An air conditioner carbon reduction control method, device, equipment and storage medium

Abstract

The present application belongs to the technical field of air conditioner carbon reduction control, and discloses an air conditioner carbon reduction control method, device, equipment and storage medium. The encoder-decoder based on the present application can effectively analyze the connection between regions and accurately predict the building hourly carbon emission. The outdoor, indoor hourly temperature and humidity, holiday, time, partition carbon emission and hourly wind volume information of each partition are used as input features. The historical information of the above features is used to predict the carbon emission intensity of each partition in the next hour. The encoder-decoder model in this study can accurately capture the mutual relationship between partitions and achieve high-precision carbon emission prediction. This method can be used as an auxiliary means to quickly locate the abnormal area or period of carbon emission. At the same time, the prediction result can be compared with the historical same period or reference value, solving the problem of difficult detection of small area carbon emission anomaly.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of air conditioner carbon reduction control, and particularly relates to an air conditioner carbon reduction control method, device, equipment and storage medium. BACKGROUND

[0002] The building industry accounts for about one-third of global carbon emissions, and excessive carbon emissions can cause a series of ecological problems. In view of the low degree of digitization and extensive management of the traditional building industry, the application divides the target floor into three zones of west, south and east, and provides an encoder-decoder model structure integrating LSTM, GRU, SimpleRNN, CNN and attention algorithm, which is used for real-time prediction of building operation carbon emissions. At the same time, the online optimization control of air conditioning equipment is carried out. The current research focuses on the carbon reduction optimization of the whole building, mainly starting from a single building and a single floor, and has not carried out carbon emission prediction from the perspective of building zoning and considering the mutual relationship of each zone. At the same time, the current air conditioning terminal air volume control is subjective, and there are few relevant basis, and dynamic control has not been realized.

[0003] Through the above analysis, the problems and defects of the prior art are:

[0004] (1) Previous research focuses on the overall carbon emission of the building, and cannot predict the hourly carbon emission of the high-value and high-energy density area in the building. It is difficult to identify abnormal carbon emission period / area from the overall carbon emission of the building, and the building carbon reduction research is not fine enough.

[0005] (2) In the traditional carbon emission time series prediction research, the information such as the correlation of carbon emissions at different times and the spatial characteristics of carbon emissions in different areas cannot be considered at the same time, resulting in low carbon emission prediction accuracy and inability to be used for actual building carbon reduction decision-making assistance.

[0006] (3) The traditional air conditioning terminal air volume control is subjective, and cannot control the air volume according to the actual needs, resulting in long-term uncomfortable environment and resource waste in the local area. SUMMARY

[0007] In view of the problems existing in the prior art, the application provides an air conditioner carbon reduction control method, device, equipment and storage medium.

[0008] The application is implemented as follows: an air conditioner carbon reduction control method comprises:

[0009] Step 1, building parameter acquisition;

[0010] Step 2, constructing an encoder-decoder time series prediction model for time series prediction;

[0011] Step three, online optimization of air conditioning terminal air volume control, to achieve building energy saving and carbon reduction.

[0012] Further, the building parameter acquisition method is as follows:

[0013] The target floor air conditioning terminal hourly energy consumption data in cooling season, outdoor hourly environmental parameters, indoor hourly temperature and humidity, time, floor holiday calendar, and air conditioning terminal hourly air volume information are acquired.

[0014] The collected data is preprocessed, specifically as follows:

[0015]

[0016] Formula (1), t l is the time to be interpolated, t l-1 and t l+1 are the previous time and the next time of time t l ; y l-1 , y l and y l+1 are the carbon emissions corresponding to time t l-1 , t l and t l+1 .

[0017] CE elec = E elec × EF elec (2)

[0018] E elec and EF elec are the power consumption and carbon emission factor (EF elec = 0.5703), and CE elec is the carbon emission corresponding to the power consumption.

[0019] Further, the encoder-decoder time series prediction model is constructed for time series prediction:

[0020] (1) The encoder module is composed of SimpleRNN, GRU and CNN;

[0021] (2) The RNN model structure is as follows:

[0022] h t = σ(wx t + Uh t-1 + b) (1)

[0023] In the formula, x t and h t correspond to the input variable and hidden state at time t, w and U are the matrices corresponding to the current unit and input variable, and b is the bias of the neuron.

[0024] (3) The GRU model structure is as follows:

[0025] z t = σ(x t w xr +h t-1 w hr +b r )(2)

[0026] r t = σ(x t w xz +h t-1 w hz +b z )(3)

[0027]

[0028] In the formula, σ and tanh are sigmoid function and activation function respectively, w xr and w hr are update gate weights, w xz and w hz are reset gate weights, w xh and w hh are candidate hidden layer states; x t is the input at time t, h t-1 is the hidden layer t-1 matrix, r t is the reset gate, z t is the update gate, b r , b z and b h are bias terms;

[0029] (4) Bi-GRU hybrid structure and attention mechanism;

[0030]

[0031] In the formula, and are the hidden layers of the forward GRU and the reverse GRU respectively, F is the combination mode of the bidirectional GRU;

[0032] (5) The calculation method of the attention mechanism is as follows:

[0033] e t = u tanh(w t +b)(7)

[0034]

[0035] In the formula, e t is a probability distribution, u and w are weight matrices, b and st are bias term and attention mechanism output, respectively; t is distribution coefficient.

