Method for predicting load of air conditioner in office building and storage medium

Abstract

The invention provides an office building air conditioner load prediction method and a storage medium, a preheating method comprises the following steps: firstly associating air conditioner operation parameter information, weather information and holiday and festival information according to date information to form air conditioner operation sample data; preprocessing the air conditioner operation sample data to obtain an air conditioner operation preprocessing sample data set, and dividing the air conditioner operation preprocessing sample data set into a test set and a training set; based on the training set, constructing an air conditioner load prediction model based on a Transform neural network, and performing hyper-parameter optimization on the air conditioner load prediction model to obtain an air conditioner load optimization prediction model; inputting air conditioner operation parameters and weather information which are obtained in real time into the trained air conditioner load optimization prediction model to obtain a prediction result of the air conditioner load; the method can accurately and reliably analyze the air conditioner load prediction result in the commercial building operation stage.

Description

Technical Field

[0001] The present invention relates to the field of building energy intelligent technology, and in particular to an air conditioning load prediction method and storage medium in an office building based on an improved transformer. Background Art

[0002] In the current actual situation of many buildings, property management personnel lack awareness of air conditioning. They are unable to make advance predictions about changes in the external environment and are unable to make adjustments to the air conditioning in advance, which leads to a waste of air conditioning resources.

[0003] Intelligent air conditioning control technology is an important part of building energy conservation and reducing the cost of building air conditioning. Among them, air conditioning load prediction is a key pre-strategy for intelligent control. Only by accurately predicting the future air conditioning load of the building can the air conditioning control be effectively implemented. Accurate prediction of building load is the basis for realizing building energy-saving design and operation, and is also the basis for equipment optimization operation and feedforward control. Accurate load prediction can not only help building managers formulate energy use plans and quota plans, but also assist operation and maintenance personnel to grasp the use of various energy sources in the building and promptly discover abnormal energy use and energy waste.

[0004] A Chinese patent with patent number ZL 202110053199.4 discloses a deep transfer learning energy consumption prediction method and system for centrally air-conditioned residences, including a data collection module, a data preprocessing module, a model building module, and a prediction result output module. However, the prediction object of this patent is based on residential buildings, and whether to turn on the air conditioner mostly depends on the user's own wishes. The air conditioners in commercial office buildings are mostly turned on and managed by the property. The prediction method provided by this patent cannot be directly applied to the air conditioning load prediction of commercial office buildings. In addition, in this patent, the sample data is relatively small, which means that the number of selected features is small and the reliability of its prediction results is low.

[0005] Therefore, there is an urgent need for an air conditioning load prediction method and storage medium in office buildings, which can accurately and reliably analyze the air conditioning load prediction results during the operation stage of commercial buildings. Summary of the invention

[0006] The purpose of the present invention is to provide a method and storage medium for predicting air conditioning load in an office building, aiming to solve the technical problems of waste of air conditioning data in commercial office buildings and low prediction reliability of traditional prediction methods.

[0007] To achieve the above object, in a first aspect, the present invention provides a method for predicting air conditioning load in an office building, the steps of which include:

[0008] S1. Collect air conditioning operation sample data:

[0009] The sample data includes multiple sets of air conditioning operation parameter information, weather information in a future preset time period of the area where the current office building is located, and holiday information in a future preset time period and historically omitted; and the air conditioning operation parameter information, weather information and holiday information are associated according to date information to form the air conditioning operation sample data;

[0010] S2, preprocessing the air-conditioning operation sample data to obtain an air-conditioning operation preprocessed sample data set, and dividing the data set into a test set and a training set;

[0011] S3. Based on the obtained training set, an air conditioning load prediction model based on a Transformer neural network is constructed, and hyperparameter optimization is performed on the air conditioning load prediction model to obtain an air conditioning load optimization prediction model, and a model evaluation is performed on the air conditioning load optimization prediction model;

[0012] S4. Inputting the air-conditioning operating parameters and weather information acquired in real time into the trained air-conditioning load optimization prediction model to obtain the prediction result of the air-conditioning load.

