Building air conditioner load monitoring system and method based on deep learning

Abstract

The invention relates to the technical field of building air conditioner load monitoring, and particularly discloses a building air conditioner load monitoring system and method based on deep learning. A data processing and analysis algorithm based on artificial intelligence and a deep learning technology is introduced at a rear end to carry out combined analysis on active power time sequence data and reactive power time sequence data of the building air conditioner; in this way, the interaction incidence relation between the active power time sequence characteristic and the reactive power time sequence characteristic of the building air conditioner and fusion characteristic representation information are captured, and therefore the air conditioner load mode type is recognized and detected. Therefore, the defects existing in a traditional NILM method can be effectively overcome, the time sequence fusion semantics between the reactive power and the active power of the building air conditioner are mined in a more intelligent mode to achieve air conditioner load mode type identification, and therefore the method can adapt to the change characteristics of different types of air conditioner load modes; powerful support is provided for optimizing the operation of the air conditioning system, and the goal of energy conservation and emission reduction is assisted to be achieved.

Description

Technical Field

[0001] The present application relates to the technical field of building air conditioning load monitoring, and more specifically, to a building air conditioning load monitoring system and method based on deep learning. Background Art

[0002] With the acceleration of urbanization and the improvement of people's living standards, the power consumption of air conditioning in commercial buildings has increased significantly, resulting in a sharp increase in electricity demand during peak hours. Therefore, it is becoming increasingly important to effectively monitor and analyze building air conditioning loads. Traditional building air conditioning load monitoring usually relies on invasive means, that is, by installing sensors on each air conditioning device to directly measure various parameters. This method is not only costly to deploy and difficult to construct, but also very difficult to implement for existing buildings.

[0003] In recent years, non-intrusive load monitoring (NILM) technology has gradually attracted attention. NILM does not require the installation of additional hardware on each electrical device (such as building air conditioners). Instead, it collects data at the total power inlet and combines it with advanced algorithm analysis to identify the working status of different electrical appliances. This method has the advantages of low deployment cost, high reliability, and easy scalability, and is particularly suitable for large-scale applications. However, traditional NILM methods usually rely only on simple statistical features or frequency domain analysis to identify the working status of air conditioners, and it is difficult to capture the complexity and diversity of air conditioner load patterns. For example, the spatial load pattern will change due to the influence of seasonal climate. When faced with different air conditioner load pattern types and power change characteristics, traditional air conditioner load monitoring methods are difficult to provide sufficient resolution.

[0004] Therefore, an optimized building air conditioning load monitoring solution is desired. Summary of the invention

[0005] The present application provides a building air-conditioning load monitoring system and method based on deep learning, which can effectively overcome the defects existing in the traditional NILM method, and realize the identification of air-conditioning load mode type by mining the time series fusion semantics between the reactive power and active power of the building air-conditioning in a more intelligent way, so as to adapt to the changing characteristics of different types of air-conditioning load modes, provide strong support for optimizing the operation of the air-conditioning system, and help achieve the goals of energy conservation and emission reduction.

[0006] In a first aspect, a building air conditioning load monitoring method based on deep learning is provided, comprising:

[0007] Using smart meters to sample the air conditioners in the monitored buildings to obtain the time queues of active power and reactive power;

[0008] Using a singular spectrum decomposition module, time series encoding is performed on the time queue of active power and the time queue of reactive power to obtain a plurality of active power time series subsequences and a plurality of reactive power time series subsequences;

[0009] Performing time series feature extraction on the multiple active power time series subsequences and the multiple reactive power time series subsequences respectively to obtain active power time series multi-scale coding features and reactive power time series multi-scale coding features;

[0010] The active power time series multi-scale coding features and the reactive power time series multi-scale coding features are subjected to feature alignment and fusion processing based on semantic information field modulation to obtain active power-reactive power time series fine-grained alignment and fusion features, including: performing semantic information field calculation on the active power time series multi-scale coding features and the reactive power time series multi-scale coding features to obtain a fine-grained semantic information field between active power-reactive power time series coding features; based on the fine-grained semantic information field between the active power-reactive power time series coding features, performing feature fusion based on field mapping fine-grained alignment on the active power time series multi-scale coding features and the reactive power time series multi-scale coding features to obtain the active power-reactive power time series fine-grained alignment and fusion features;

[0011] Based on the active power-reactive power time series fine-grained alignment fusion feature, a load monitoring result is determined, and the load monitoring result is used to represent an air conditioning load mode type label.