[0036] (6) coefficient of determination R 2 , root mean square error RMSE and mean absolute error MAE.

[0037]

[0038]

[0039] wherein, y i and represent predicted value, observed value and predicted mean value, and n is sample number.

[0040] Further, the online optimization air conditioner terminal air volume control realizes building energy saving and carbon reduction:

[0041] (1) target area comfort calculation;

[0042] According to the hourly temperature and humidity information detected by the indoor sensor, the hourly PMV value can be calculated.

[0043]

[0044] wherein, PMV is comfort value, and are constant coefficients, and T and RH are temperature and relative humidity, respectively.

[0045] (2) accurate prediction of carbon emission considering subarea connection;

[0046] (3) real-time online optimization of air conditioner terminal air volume; multi-objective optimization.

[0047] Further, the accurate prediction of carbon emission considering subarea connection; the data collected in the early stage is directly input to the model after pretreatment, and the carbon emission intensity in the next hour is predicted through the sliding window method; at the same time, the constructed proxy model is used to predict the hourly PMV value.

[0048] Further, the multi-objective optimization process is as follows:

[0049] min.f(x)={f1(x),f2(x),f3(x),f4(x),f5(x),f6(x)}(14)

[0050] s.t.g(x)={g1(x),g2(x),f3(x),g4(x)}<g req (15)

[0051] wherein, f i(x), g(x) and g req are fitness function, constant function and preset threshold respectively;

[0052] Parent population and offspring population are randomly generated; the non-dominant solutions are grouped and distinguished by the crowding distance; the crowding distance is calculated as follows:

[0053]

[0054] In the formula, d(x) i is the crowding distance, and M represents the number of optimization objects. m (i+1) and p m (i-1) are the target values of the next individual and the previous individual of individual i on target m respectively.

[0055] Another object of the present application is to provide an air conditioner carbon reduction control system comprising:

[0056] The building parameter acquisition module is configured to acquire the hourly energy consumption data of the air conditioner terminal on the target floor in the cooling season, the hourly outdoor environmental parameters, the hourly indoor temperature and humidity, the time, the floor holiday calendar, and the hourly air volume information of the air conditioner terminal.

[0057] The prediction module is configured to construct an encoder-decoder time series prediction model for time series prediction.

[0058] The optimization module is configured to optimize the air volume control of the air conditioner terminal online to achieve building energy saving and carbon reduction.

[0059] Another object of the present application is to provide a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the computer program is executed by the processor to enable the processor to perform the steps of the air conditioner carbon reduction control method.

[0060] Another object of the present application is to provide a computer readable storage medium storing a computer program, wherein the computer program is executed by a processor to enable the processor to perform the steps of the air conditioner carbon reduction control method.

[0061] Another object of the present application is to provide an information data processing terminal for implementing the air conditioner carbon reduction control system.

[0062] In combination with the above technical solutions and the technical problems solved, the technical solution to be protected by the present application has the following advantages and positive effects:

[0063] First, the purpose of the present application is to propose an encoder-decoder model based on LSTM, GRU, SimpleRNN, CNN, attention algorithm, considering real-time high-precision prediction of carbon emissions of building partitions, and online optimization of air supply volume of single partition air conditioner terminal. The main problems to be solved are as follows:

[0064] (1) In order to solve the problems that the traditional building carbon reduction research is not fine enough and cannot identify and locate abnormal carbon emission area and period, the target floor (the ninth floor) is divided into west, south and east areas according to the electric well coverage range, the mutual relationship between single partitions is considered, and the real-time carbon emission data of each partition is taken as the model input for time series prediction.

[0065] (2) In order to solve the problems that the traditional deep learning model cannot process time series prediction and there are gradient disappearance and gradient explosion problems in the process of processing long time series prediction, an encoder-decoder algorithm framework integrating LSTM, GRU, SimpleRNN, CNN and attention module is constructed to improve the accuracy and robustness of time series prediction.

[0066] (3) In order to solve the problem that the traditional air conditioner terminal air supply volume is fixed and the setting is subjective, the air conditioner terminal hourly air volume online optimization is proposed, which can maximize the reduction of building carbon emissions while ensuring indoor personnel comfort.

[0067] The building operation carbon reduction theory framework provided by the present application has the following advantages:

[0068] (1) The encoder-decoder based on the present application can effectively analyze the relationship between regions and accurately predict the hourly carbon emission of buildings. The outdoor, indoor hourly temperature, humidity, holiday, time, partition carbon emission and hourly air volume information of each partition are taken as input features, and the historical information of the above features is used to predict the carbon emission intensity of each partition in the future one hour. The encoder-decoder model in the present application can accurately capture the mutual relationship between partitions and realize high-precision prediction of carbon emission. This method can be used as an auxiliary means to quickly locate the abnormal carbon emission area or period. At the same time, the prediction results can be compared with the historical same period or reference value, which solves the problem of difficult detection of small area carbon emission anomaly.

