task scenario, there is real-world labeled data.

[0137] In step S503 of some embodiments, the model parameters are updated through multi-step convergence training to minimize the total loss function L, and the prediction accuracy is verified by validating the test set. When the accuracy meets the actual business requirements, the model adjustment parameters can be saved and used as training loss data for subsequent adjustment and optimization of the prediction model in the original system.

[0138] Steps S501 to S503, as illustrated in this embodiment, involve constructing a training loss function for each task scenario based on the validation dataset and the training and prediction data for each task scenario. Then, a joint training approach is used, and the training loss functions for each task scenario are weighted and summed according to the training weight information to obtain the total training loss function. Finally, the total training loss function is converged to obtain training loss data that can be used for subsequent optimization of the original system's regulation and prediction model.

[0139] Please see Figure 6 In some embodiments, step S502 includes, but is not limited to, steps S601 to S602:

[0140] Step S601: Input the candidate regulation prediction data into the optimization model to obtain the objective function;

[0141] Step S602: Adjust the variables in the candidate control prediction data until the objective function meets the preset conditions, and then use the corresponding candidate control prediction data as the target control prediction data.

[0142] In step S601 of some embodiments, particle swarm optimization is used to solve for the optimal control strategy for system energy saving, and the solution variables are a control variables C. The objective function J of the optimization algorithm model is designed based on candidate regulation prediction data, and its expression is as follows:

[0143]

[0144] y1,y2,......,y k =Model(x)

[0145] Where y1, y2, ..., y k For each task scenario, β represents the candidate control prediction data. k For each task scenario, the weight x is composed of control variable C, other state variables, and historical data. C sets the lower and upper limits of each control variable as needed, and together with other constraints, constitutes the description of this optimization problem.

[0146] It should be noted that the particle swarm optimization method first randomly initializes the velocity and position of particles. Each particle can be regarded as a search individual in an N-dimensional search space. The current position of the particle is a candidate solution for the corresponding optimization problem, and the particle's flight process is the search process for that individual. The particle's flight speed can be dynamically adjusted based on the particle's historical best position and the population's historical best position. Particles have only two attributes: velocity and position. Velocity represents the speed of movement, and position represents the direction of movement. The optimal solution searched by each particle individually is called the individual extreme value, and the optimal individual extreme value in the particle swarm is taken as the current global optimal solution. The process is continuously iterated, updating the particle's velocity and position. Finally, the optimal solution of the control variables that meets the termination condition is obtained, which, for each task scenario, is the optimal / second-best control quantity that minimizes host power consumption, chiller pump power consumption, cooling pump power consumption, cooling tower power consumption, and cooling water outlet temperature at the cooling tower end.

[0147] In step S602 of some embodiments, by adjusting the variable C in the candidate regulation prediction data until the objective function J satisfies the optimal solution, the control variable C in the corresponding task scenario is used as the target regulation prediction data.

[0148] Steps S601 to S602 as shown in the embodiments of this application involve inputting candidate control prediction data into the optimization model to obtain the objective function, adjusting the variables in the candidate control prediction data until the objective function meets the preset conditions, and using the corresponding candidate control prediction data as the target control prediction data to achieve the optimal solution control data for the total power consumption of the central air conditioning system in each task scenario.

[0149] Please see Figure 7 In some embodiments, step S107 may include, but is not limited to, steps S701 to S702:

[0150] Step S701: Obtain historical training operation data of the central air conditioning system;

[0151] Step S702: After filtering out abnormal data through data filtering preprocessing, the historical training data is used to obtain the initial selected historical data.

[0152] Step S703: Slice the initially selected historical data to obtain training input variable data and training target variable data;

[0153] Step S704: Construct a training dataset and a validation dataset based on the training input variable data and the training target variable data.

[0154] In step S701 of some embodiments, historical training operation data of the central air conditioning system can be obtained by acquiring the historical operation logs of the central air conditioning system. Alternatively, historical training operation data can be obtained through other third-party software, and is not limited to this.

[0155] In step S702 of some embodiments, after acquiring historical training data, the historical training data is preprocessed to obtain preliminary historical data. For example, the historical training data is cleaned to remove 0 values / abnormal samples collected due to sensor malfunctions or other abnormal reasons; some missing values ​​are filled using the previous training data. Other data preprocessing methods can also be used to preprocess the data, and are not limited to these.

[0156] In step S703 of some embodiments, the initial historical data is sliced, and the historical training data is divided into segments of input variable x and its corresponding target variable y.

[0157] In step S704 of some embodiments, a training dataset is finally constructed based on the input variable x and the target variable y, and the training dataset is standardized to reduce the influence of different units between variables.

[0158] Steps S701 to S704, as illustrated in this embodiment, involve acquiring historical training operation data of the central air conditioning system, filtering out abnormal data through data filtering preprocessing to obtain preliminary historical data, then slicing the preliminary historical data to obtain training input variable data and training target variable data, and finally constructing training datasets and validation datasets based on the training input variable data and training target variable data for subsequent model training.