[0013] As a further improvement of the above scheme, in step S3, when constructing the air conditioning load prediction model based on the Transformer neural network,

[0014] The air conditioning load prediction model is constructed based on the DNN deep neural network and the Transformer neural network model to obtain the air conditioning load prediction model based on the Dnn-Transformer neural network;

[0015] The air conditioning load prediction model based on the Dnn-Transformer neural network includes a DNN layer, a Transformer layer and a fully connected layer.

[0016] As a further improvement of the above solution, the DNN layer includes a first DNN module and a second DNN module;

[0017] The first DNN module is used to receive and process weather information and air conditioning operation parameter information;

[0018] The second DNN module is used to receive the n-dimensional vector of the time series after the time series passes through the Transformer, and is also used to receive the m-dimensional vector output by the first DNN module;

[0019] Preferably, the n-dimensional vector and the m-dimensional vector are concatenated into tensors through a concat function to form (n+m)-dimensional vector data.

[0020] As a further improvement of the above solution, the Transformer layer includes a position encoding layer, an encoder layer and a decoder layer;

[0021] The encoder layer includes a multi-head attention module, a layer normalization module, a feedforward network module, a residual link and normalization module;

[0022] The decoder layer includes a masked multi-head attention module, a feedforward network module, a residual link and a normalization module.

[0023] As a further improvement of the above scheme, in step S3, the hyperparameters of the air-conditioning load prediction model are optimized by methods including but not limited to genetic optimization algorithm and particle swarm optimization algorithm; preferably, the hyperparameters of the air-conditioning load prediction model are optimized by genetic optimization algorithm.

[0024] As a further improvement of the above solution, in step S1, multiple groups of air-conditioning operation parameter information are collected and obtained through deployed sensors, and the air-conditioning operation parameter information is transmitted through the Internet of Things data transmission protocol.

[0025] As a further improvement of the above scheme, the air conditioning operation parameter information includes but is not limited to the temperature difference between supply and return water, chilled water inlet temperature, chilled water outlet temperature, chilled water inlet pressure, chilled water outlet pressure, freezing pump inlet pressure, freezing pump outlet pressure and cooling tower outlet temperature.

[0026] As a further improvement of the above scheme, in step S1, the weather information is the climate-related characteristics of the day; preferably, the climate-related characteristics include but are not limited to temperature, humidity, wind direction and wind speed.

[0027] As a further improvement of the above solution, in step S1, when obtaining holiday information within a future preset time period and historically omitted holiday information, holiday-related date information is collected by web crawlers and / or manual sorting.

[0028] As a further improvement of the above scheme, in step S2, when the air conditioner operation sample data is preprocessed, it includes performing outlier processing, normalization processing, text processing, and labeling processing on the air conditioner operation sample data;

[0029] The normalization process adopts Min-max normalization process, and the calculation formula of Min-max standardization is as follows:

[0030]

[0031] In the formula, max(x) is the maximum value of the data, and min(x) is the minimum value of the data.

[0032] As a further improvement of the above scheme, in step S2, when preprocessing the air-conditioning operation sample data, for the processing of holiday information, firstly a holiday database is constructed to determine whether the current time is a holiday based on the specific corresponding date and time of the collected data.

[0033] As a further improvement of the above scheme, in step S2, when preprocessing the air-conditioning operation sample data and processing the historical data, deduplication, outlier replacement, and data type conversion are uniformly performed to make the historical data more consistent with the actual input data style of the algorithm and easier to fit.

[0034] As a further improvement of the above scheme, the specific expression of the internal structure process of the Transformer neural network model is:

[0035]

[0036] z=x+f(x)

[0037]

[0038]

[0039] FFN(x)=max(0,xW1+b1)W2+b2

[0040] In the formula, pos represents the position of the variable, i represents the i-th dimension, and d model represents the word embedding dimension; x is the input tensor value, f(x) is the value after function transformation, z represents the residual layer; Q, K, and V are the input weight matrices, It is a scaling process to prevent the value from being too large; y is the value after hierarchical normalization of the input x, γ and β parameters are both learnable parameters, mean represents the mean, and var represents the variance; FFN(x) is the feature capture function, W1, W2, b1, b2 are learnable weight matrices and bias vectors.