[0069] (2) The encoder-decoder framework integrating SimpleRNN, GRU, CNN, Bi-GRU and attention module proposed in this study can accurately extract key information in time series. SimpleRNN and CNN extract local features of time series, GRU extracts time series information, attention focuses on key information in the sequence, and Bi-GRU can solve the problem of long time series memory forgetting. The designed encoding-decoding model structure can ensure the scientificity of time series information extraction and improve the accuracy of carbon emission prediction.

[0070] (3) The online optimization of air conditioner terminal air volume proposed in this study can effectively reduce the carbon emission of buildings in the cooling season. Based on the constructed encoder-decoder proxy model, the carbon emission and comfort information at the next moment are predicted, and the change range of air conditioner terminal air volume is set. The optimal air volume parameter at each moment is searched by using NSGA-II, and the initial air conditioner terminal air volume setting parameter is replaced. This online optimization process can realize adaptive control of air conditioner terminal air volume, and realize building carbon reduction under the premise of ensuring indoor comfort.

[0071] The technical key points of the present application are:

[0072] A multi-partition carbon emission real-time prediction based on building partition interval connection is proposed, which can quickly locate the carbon emission abnormal area / time period and realize fine management of building carbon reduction.

[0073] An encoder-decoder hybrid model for time series prediction is constructed, which makes up for the problems of single deep learning model such as unable to predict time series and long time series information forgetting.

[0074] The theoretical method of online optimization of air conditioner terminal air volume is proposed, which can dynamically adjust the air volume of air conditioner according to the actual comfort needs of personnel, and greatly save building energy.

[0075] Secondly, as the creative auxiliary evidence of the claims of the present application, it is also embodied in the following important aspects:

[0076] (1) The expected income and commercial value of the technical scheme of the present application after transformation are:

[0077] The proposed online prediction and optimization scheme of high-rise building air conditioning system can improve the intelligent level of buildings, realize fine management of buildings, and greatly improve the energy utilization efficiency of buildings.

[0078] (2) The technical scheme of the present application fills the technical gap in the industry at home and abroad:

[0079] Through the prospect of existing literature and patents, we found that carbon emission timing prediction and carbon reduction optimization related research has achieved remarkable success in the field of building energy saving and emission reduction. Internet of Things, big data and other technologies provide new ideas for building operation and carbon reduction. Although there are many studies in this field, the existing research is not enough in terms of equipment optimization control, and it is difficult to be directly applied to building air conditioning system dynamic response and carbon reduction optimization. In view of this, this study proposes a real-time prediction and online optimization theoretical framework based on building partitioning, which greatly reduces the carbon emission intensity of buildings.

[0080] The scheme subdivides the target floor into west, south and east regions, and divides them based on the coverage range of the electric well, while considering the interconnection between the partitions. This solves the problem of insufficient refinement in traditional building carbon reduction research, allowing the research to accurately identify and locate abnormal carbon emission areas and periods. This refined division method greatly improves the accuracy of carbon emission monitoring and management, providing more reliable data support for building energy optimization.

[0081] In view of the limitations of traditional deep learning models in time series prediction, such as the difficulty in dealing with gradient disappearance and gradient explosion problems in long time series prediction, the scheme proposes an encoder-decoder algorithm framework that integrates LSTM, GRU, SimpleRNN, CNN and attention modules. This framework combines the advantages of different models, not only effectively capturing long-term dependencies in time series, but also enhancing the robustness and prediction accuracy of the model. Compared with a single model, the integrated framework performs well in complex and variable building carbon emission time series prediction, helping to achieve more accurate carbon emission control.

[0082] The scheme proposes an online optimization strategy for air conditioning terminal hourly air volume. In response to the shortcomings of traditional air conditioning terminal air volume being fixed and set subjectively, major improvements have been made. By adjusting the air volume of the air conditioner in real time, the scheme can minimize the carbon emissions of the building while ensuring the comfort of indoor personnel. This optimization strategy not only improves the energy efficiency of the air conditioning system, but also provides an intelligent and dynamic adjustment solution for building energy management.

[0083] (3) The technical scheme of the present invention solves the technical problems that people have been eager to solve but have always failed to succeed:

[0084] Building energy saving and carbon reduction still mainly targets the whole building, facing problems such as low building refinement, low resource utilization, and poor personnel comfort. The correlation between carbon emissions in different building partitions is rarely explored, and existing deep learning models have low accuracy in carbon emission time series prediction. In addition, the control of building air conditioning systems is subjective and cannot simultaneously optimize building carbon emissions and comfort. Traditional building optimization design focuses on the design stage, but most buildings have entered the operation period, and related parameters are difficult to adjust. Therefore, academia has been committed to innovative research on building energy saving and carbon reduction, although great efforts have been made, but so far no major breakthrough has been made.

[0085] Traditional building carbon emission research usually focuses on the whole or extensive management, and it is difficult to conduct fine analysis on different areas inside the building. This leads to difficulties in identifying and locating specific high-emission areas and abnormal emission periods in carbon emission control, which affects the effectiveness of overall energy efficiency optimization. The present invention breaks through this technical bottleneck by subdividing the target floor of the building into multiple areas and dividing based on the coverage range of the power well, achieving precise monitoring and management of carbon emissions in the internal areas of the building, solving the long-standing problem of fine monitoring.