[0159] Please see Figure 8 This application also provides a control device for a central air conditioning system, which can implement the above-mentioned control method for a central air conditioning system. The device includes:

[0160] The data acquisition module 801 is used to acquire historical operating data of the central air conditioning system and at least two task scenarios; wherein, different task scenarios focus on different preset performance indicators of the central air conditioning system.

[0161] The sample construction module 802 is used to construct a sample dataset for each of the task scenarios based on the preset performance indicators and the historical running data.

[0162] The sample input module 803 is used to input the sample dataset into the preset target system regulation and prediction model, wherein the target system regulation and prediction model includes: a gating network, a feature extraction network, and a regulation and prediction network;

[0163] The weight allocation module 804 is used to allocate weights to the sample dataset through a gating network to obtain the target weight information for each task scenario.

[0164] The feature extraction module 805 is used to perform feature mapping on the sample dataset through the feature extraction network and target weight information to obtain the target feature vector;

[0165] The regulation and prediction module 806 is used to regulate and predict the target weight information and target feature vector through the regulation and prediction network to obtain candidate regulation and prediction data.

[0166] The control optimization module 807 is used to optimize the candidate regulation prediction data according to the preset optimization model to obtain the target regulation prediction data.

[0167] The control module 808 is used to control the central air conditioning system based on the target control prediction data.

[0168] The specific implementation method of the control device of the central air conditioning system is basically the same as the specific implementation method of the control method of the central air conditioning system described above, and will not be repeated here.

[0169] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the control method of the central air conditioning system described above. This electronic device can be any smart terminal, including a tablet computer, an in-vehicle computer, or similar device.

[0170] Please see Figure 9 , Figure 9 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes:

[0171] The processor 901 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application.

[0172] The memory 902 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 902 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 902 and is called and executed by the processor 901 to execute the control method of the central air conditioning system of the embodiments of this application.

[0173] The input / output interface 903 is used to implement information input and output;

[0174] The communication interface 904 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0175] Bus 905 transmits information between various components of the device (e.g., processor 901, memory 902, input / output interface 903, and communication interface 904);

[0176] The processor 901, memory 902, input / output interface 903, and communication interface 904 are connected to each other within the device via bus 905.

[0177] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the control method of the central air conditioning system described above.

[0178] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0179] The control method, control device, electronic device, and storage medium for a central air conditioning system provided in this application embodiment acquire historical operating data and at least two task scenarios of the central air conditioning system. Different task scenarios focus on different preset performance indicators in the central air conditioning system. A sample dataset for each task scenario is constructed based on the preset performance indicators and historical operating data. The sample dataset is then input into a preset target system regulation prediction model. The target system regulation prediction model includes a gating network, a feature extraction network, and a regulation prediction network. The gating network assigns weights to the sample dataset to obtain target weight information for each task scenario. The feature extraction network includes a shared underlying feature extraction network. The system employs a feature extraction subnetwork and a task feature extraction subnetwork. By sharing the underlying feature extraction subnetwork, task feature extraction subnetwork, and target weight information, it performs feature mapping on the sample dataset to obtain target feature vectors that reflect both common features and unique features specific to each task scenario. Then, a regulation and prediction network is used to regulate and predict the target weight information and target feature vectors to obtain candidate regulation and prediction data that meet the basic power consumption and energy-saving requirements of practical applications. Finally, the candidate regulation and prediction data is optimized and processed according to a preset optimization model to obtain target regulation and prediction data. The central air conditioning system is then controlled based on the target regulation and prediction data to achieve the lowest possible total power consumption.

[0180] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0181] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0182] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0183] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0184] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0185] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0186] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0187] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0188] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0189] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0190] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A control method for a central air conditioning system, characterized in that, The method includes: Obtain historical operating data of the central air conditioning system and at least two task scenarios; wherein, different task scenarios focus on different preset performance indicators of the central air conditioning system; A sample dataset for each task scenario is constructed based on the preset performance indicators and the historical running data; The sample dataset is input into a preset target system regulation and prediction model; wherein, the target system regulation and prediction model includes: a gating network, a feature extraction network, and a regulation and prediction network; The target weight information for each task scenario is obtained by weighting the sample dataset through the gating network. The target feature vector is obtained by performing feature mapping on the sample dataset using the feature extraction network and the target weight information; The target weight information and the target feature vector are controlled and predicted by the control and prediction network to obtain control and prediction data. The control and prediction data are optimized according to a preset optimization model to obtain the target control data; The central air conditioning system is controlled based on the target control data; The feature extraction network includes: a shared underlying feature extraction sub-network, a task feature extraction sub-network for each task scenario, and a feature filtering sub-network. The step of performing feature mapping on the sample dataset using the feature extraction network and the target weight information to obtain the target feature vector includes: The shared underlying feature extraction subnetwork is used to extract shared features from the sample dataset to obtain a shared underlying feature vector. The shared underlying feature extraction subnetwork consists of three task feature extraction subnetworks, namely BiLSTM, CNN and Transformer. The task feature extraction sub-network is used to extract task features from the sample dataset to obtain the task-level sub-feature vector for each task scenario. Each task scenario has at least one task-level sub-feature vector that is unique to that task scenario. The shared underlying feature vector and the task underlying sub-feature vector are concatenated according to the target weight information to obtain the candidate feature vector; The candidate feature vectors are filtered through the feature filtering sub-network to obtain the target feature vector.