[0041] In a second aspect, the present invention further provides a device, comprising a memory for storing computer program instructions and a processor for executing computer program instructions, wherein when the computer program instructions are executed by the processor, the device is triggered to execute a method for predicting air conditioning load in an office building provided in the first aspect.

[0042] In a third aspect, the present invention further provides a storage medium having a computer program stored thereon, wherein when the program is executed, the device where the storage medium is located is controlled to execute a method for predicting air conditioning load in an office building provided in the first aspect.

[0043] Since the present invention adopts the above technical solution, the beneficial effects of this application are:

[0044] The present invention provides a method for predicting air conditioning load in an office building. First, multiple sets of air conditioning operation parameter information, weather information within a preset future time period in the area where the current office building is located, and holiday information within the preset future time period and historically omitted holiday information are collected, and the air conditioning operation parameter information, weather information and holiday information are associated according to date information to form the air conditioning operation sample data. With such a setting, not only the air conditioning operation parameter information but also the weather information of the area and the preset future time period and historically omitted holiday information are considered in the air conditioning operation sample data, and these data information are sorted and associated according to date. By mining time series data, the time series data can be turned into a sequence, which is convenient for complete input into the Transformer network structure, thereby making better use of the characteristics of the time series data. In addition, when constructing the basic When the air conditioning load prediction model of the Transformer neural network is used, the air conditioning load prediction model is built based on the DNN deep neural network and the Transformer neural network model to obtain the air conditioning load prediction model based on the Dnn-Transformer neural network; the air conditioning load prediction model based on the Dnn-Transformer neural network includes a DNN layer, a Transformer layer and a fully connected layer; the present invention adopts a hybrid structure model composed of Dnn-Transformer to predict the air conditioning load, and the specific DNN layer mainly processes discrete features and weather-related features, so that the weights and information of these features can be maximized; so that the air conditioning load prediction model constructed by the present invention has higher prediction accuracy and reliability than that of only considering a single load sequence;

[0045] In some preferred embodiments, Internet of Things devices are used to collect air conditioning operating parameter information and weather information; thereby ensuring the accuracy and real-time nature of sample data. In addition, when collecting historical data, a large number of phase characteristics are manually collected, so that more useful information can be mined, making the prediction method provided by the present invention more effective. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying any creative work.

[0047] Figure 1A schematic diagram of a data collection process of a commercial building involved in a method for predicting air conditioning load in an office building disclosed in the present invention;

[0048] Figure 2 A schematic diagram of correlation analysis of relevant parameters collected from commercial buildings involved in a method for predicting air conditioning load in an office building disclosed in the present invention;

[0049] Figure 3 It is a schematic diagram of the overall process of an air conditioning load prediction method in an office building disclosed in the present invention;

[0050] Figure 4 The present invention discloses a method for predicting air conditioning load in an office building based on a specific structural diagram of a Dnn-Transformer algorithm model.

[0051] The purpose, features and advantages of the present invention will be further described in conjunction with the embodiments with reference to the accompanying drawings. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] Furthermore, the technical solutions between the various embodiments of the present invention may be combined with each other, but this must be based on the fact that they can be implemented by ordinary technicians in the field. When the combination of technical solutions is mutually contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0054] Embodiment 1:

[0055] See also Figure 1-4 The present invention provides a method for predicting air conditioning load in an office building, the steps of which include:

[0056] S1. Collect air conditioning operation sample data:

[0057] The sample data includes multiple sets of air conditioning operation parameter information, weather information in a future preset time period of the area where the current office building is located, and holiday information in a future preset time period and historically omitted; and the air conditioning operation parameter information, weather information and holiday information are associated according to date information to form an air conditioning operation sample data set;

[0058] For details, see Figure 1, multiple sets of air-conditioning operation parameter information and weather information are collected by sensors deployed on the air-conditioning equipment and sensors deployed in the operating environment of the air-conditioning equipment; the air-conditioning operation parameter information and weather information obtained by the sensors are transmitted through the Internet of Things data transmission protocol MQTT; at the same time, through the message parsing method, the relevant data information in each transmission interface is obtained and stored in the database;

[0059] The air conditioning operation parameter information includes but is not limited to the temperature difference between supply and return water, chilled water inlet temperature, chilled water outlet temperature, chilled water inlet pressure, chilled water outlet pressure, chilled pump inlet pressure, chilled pump outlet pressure and cooling tower outlet temperature;