[0086] In the field of time series prediction, traditional deep learning models have problems such as gradient disappearance and gradient explosion when processing long time series data, making it difficult to obtain high-precision prediction results. This technical challenge has plagued researchers, especially in complex building energy management, which is difficult to overcome. The present invention successfully solves this problem by constructing an encoder-decoder algorithm framework that integrates LSTM, GRU, SimpleRNN, CNN, and attention modules, achieving high-precision time series prediction of building carbon emissions and providing strong technical support for energy optimization.

[0087] The traditional air conditioning system has fixed terminal air volume and subjective settings, making it difficult to adjust dynamically at different times, resulting in low energy efficiency management. This problem has long existed in building carbon reduction and has not been successfully solved. The online optimization strategy for air conditioning terminal hourly air volume proposed by the present invention successfully breaks through this technical barrier, by adjusting the air volume in real time, not only ensuring indoor comfort, but also significantly reducing carbon emissions.

[0088] Fourth, the present application addresses the shortcomings of existing air conditioning systems in terms of energy saving and carbon reduction, and proposes a new control method to solve many problems in energy consumption and carbon emission management of traditional air conditioning systems. Existing technologies usually rely on static and fixed control strategies, which cannot adapt to complex building environments and variable external conditions, resulting in low operating efficiency of air conditioning systems, serious energy waste and high carbon emissions. In addition, existing systems lack real-time monitoring and accurate prediction means for carbon emissions, making it difficult to effectively implement energy saving and carbon reduction strategies.

[0089] The present application significantly improves the intelligent level of air conditioning systems by introducing a dynamic control method based on time series prediction. By obtaining the hourly energy consumption, environmental parameters and air conditioning operation data of the building, the system can build an accurate encoder-decoder time series prediction model to accurately predict future energy consumption and carbon emissions. Compared with traditional methods, the present application can dynamically adjust according to real-time data, avoiding the rigidity and lag of traditional control strategies, and significantly improving the energy saving effect and operating efficiency of air conditioning systems.

[0090] On this basis, the present application further introduces an online optimization mechanism to real-time regulate the air volume of air conditioning terminals. By calculating the comfort index (such as PMV value) of indoor environment and combining with the carbon emission prediction of partitioned areas, the system can accurately control the air volume of each area, ensuring the comfort of indoor environment and achieving low carbon emission. This optimization strategy breaks the one-size-fits-all operation mode of traditional air conditioning systems, and can flexibly cope with different environments and demands, further reducing overall energy consumption and carbon emissions.

[0091] The present application has made significant technical progress on the basis of existing technologies. Through the combination of dynamic prediction and real-time optimization control strategy, the present application effectively solves the shortcomings of traditional air conditioning systems in energy saving and carbon reduction, and realizes more refined and intelligent air conditioning system management. This technology not only improves the energy utilization efficiency of buildings and reduces operating costs, but also provides strong technical support for realizing green buildings and sustainable development goals. BRIEF DESCRIPTION OF DRAWINGS

[0092] Figure 1 is the air conditioning carbon reduction control method flowchart provided by the embodiment of the present application.

[0093] Figure 2 is the air conditioning carbon reduction control system structure block diagram provided by the embodiment of the present application.

[0094] Figure 3 is the target floor function partitioning diagram provided by the embodiment of the present application.

[0095] Figure 4is a schematic diagram of an Encoder module structure provided by an embodiment of the present application.

[0096] Figure 5 is a schematic diagram of a Decoder module structure provided by an embodiment of the present application.

[0097] Figure 6 is a structure diagram of an Encoder-decoder model provided by an embodiment of the present application.

[0098] Figure 7 is a process diagram of NSGA-II individual selection provided by an embodiment of the present application.

[0099] Figure 8 is a schematic diagram of air conditioner terminal online optimization provided by an embodiment of the present application. DETAILED DESCRIPTION

[0100] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in combination with embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0101] The following are two specific embodiments of industrial applications, which demonstrate the application of the air conditioner carbon reduction control method of the present application in different fields:

[0102] Embodiment one: optimization of air conditioning system in commercial office buildings

[0103] In large commercial office buildings, the air conditioning system is usually one of the largest energy-consuming devices and a major source of carbon emissions. Through the air conditioner carbon reduction control method of the present application, the energy-saving and carbon-reducing goals of the air conditioning system can be achieved according to the actual use of the office building. First, the system obtains the energy consumption data of each floor of the office building, the indoor and outdoor environmental parameters, and the air volume information of the air conditioner terminal from the building management system and performs data preprocessing. Then, based on these data, an encoder-decoder time series prediction model is constructed to predict future energy consumption and carbon emissions. On this basis, the system optimizes the air volume control of the air conditioner terminal online, adjusts the air volume according to real-time demand to ensure that the comfort level is met while minimizing energy consumption and carbon emissions. This method can significantly reduce the operating cost of the air conditioning system in large office buildings, while improving energy utilization efficiency and achieving sustainable development goals.