2. The method according to claim 1, characterized in that, Before inputting the sample dataset into the preset target system regulation prediction model, the method further includes: Constructing the target system regulation and prediction model specifically includes: Obtain the training dataset for each of the aforementioned task scenarios in the central air conditioning system; wherein, the training dataset includes: historical training dataset and validation dataset; The historical training dataset is input into a preset original system regulation and prediction model; wherein, the original system regulation and prediction model includes: a gating network, a feature extraction network, and regulation and prediction networks for each sub-network; The training dataset is weighted using the gating network to obtain training weight information for each task scenario; The historical training dataset is mapped using the feature extraction network to obtain training feature vectors; The training weight information and the training feature vector are adjusted and predicted by the prediction network through each sub-network to obtain the training adjustment and prediction data for each task scenario. Based on the training control prediction data for each task scenario, the validation dataset, and the training weight information, loss calculation is performed to obtain training loss data; The parameters of the preset original system regulation and prediction model are adjusted based on the training loss data to obtain the target system regulation and prediction model.

3. The method according to claim 1, characterized in that, The step of concatenating the shared underlying feature vector and the task underlying sub-feature vector according to the target weight information to obtain the candidate feature vector includes: Based on the target weight information, the weighted concatenation of all task-level sub-feature vectors for each task scenario is used to obtain the training-level feature vector for each task scenario. The candidate feature vector is obtained by concatenating the training low-level feature vector and the shared low-level feature vector.

4. The method according to claim 2, characterized in that, The step of calculating the training loss data based on the training adjustment prediction data for each task scenario, the validation dataset, and the training weight information includes: The training loss function for each task scenario is constructed based on the validation dataset and the training and prediction data for each task scenario. The training loss function for each task scenario is weighted and summed based on the training weight information to obtain the total training loss function. The training loss data is obtained by converging the total training loss function.

5. The method according to claim 1, characterized in that, The target control prediction data includes state variable information; the optimization process of the candidate control prediction data according to the preset optimization model to obtain the target control prediction data specifically includes: The candidate regulation prediction data are input into the optimization model to obtain the objective function; By adjusting the variables within the candidate regulation prediction data until the objective function meets the preset conditions, the corresponding candidate regulation prediction data is used as the target regulation prediction data.

6. The method according to claim 2, characterized in that, The acquisition of the training dataset for the central air conditioning system includes: Obtain the historical training operation data of the central air conditioning system; After filtering out abnormal data through data filtering preprocessing, the historical training data is used to obtain the initial selected historical data. The initial historical data is sliced ​​to obtain training input variable data and training target variable data; The training dataset and the validation dataset are constructed based on the training input variable data and the training target variable data.

7. A control device for a central air conditioning system, characterized in that, The device includes: The data acquisition module is used to acquire historical operating data of the central air conditioning system and at least two task scenarios; wherein, different task scenarios focus on different preset performance indicators of the central air conditioning system. The sample construction module is used to construct a sample dataset for each of the task scenarios based on the preset performance indicators and the historical running data. The sample input module is used to input the sample dataset into a preset target system regulation and prediction model; wherein, the target system regulation and prediction model includes: a gating network, a feature extraction network, and a regulation and prediction network; The weight allocation module is used to allocate weights to the sample dataset through the gating network to obtain target weight information for each task scenario; The feature extraction module is used to perform feature mapping on the sample dataset through the feature extraction network and the target weight information to obtain the target feature vector; The regulation prediction module is used to perform regulation prediction on the target weight information and the target feature vector through the regulation prediction network to obtain candidate regulation prediction data. The control optimization module is used to optimize the candidate regulation prediction data according to a preset optimization model to obtain the target regulation prediction data. The control module is used to control the central air conditioning system according to the target regulation prediction data; The feature extraction network includes: a shared underlying feature extraction sub-network, a task feature extraction sub-network for each task scenario, and a feature filtering sub-network. The step of performing feature mapping on the sample dataset using the feature extraction network and the target weight information to obtain the target feature vector includes: The shared underlying feature extraction subnetwork is used to extract shared features from the sample dataset to obtain a shared underlying feature vector. The shared underlying feature extraction subnetwork consists of three task feature extraction subnetworks, namely BiLSTM, CNN and Transformer. The task feature extraction sub-network is used to extract task features from the sample dataset to obtain the task-level sub-feature vector for each task scenario. Each task scenario has at least one task-level sub-feature vector that is unique to that task scenario. The shared underlying feature vector and the task underlying sub-feature vector are concatenated according to the target weight information to obtain the candidate feature vector; The candidate feature vectors are filtered through the feature filtering sub-network to obtain the target feature vector.

8. An electronic device, characterized in that, The electronic device includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the control method of the central air conditioning system according to any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the control method of the central air conditioning system according to any one of claims 1 to 6.

Images