[0060] The weather information is the climate-related characteristics of the day; the climate-related characteristics include but are not limited to temperature, humidity, wind direction and wind speed; specifically, in this embodiment, the weather API interface is obtained to mine the climate-related characteristics of the area where the current building is located in the next 24 hours;

[0061] When obtaining holiday information within a preset time period in the future and missed in history, collect holiday-related date information through web crawlers and / or manual sorting;

[0062] S2, preprocessing the air-conditioning operation sample data to obtain an air-conditioning operation preprocessed sample data set, and dividing the data set into a test set and a training set;

[0063] Specifically, when the air conditioner operation sample data is preprocessed, it includes performing outlier processing, normalization processing, text processing, and labeling processing on the air conditioner operation sample data;

[0064] The normalization process adopts Min-max normalization process, and the calculation formula of Min-max standardization is as follows:

[0065]

[0066] Among them, max(x) is the maximum value of the data, and min(x) is the minimum value of the data;

[0067] When preprocessing the air conditioner operation sample data, for the processing of holiday information, firstly, a holiday database is constructed, and the specific corresponding date and time of the collected data is used to determine whether it is a holiday;

[0068] When processing historical data, duplicate removal, outlier replacement, and data type conversion are uniformly performed to make the historical data more consistent with the actual input data style of the algorithm and easier to fit;

[0069] In this embodiment, the air conditioner operation preprocessing sample data set is divided into 80% as a training set and 20% as a test set, and the data are ensured to be two days apart at the specific segmentation position;

[0070] S3. Based on the obtained training set, an air conditioning load prediction model based on a Transformer neural network is constructed, and hyperparameter optimization is performed on the air conditioning load prediction model to obtain an air conditioning load optimization prediction model, and a model evaluation is performed on the air conditioning load optimization prediction model;

[0071] When building an air conditioning load prediction model based on Transformer neural network,

[0072] The air conditioning load prediction model is constructed based on the DNN deep neural network and the Transformer neural network model to obtain the air conditioning load prediction model based on the Dnn-Transformer neural network;

[0073] The air conditioning load prediction model based on the Dnn-Transformer neural network includes a DNN layer, a Transformer layer and a fully connected layer;

[0074] S4, inputting the air conditioning operation parameters and weather information obtained in real time into the trained air conditioning load optimization prediction model to obtain the prediction result of the air conditioning load;

[0075] The present invention first collects multiple sets of air-conditioning operation parameter information, weather information within a preset future time period of the area where the current office building is located, and holiday information within a preset future time period and historically omitted holiday information, and associates the air-conditioning operation parameter information, weather information and holiday information according to date information to form the air-conditioning operation sample data; such a setting not only considers the air-conditioning operation parameter information, but also considers the weather information of the area and the preset future time period and historically omitted holiday information in the air-conditioning operation sample data, and at the same time sorts and associates these data information according to date; through time series data mining, the time series data can be turned into a sequence, which is convenient for complete input into the Transformer network structure, thereby better utilizing the characteristics of the time series data; in addition, in constructing a Transformer-based When constructing an air conditioning load prediction model based on a mer neural network, the air conditioning load prediction model is constructed based on a DNN deep neural network and a Transformer neural network model to obtain an air conditioning load prediction model based on a Dnn-Transformer neural network; the air conditioning load prediction model based on a Dnn-Transformer neural network includes a DNN layer, a Transformer layer and a fully connected layer; the present invention uses a hybrid structure model composed of a Dnn-Transformer combination to predict the air conditioning load, and the specific DNN layer mainly processes discrete features and weather-related features, so that the weights and information of these features can be maximized; the air conditioning load prediction model constructed by the present invention has higher prediction accuracy and reliability than that of a single load sequence only;

[0076] In certain preferred embodiments, Internet of Things devices are used to collect air conditioning operating parameter information and weather information; thereby ensuring the accuracy and real-time nature of the sample data. In addition, when collecting historical data, holiday-related date information is also collected through web crawlers and / or manual sorting, thereby being able to mine useful information with more changes, making the prediction effect of the prediction method provided by the present invention better.