[0104] Embodiment two: air conditioning system management in intelligent industrial plants

[0105] In intelligent industrial plants, the energy consumption and carbon emission management of air conditioning systems is also crucial. Especially in some production workshops with strict requirements on temperature and humidity, the operation efficiency of air conditioning systems directly affects the stability of production processes. The method of the present application monitors the temperature and humidity and energy consumption data in the plant in real time, combines external environmental parameters, constructs a time series prediction model, and accurately predicts the energy consumption and carbon emission of the air conditioning system. Subsequently, the system adjusts the air volume and outlet air temperature of the air conditioning terminal in each workshop online according to the actual needs of different workshops through a multi-objective optimization algorithm. This method not only ensures the stability of the production environment, but also effectively reduces the overall energy consumption and carbon emission of the plant, helping enterprises achieve the goal of green manufacturing and low-carbon production.

[0106] The air conditioning carbon reduction control method of the present application achieves the energy saving and carbon reduction goal of building air conditioning systems through the coordinated work of multiple steps. First, the system obtains necessary parameter data from the actual operation of the building, including hourly energy consumption data of air conditioning terminals, outdoor environmental parameters, indoor temperature and humidity, floor holiday calendar, etc. After preprocessing, these data can ensure that the model construction and optimization process in the subsequent steps have accurate and complete basic data support. During the preprocessing process, the data is interpolated to fill in the missing data in the time series, and the carbon emission at each time is calculated according to the relationship between power consumption and carbon emission factors, laying a foundation for subsequent carbon emission prediction.

[0107] Next, using these processed building parameter data, the system constructs a time series prediction model based on an encoder-decoder structure. The Encoder module of the model is composed of SimpleRNN, GRU and CNN, responsible for time series encoding of input data, capturing the time sequence characteristics of data. The GRU model effectively retains and updates the information in the hidden state through the structure of the update gate and the reset gate, thereby improving the model's ability to capture long sequence dependencies. The Bi-GRU hybrid structure combines the advantages of forward and reverse GRU, further enhancing the model's understanding of data time sequence dependencies, and the introduction of attention mechanism enables the model to focus on the most critical input data, improving the accuracy and efficiency of prediction.

[0108] After the model construction and training are completed, the system enters the online optimization phase to control the air outlet volume of the air conditioner terminal in real time to achieve the goal of building energy saving and carbon reduction. Specifically, the system calculates the PMV value of each hour according to the temperature and humidity data obtained by the indoor sensor to evaluate the comfort level of the target area, and accurately predicts the carbon emissions according to the relationship between different partitions. On this basis, the system uses a multi-objective optimization algorithm to adjust the air volume output of the air conditioner terminal in real time to ensure that the comfort requirement is met while the carbon emissions are minimized. This optimization process combines the prediction results of the surrogate model and the relationship between partitions, making the control more intelligent and accurate.

[0109] Finally, in order to further optimize the performance of the system, the system introduces an algorithm based on the fitness function in the multi-objective optimization process, and uses the concept of crowding distance to group the solution space. By calculating the crowding distance of non-dominant solutions, the optimization target is divided, thereby effectively avoiding the local optimal problem in the optimization process and improving the overall optimization effect. This optimization strategy not only ensures the operation efficiency of the air conditioning system, but also significantly reduces energy consumption and carbon emissions, fully realizing the technical progress of the invention in building energy saving and carbon reduction.

[0110] As shown in Figure 1 , the air conditioning carbon reduction control method provided by the embodiment of the invention includes the following steps:

[0111] S101: Building parameter acquisition;

[0112] (1) Obtain the following information: target floor air conditioner terminal hourly energy consumption data, outdoor hourly environmental parameters, indoor hourly temperature and humidity, time, floor holiday calendar, and air conditioner terminal hourly air volume. Figure 3 The building energy consumption data mainly includes air conditioner terminal hourly energy consumption, which is measured by each floor single partition electric meter. The partition result is shown in

[0113]

[0114] Formula (1), t l is the time to be interpolated, t l-1 and t l+1 are the previous time and the next time of time t l , respectively. y l-1 , y l and y l+1 are the carbon emissions corresponding to time t l-1 , t l and t l+1 .

[0115] CE elec = Eelec ×EF elec (2)

[0116] E elec and EF elec are the electricity consumption and carbon emission factor (EF elec = 0.5703), respectively, and CE elec is the carbon emission corresponding to the electricity consumption.

[0117] S102: Build an encoder-decoder time series prediction model for time series prediction.

[0118] (1) The encoder module is composed of SimpleRNN, GRU and CNN, and the structure is as follows. Seq2seq model is a common algorithm used in natural language processing, and it has attracted more and more attention in the field of civil engineering. This model generally includes encoder and decoder modules, which can convert time series into sequences that are easy for computers to understand, and output the final sequence. Considering the special structure of time series, RNN usually plays an important role in capturing long-term dependencies in time series. However, when extracting valuable information from long time series, RNN will have the problem of gradient disappearance or gradient explosion. Therefore, we further optimize the model structure on the basis of previous research to improve its building carbon emission prediction performance. The encoder is composed of three neural network structures, namely SimpleRNN, GRU and CNN, and its framework is shown in Figure 4 .

[0119] (2) CNN can extract valuable information from time series, and we use one-dimensional convolutional neural network (1D-CNN) as a feature extraction module. This structure cannot be directly used for feature extraction of time series. Therefore, the other two improved RNN algorithm structures (i.e. SimpleRNN and GRU) are used for time series prediction. SimpleRNN, as one of the basic structures of RNN, can associate the current output with the output of the previous unit. More specifically, this model can remember previous information and be used for time series prediction. These features enable the traditional RNN model to learn an excellent structure that integrates historical data in the case of the latest prediction being divided into a small range of specific cases. However, when dealing with long-term time series, SimpleRNN usually has the problem of gradient disappearance or explosion. The RNN model structure is as follows:

[0120] h t = σ (wx t + Uh t-1 + b) (1)

[0121] In the formula, x t and h tThe input variable and hidden state corresponding to time t, w and U are the matrices corresponding to the current unit and input variable, respectively, and b is the bias of the neuron.

[0122] (3) Considering the performance deficiency of SimpleRNN in realizing long-term time series prediction, scholars have proposed LSTM and GRU to solve this problem. As an improved form of RNN, LSTM has three internal gates, namely input gate, forget gate and output gate, which are used to control the block state and output. Like LSTM, GRU also uses gates to achieve long-term time series prediction, and since it combines the input gate and the forget gate into an update gate, it is more efficient in training. In this study, GRU is used for long-term time feature extraction. The GRU model structure is as follows:

[0123] z t =σ(x t w xr +h t-1 w hr +b r ) (2)

[0124] r t =σ(x t w xz +h t-1 w hz +b z ) (3)

[0125]

[0126] In the formula, σ and tanh are sigmoid function and activation function, respectively, w xr and w hr are update gate weights, w xz and w hz are reset gate weights. w xh and w hh are candidate hidden layer states; x t is the input at time t, h t-1 is the hidden layer t-1 matrix, r t is the reset gate, z t is the update gate, b r , b z and b h are bias terms.

[0127] (4) In the encoder, these initial data are encoded and stored in the hidden layer, which is decoded into a new sequence and outputs the prediction results to the decoder. We designed the encoder using GRU, 1D-CNN and SimpleRNN, and the output of this part includes the feature information and time series information extracted by GRU and SimpleRNN, respectively. As the length of the time series increases, the performance of GRU will decline due to the generation of too much ambiguous information. Therefore, we use Bi-GRU composed of forward GRU and backward GRU to decode the sequence, so as to mine valuable information between the input and the prediction variance, and thus improve the prediction performance of the target floor air conditioning end carbon emission intensity. Figure 5 The proposed Bi-GRU hybrid structure and attention mechanism are demonstrated.

[0128]

[0129] where, and are the hidden layers of forward GRU and reverse GRU, respectively, and F is the combination method of Bi-GRU.

[0130] (5) The carbon emission intensity of buildings will be affected by many uncertain factors, such as weather conditions, seasons, building calendar information, etc. We want to pay more attention to these key factors and exclude the interference of unimportant factors. Therefore, we introduce the attention mechanism to help us identify these factors by giving more weight to the main factors. The attention mechanism is designed based on the model of human brain behavior, which can automatically focus on the main features of the target we observe. The calculation method of attention mechanism is as follows:

[0131] e t = utanh(wh t +b) (7)

[0132]

[0133] where e t is the probability distribution, u and w are weight matrices, b and s t are bias terms and attention mechanism outputs, respectively, and a t is the distribution coefficient.

[0134] (6) Figure 6The encoder module based on SimpleRNN-CNN-GRU and the decoder module based on Bi-GRU-Attention constitute the structure of the encoder-decoder model in this study. The role of the encoder module is to convert the input variables into a fixed-length sequence according to its neural layers, while the decoder outputs our prediction results. The asymmetric design of the encoder and decoder structure enables our model to complete more challenging encoding tasks, capture more complex correlations, and improve the performance of the model.

[0135] In this study, three commonly used regression indicators, including the coefficient of determination (R2), the root mean square error (RMSE), and the mean absolute error (MAE), are used to highlight the prediction performance of the encoder-decoder model constructed. R2 ranges from 0 to 1, and an index close to 1 indicates excellent model performance. Conversely, when the remaining error-related evaluation indicators are 0, it indicates excellent model performance.

[0136]

[0137] where, y i and represent the predicted value, observed value, and predicted mean value, and n is the sample size.

[0138] S103: Online optimization of air conditioning terminal air volume control to achieve building energy saving and carbon reduction;

[0139] (1) Target area comfort calculation. Personnel comfort is a qualitative indicator and is difficult to quantify. Common comfort indicators include PMV, PPD, and uncomfortable hours (hrs). In this study, the PMV index is used to describe personnel comfort, and the empirical calculation formula is as follows. As shown in equation (13), comfort is a binary linear relationship of indoor temperature and humidity. Therefore, the hourly PMV value can be calculated based on the hourly temperature and humidity information detected by the indoor sensor.

[0140]

[0141] where PMV is the comfort value, and are constant coefficients, and T and RH are temperature and relative humidity, respectively.

[0142] (2) Precise prediction of carbon emissions considering the relationship between zones. The preprocessed data is directly input into the model, and the carbon emission intensity of the next hour is predicted through a sliding window method. At the same time, the constructed proxy model (encoder-decoder) is used to predict the hourly PMV value.