[0077] As a preferred embodiment, in step S1, after the various collected original raw data are parsed and stored in the database, the original data also needs to be abnormalized and normalized. Specifically, the data is deduplicated and de-empty using the relevant interface in Python, and the return water temperature, supply water temperature and other characteristics are normalized using the data normalization method to eliminate the influence of large variable differences.

[0078] After the raw data is anomalyized and normalized, it is also necessary to transform the features of the raw data, such as encoding the weather, wind direction, wind speed and other features to make them conform to the parameter format of the algorithm input. On this basis, the weather and climate related features are combined through feature engineering and other methods to form more effective feature data.

[0079] After the various collected raw data are transformed into features, they are also spliced. Specifically, the load forecast values ​​of related equipment are recorded through sensors, and the data transmitted by sensors are analyzed in the background. The corresponding data are stored in the warehouse, and the load data of the last 48 hours are spliced. The data is processed in segments every half an hour, and finally a load sequence data with a length of 96 is formed; the air-conditioning operation sample data set is preliminarily constructed, and the correlation analysis of various parameters in the constructed air-conditioning operation sample data set is also carried out. The specific analysis results are as follows: Figure 2 As shown; in combination with the results of the correlation analysis, parameters with greater correlation are selected as the final air conditioning operation sample data set.

[0080] As a preferred embodiment, the DNN layer includes a first DNN module and a second DNN module;

[0081] The first DNN module is used to receive and process weather information and air conditioning operation parameter information;

[0082] The second DNN module is used to receive the n-dimensional vector of the time series after the time series passes through the Transformer, and is also used to receive the m-dimensional vector output by the first DNN module;

[0083] Preferably, the n-dimensional vector and the m-dimensional vector are concatenated into tensors through a concat function to form (n+m)-dimensional vector data;

[0084] For details, see Figure 4 In this embodiment, relevant information such as climate, temperature, holidays, etc. is taken and input into the DNN layer, the number of layers of the DNN layer is set to 2, the number of neural network units is set to 20, and the output data dimension is 32 dimensions;

[0085] The first DNN module uses the parameters obtained from the input weather, air conditioning, and sensors as the input layer of the DNN; the specific formula of the middle hidden layer is: f(x)=w*x+b, where x is the input data, w is the hidden layer weight, and b is the bias;

[0086] The second DNN module is a one-layer DNN structure. The input is divided into two: one is the 16-dimensional vector of the time series after the transformer, and the other is the 32-dimensional vector input by the first DNN layer. Finally, the tensors are concatenated through the concat function to form the input 48-dimensional vector data.

[0087] As a preferred embodiment, the Transformer layer includes a position encoding layer, an encoder layer and a decoder layer;

[0088] The encoder layer includes a multi-head attention module, a layer normalization module, a feedforward network module, a residual link and normalization module;

[0089] The decoder layer includes a masked multi-head attention module, a feedforward network module, a residual link and a normalization module;

[0090] Specifically, the internal structure process of the Transformer neural network model is specifically expressed as follows:

[0091]

[0092] z=x+f(x)

[0093]

[0094]

[0095] FFN(x)=max(0,xW1+b1)W2+b2

[0096] In the formula, pos represents the position of the variable, i represents the i-th dimension, and d model represents the word embedding dimension; x is the input tensor value, f(x) is the value after function transformation, z represents the residual layer; Q, K, and V are the input weight matrices, It is a scaling process to prevent the value from being too large; y is the value after hierarchical normalization of the input x, γ and β parameters are both learnable parameters, mean represents the mean value, and var represents the variance; FFN(x) is the feature capture function, W1, W2, b1, b2 are learnable weight matrices and bias vectors;

[0097] The feature capture function FFN(x) introduces necessary nonlinearity into the model through a combination of two linear transformations and a nonlinear activation function, which enables the model to learn more complex and abstract data representations; it enables the Transformer neural network model to capture more complex features and improve the model's expressiveness;

[0098] Specifically, a layer of encoder and a layer of decoder are used to design the time series data algorithm, Q is used as the input time series sequence, K is used as the input time parameter sequence, and V is also the input time series.