[0143] (3) Figure 7Air conditioner terminal wind volume real-time online optimization. Based on the encoder-decoder time series prediction agent model constructed in the early stage, the carbon emission and comfort at a certain time are first predicted, and then the air outlet volume corresponding to each time is optimized by using the NSGA-II algorithm. The optimized air outlet volume replaces the initial air conditioner terminal air outlet volume. In this way, the terminal hourly air outlet volume is gradually optimized. NSGA-II is an optimization algorithm developed for multi-objective optimization problems. Compared with NSGA (non-dominated sorting genetic algorithm), the second generation algorithm comprehensively considers the dominated solution and the solution dominated by it. The crowding distance, which calculates the proximity of the solution to its neighboring solutions, is used to evaluate the fitness. The multi-objective optimization process is as follows:

[0144] min.f(x)={f1(x),f2(x),f3(x),f4(x),f5(x),f6(x)} (14)

[0145] s.t.g(x)={g1(x),g2(x),g3(x),g4(x)}<g req (15)

[0146] In the formula, f i (x), g(x) and g req are fitness function, constant function and preset threshold value respectively.

[0147] Parent population and offspring population are randomly generated. Non-dominant solutions are grouped and distinguished by crowding distance. The crowding distance is calculated as follows:

[0148]

[0149] In the formula, d(x) i is the crowding distance, M represents the number of optimization objects, p m (i+1) and p m (i-1) are the target values of the next individual and the previous individual of individual i on target m respectively.

[0150] Figure 8 Air conditioner terminal online optimization schematic diagram

[0151] As Figure 2 shown, another object of the present application provided by the embodiment of the present application is to provide a carbon reduction control system for an air conditioner, which comprises:

[0152] A building parameter acquisition module is configured to acquire hourly energy consumption data of an air conditioner terminal of a target floor in a cooling season, hourly outdoor environmental parameters, hourly indoor temperature and humidity, time, floor holiday calendar and hourly air volume information of the air conditioner terminal.

[0153] A prediction module is configured to construct an encoder-decoder time series prediction model for time series prediction.

[0154] An optimization module is configured to optimize the air conditioner terminal air volume control online to achieve building energy saving and carbon reduction.

[0155] Another object of the present application is to provide a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the computer program is executed by the processor to enable the processor to perform the steps of the air conditioner carbon reduction control method.

[0156] Another object of the present application is to provide a computer readable storage medium storing a computer program, wherein the computer program is executed by a processor to enable the processor to perform the steps of the air conditioner carbon reduction control method.

[0157] Another object of the present application is to provide an information data processing terminal for implementing the air conditioner carbon reduction control system.

[0158] On the ninth floor of a commercial complex, we subdivide the floor into west, south, and east regions according to the coverage range of the electrical well, and collect and input the carbon emission data of each region into the prediction model in real time. Compared with the traditional overall monitoring method, our scheme can accurately locate the carbon emission anomalies in specific regions, significantly improving the monitoring accuracy and response efficiency of carbon emissions. This fine division scheme verifies the creativity of the present application in regional carbon emission management, and can effectively improve the building energy efficiency.

[0159] For the historical carbon emission data of the building floor, we use an encoder-decoder algorithm framework integrating LSTM, GRU, SimpleRNN, CNN, and attention modules for prediction. During the experiment, the model of the present application not only shows better prediction accuracy than traditional single models, but also overcomes the common problems of gradient disappearance and gradient explosion in time series prediction, further proving the technical advantages of the present application in handling complex time series data.

[0160] In the building air conditioning system, we apply the online optimization technology of air conditioner terminal hourly air volume, which adjusts the air volume of the air conditioner in real time to ensure that the carbon emission is reduced to the maximum extent without reducing the indoor comfort. Compared with the traditional fixed air volume control, this scheme reduces energy consumption and significantly reduces carbon dioxide emissions. This optimization strategy verifies the practical application value of the present application in improving the energy efficiency management of the air conditioning system.

[0161] Through the verification of the above embodiments, we further confirm the creativity and technical value of the technical solutions of the present application, which not only solves the long-standing technical problems, but also has a wide application prospect.

[0162] The traditional building air conditioning equipment control mode is single and not fine enough, and the phenomenon of too high or too low regional temperature often occurs on the same floor, as shown in Table 1. This phenomenon not only causes a huge waste of building energy, but also causes indoor discomfort. Therefore, the floors are divided in this study. On the one hand, the temperature of the air conditioner in the partition is low, which will inevitably increase the carbon emissions of the building. Predicting the carbon emissions of the partition can quickly predict the carbon emissions of the partition, and compared with the historical same period, it can quickly judge whether the carbon emissions are abnormal.

[0163] Table 1 Indoor personnel comfort feedback of part of the floors

[0164]

[0165] The demand for cooling and resource consumption are different for each partition due to different personnel distribution. Optimizing the comfort and carbon emission targets of each partition can minimize carbon emissions. Traditional air conditioning experience control only uses one set temperature or air volume per day, but the fluctuation of cooling demand is caused by the hourly change of personnel in each partition. This study optimizes the comfort and carbon emissions of each partition every hour to obtain the optimal air volume at the next moment.