[0099] As a preferred embodiment, in step S3, the hyperparameters of the air conditioning load prediction model are optimized by methods including but not limited to genetic optimization algorithm and particle swarm optimization algorithm; preferably, the hyperparameters of the air conditioning load prediction model are optimized by genetic optimization algorithm.

[0100] As a preferred embodiment, in step S3, the air conditioning load optimization prediction model is evaluated. Specifically, the Dnn-Transformer neural network is trained, and the predicted values ​​in the Dnn-Transformer neural network are compared with the actual values ​​in the test set. RMSE (root mean square error) and MAE (mean absolute error) are used as verification criteria. The relevant calculation formulas are as follows:

[0101]

[0102] Where: θ is the true value of the parameter, n is the number of data sets, is the model prediction value, y i is the actual value.

[0103] As a preferred embodiment, in step S4, the trained air conditioning load optimization prediction model is deployed to an online application, and a prediction interface and service are provided for the front-end to call. Specifically, the weather data corresponding to the hour currently required for prediction is obtained from the database through the weather data warehouse, the characteristic information of the relevant equipment is obtained through the database corresponding to the sensor device, and the load data based on statistics and the load series data based on the time series are obtained through the historical load. Specifically, as an optional air conditioning energy consumption optimization method, if the air conditioning load prediction result is too high and exceeds the budgeted energy consumption value, it is necessary to properly control the use of the air conditioner. If the air conditioning energy consumption prediction result is below or equal to the budgeted energy consumption value, the use of the air conditioner can maintain the status quo or further reduce the use of the air conditioner to achieve maximum energy saving. The specific air conditioning start and stop plan can be formulated by professional and technical personnel in this field according to specific needs.

[0104] Embodiment 2:

[0105] The present invention also provides a device, comprising a memory for storing computer program instructions and a processor for executing the computer program instructions, wherein when the computer program instructions are executed by the processor, the device is triggered to execute some or all of the steps in Embodiment 1;

[0106] The processor may include one or more processing units, for example, the processor may include an application processor (AP), a modem processor, a graphics processor (GPU), an image signal processor (ISP), a controller, a video codec, a digital signal processor (DSP), a baseband processor, and / or a neural-network processing unit (NPU), etc. Different processing units may be independent devices or integrated into one or more processors.

[0107] The controller can be the nerve center and command center of the electronic device. The controller can generate operation control signals according to the instruction operation code and timing signal to complete the control of fetching and executing instructions.

[0108] A memory may also be provided in the processor for storing instructions and data. In some embodiments, the memory in the processor is a cache memory. The memory may store instructions or data that the processor has just used or is cyclically used. If the processor needs to use the instruction or data again, it may be directly called from the memory. This avoids repeated access, reduces the waiting time of the processor, and thus improves the efficiency of the system.

[0109] Embodiment 3:

[0110] The present invention also provides a storage medium having a computer program stored thereon, wherein when the program is executed, the device where the storage medium is located is controlled to execute part or all of the steps in Example 1.

[0111] The storage medium may include a high-speed RAM memory, and may also include a nonvolatile memory (nonvolatile memory), such as at least one disk memory. It is understandable that the storage medium can be a random access memory (Random Access Memory, RAM), a disk, a hard disk, a solid state disk (solid state disk, SSD) or a nonvolatile memory, etc., which can store program codes and other machine-readable media.

[0112] Those skilled in the art will appreciate that the embodiments of the present invention may be provided as methods or storage media. Therefore, the embodiments of the present invention may take the form of complete hardware embodiments, complete software embodiments, or embodiments combining software and hardware. Moreover, the embodiments of the present invention may take the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0113] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. All equivalent structural changes made using the contents of the present invention's specification and drawings, or directly / indirectly applied in other related technical fields, are included in the patent protection scope of the present invention.