[0166] It should be noted that the embodiments of the present application can be realized by hardware, software or a combination of software and hardware. The hardware part can be realized by using special logic; the software part can be stored in the memory and executed by an appropriate instruction execution system, such as a microprocessor or a specially designed hardware. Those skilled in the art can understand that the above devices and methods can be realized by computer executable instructions and / or included in processor control code, such as provided on carrier media such as magnetic disk, CD or DVD-ROM, programmable memory such as read-only memory (firmware), or data carrier such as optical or electronic signal carrier. The device and its modules of the present application can be realized by hardware circuit such as ultra-large scale integrated circuit or gate array, semiconductor such as logic chip, transistor, or programmable hardware device such as field programmable gate array, programmable logic device, etc. It can also be realized by software executed by various types of processors, or by a combination of the above hardware circuit and software, such as firmware.

[0167] The above merely illustrates the specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any modification, equivalent replacement and improvement within the technical range disclosed by the present application and within the spirit and principle of the present application should be covered within the protection scope of the present application.

Claims

1. A carbon reduction control method of an air conditioner, characterized by, The method comprises the following steps: Step 1: building parameter acquisition; Step 2: constructing an encoder-decoder time series prediction model for time series prediction; Step 3: online optimization of air conditioner terminal air volume control to achieve building energy saving and carbon reduction; The building parameter acquisition method is as follows: Acquisition of air conditioner terminal hourly energy consumption data, outdoor hourly environmental parameters, indoor hourly temperature and humidity, time, floor holiday calendar, and air conditioner terminal hourly air volume information of the target floor in the cooling season; The collected data is preprocessed, specifically as follows: (1) Formula (1), is a time to be interpolated, and are respectively a previous time and a next time of the time ; , and are respectively carbon emissions corresponding to the time , and ; (2) and respectively, the power consumption and the carbon emission factor = 0.5703, is the carbon emission amount corresponding to the power consumption; The encoder-decoder time series prediction model is constructed for time series prediction as follows: (1) The encoder module is composed of SimpleRNN, GRU and CNN; (2) The RNN model structure is as follows: (1) wherein and The input variables and hidden states at time t, w and U are matrices corresponding to the current cell and input variables, respectively, and b is the bias of the neuron. (3) The GRU model structure is as follows: (2) (3) (4) (5) wherein, and tanh are the sigmoid function and the activation function, respectively, and is the update gate weight, and is the reset gate weight; is the input at time t, is the hidden layer t-1 matrix; (4) Bi-GRU hybrid structure and attention mechanism; (6) wherein and are the hidden layers of the forward and backward GRU, respectively, and F is the combination of the bidirectional GRUs; (5) The calculation method of the attention mechanism is as follows: (7) (8) (9) wherein, is a probability distribution, and is a weight matrix, b and are bias terms and attention mechanism outputs, respectively. (6) Coefficient of determination R2, root mean square error RMSE and mean absolute error MAE; (10) (11) (12) wherein and represent the predicted value, the observed value and the predicted average value, n is the number of samples; The online optimization of air conditioner terminal air volume control to achieve building energy saving and carbon reduction is as follows: (1) Target area comfort calculation; The hourly PMV value can be calculated according to the hourly temperature and humidity information detected by the indoor sensor; (13) (2) Precise prediction of carbon emissions considering the connection between partitions; (3) Real-time online optimization of air conditioner terminal air volume; multi-objective optimization; The precise prediction of carbon emissions considering the connection between partitions directly inputs the preprocessed data collected in the early stage into the model to predict the carbon emission intensity in the next hour through a sliding window method; at the same time, the constructed proxy model is used to predict the hourly PMV value; The multi-objective optimization process is as follows: (14) (15) In the formula, , and are fitness function, constant function and preset threshold value, respectively. Parent population and offspring population are randomly generated; the non-dominant solutions are grouped and distinguished by crowding distance; the crowding distance is calculated as follows: (16) In the formula, is the congestion distance, and M represents the number of optimization objects.

2. An air-conditioning carbon reduction control system for implementing the air-conditioning carbon reduction control method according to claim 1, characterized by The air conditioner carbon reduction control system comprises: A building parameter acquisition module for acquiring air conditioner terminal hourly energy consumption data, outdoor hourly environmental parameters, indoor hourly temperature and humidity, time, floor holiday calendar, and air conditioner terminal hourly air volume information of the target floor in the cooling season; A prediction module for constructing an encoder-decoder time series prediction model for time series prediction; An optimization module for online optimization of air conditioner terminal air volume control to achieve building energy saving and carbon reduction.

3. A computer device, comprising: The computer device comprises a memory and a processor, and the memory stores a computer program, which, when executed by the processor, causes the processor to execute the steps of the air conditioner carbon reduction control method of claim 1.

4. A computer readable storage medium storing a computer program, which, when executed by a processor, causes the processor to execute the steps of the air conditioner carbon reduction control method of claim 1.

5. An information data processing terminal, characterized by The information data processing terminal is used to implement the air conditioner carbon reduction control system of claim 2.

Images