Claims

1. A method for predicting air conditioning load in office buildings. It is characterized in that The steps include: S1. Collect air conditioning operation sample data: The sample data includes multiple sets of air conditioning operation parameter information, weather information in a future preset time period of the area where the current office building is located, and holiday information in a future preset time period and historically omitted; and the air conditioning operation parameter information, weather information and holiday information are associated according to date information to form the air conditioning operation sample data; S2, preprocessing the air-conditioning operation sample data to obtain an air-conditioning operation preprocessed sample data set, and dividing the data set into a test set and a training set; S3. Based on the obtained training set, an air conditioning load prediction model based on a Transformer neural network is constructed, and hyperparameter optimization is performed on the air conditioning load prediction model to obtain an air conditioning load optimization prediction model, and a model evaluation is performed on the air conditioning load optimization prediction model; S4. Inputting the air-conditioning operating parameters and weather information acquired in real time into the trained air-conditioning load optimization prediction model to obtain the prediction result of the air-conditioning load.

2. A method for predicting air conditioning load in an office building according to claim 1, It is characterized in that In step S3, when constructing the air conditioning load prediction model based on the Transformer neural network, The air conditioning load prediction model is constructed based on the DNN deep neural network and the Transformer neural network model to obtain the air conditioning load prediction model based on the Dnn-Transformer neural network; The air conditioning load prediction model based on the Dnn-Transformer neural network includes a DNN layer, a Transformer layer and a fully connected layer.

3. A method for predicting air conditioning load in an office building according to claim 2, It is characterized in that The DNN layer includes a first DNN module and a second DNN module; The first DNN module is used to receive and process weather information and air conditioning operation parameter information; The second DNN module is used to receive the n-dimensional vector of the time series after passing through the Transformer, and is also used to receive the m-dimensional vector output by the first DNN module.

4. A method for predicting air conditioning load in an office building according to claim 2 or 3, It is characterized in that The Transformer layer includes a position encoding layer, an encoder layer and a decoder layer; The encoder layer includes a multi-head attention module, a layer normalization module, a feedforward network module, a residual link and normalization module; The decoder layer includes a masked multi-head attention module, a feedforward network module, a residual link and a normalization module.

5. A method for predicting air conditioning load in an office building according to any one of claims 1 to 3, It is characterized in that In step S1, multiple groups of air-conditioning operation parameter information are collected and acquired through deployed sensors, and the air-conditioning operation parameter information is transmitted through the Internet of Things data transmission protocol.

6. A method for predicting air conditioning load in an office building according to any one of claims 1 to 3, It is characterized in that The air conditioning operation parameter information includes the temperature difference between supply and return water, chilled water inlet temperature, chilled water outlet temperature, chilled water inlet pressure, chilled water outlet pressure, chilled pump inlet pressure, chilled pump outlet pressure and cooling tower outlet temperature.

7. A method for predicting air conditioning load in an office building according to any one of claims 1 to 3, It is characterized in that In step S1, the weather information is the climate-related characteristics of the day, and the climate-related characteristics include temperature, humidity, wind direction and wind speed.

8. A method for predicting air conditioning load in an office building according to any one of claims 1 to 3, It is characterized in that In step S2, preprocessing the air conditioner operation sample data includes performing outlier processing, normalization processing, text processing, and labeling processing on the air conditioner operation sample data; When preprocessing the air conditioner operation sample data, for the processing of holiday information, firstly, a holiday database is constructed, and the specific corresponding date and time of the collected data is used to determine whether it is a holiday; When preprocessing the air conditioner operation sample data, when processing historical data, duplicate removal, abnormal value replacement, and data type conversion are uniformly performed.

9. A method for predicting air conditioning load in an office building according to any one of claims 1 to 3, It is characterized in that The specific expression of the internal structure process of the Transformer neural network model is: z=x+f(x) FFN(x)=max(0,xW 1 +b 1 )W 2 +b 2 In the formula, pos represents the position of the variable, i represents the i-th dimension, and d model represents the word embedding dimension; x is the input tensor value, f(x) is the value after function transformation, z represents the residual layer; Q, K, and V are the input weight matrices, is to prevent the value from being too large. y is the value after hierarchical normalization of input x. Both γ and β parameters are learnable parameters. mean represents the mean value and var represents the variance. FFN(x) is the feature capture function. W 1 , W 2 、b 1 、b 2 are the learnable weight matrices and bias vectors.

10. A storage medium, It is characterized in that A computer program is stored thereon, wherein when the program is running, the device where the storage medium is located is controlled to execute the method for predicting air conditioning load in an office building as described in any one of claims 1-9.

Images