[0012] In the second aspect, a building air conditioning load monitoring system based on deep learning is provided, comprising:

[0013] The building air conditioning power sampling module is used to use the smart meter to sample the monitored building air conditioning to obtain the time queue of active power and the time queue of reactive power;

[0014] A power time series encoding module, used for performing time series encoding on the active power time series and the reactive power time series using a singular spectrum decomposition module to obtain a plurality of active power time series subsequences and a plurality of reactive power time series subsequences;

[0015] A power time series feature extraction module, used to extract time series features from the multiple active power time series subsequences and the multiple reactive power time series subsequences respectively to obtain active power time series multi-scale coding features and reactive power time series multi-scale coding features;

[0016] A power feature alignment and fusion processing module is used to perform feature alignment and fusion processing based on semantic information field modulation on the active power time series multi-scale coding features and the reactive power time series multi-scale coding features to obtain active power-reactive power time series fine-grained alignment and fusion features, wherein the power feature alignment and fusion processing module includes: a semantic information field calculation unit, which is used to perform semantic information field calculation on the active power time series multi-scale coding features and the reactive power time series multi-scale coding features to obtain a fine-grained semantic information field between active power-reactive power time series coding features; a fine-grained alignment and fusion processing unit, which is used to perform feature fusion based on field mapping fine-grained alignment on the active power time series multi-scale coding features and the reactive power time series multi-scale coding features based on the fine-grained semantic information field between the active power-reactive power time series coding features to obtain the active power-reactive power time series fine-grained alignment and fusion features;

[0017] The load monitoring result determination module is used to determine the load monitoring result based on the active power-reactive power time series fine-grained alignment fusion feature, and the load monitoring result is used to represent the air conditioning load mode type label.

[0018] The present application provides a building air conditioning load monitoring system and method based on deep learning, which can periodically sample the active power and reactive power of the air conditioner in the monitored building through a smart meter to form time series data, and then introduce a data processing and analysis algorithm based on artificial intelligence and deep learning technology in the back end to combine and analyze the active power time series data and reactive power time series data of the building air conditioner, so as to capture the interactive correlation relationship and fusion feature representation information between the active power time series characteristics and the reactive power time series characteristics of the building air conditioner, so as to identify and detect the air conditioning load mode type (such as switch two-state type, finite state type, continuous change type and fluctuation type). This building air conditioning load monitoring method based on deep learning can effectively overcome the defects in the traditional NILM method, and realize the identification of air conditioning load mode type by digging out the time series fusion semantics between the reactive power and active power of the building air conditioner in a more intelligent way, so as to adapt to the changing characteristics of different types of air conditioning load modes, and provide strong support for optimizing the operation of the air conditioning system, and help achieve energy conservation and emission reduction goals. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings of the embodiments of the present application are briefly introduced below. Obviously, the drawings described below only relate to some embodiments of the present application, and are not intended to limit the present application.

[0020] Figure 1 This is a schematic flowchart of a building air conditioning load monitoring method based on deep learning according to an embodiment of the present application.

[0021] Figure 2 This is a data flow diagram of a building air conditioning load monitoring method based on deep learning according to an embodiment of the present application.

[0022] Figure 3 This is a schematic flowchart of step S4 in the building air-conditioning load monitoring method based on deep learning in an embodiment of the present application.

[0023] Figure 4 This is a schematic flowchart of step S41 in the building air-conditioning load monitoring method based on deep learning in an embodiment of the present application.

[0024] Figure 5 This is a schematic flowchart of step S42 in the building air conditioning load monitoring method based on deep learning in an embodiment of the present application.

[0025] Figure 6 This is a schematic block diagram of a building air conditioning load monitoring system based on deep learning according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work also fall within the scope of protection of the present application.

[0027] In view of the above technical problems, in the technical solution of this application, a building air conditioning load monitoring method based on deep learning is proposed, which can regularly sample the active power and reactive power of the air conditioner in the monitored building through the smart meter to form time series data, and then introduce the data processing and analysis algorithm based on artificial intelligence and deep learning technology in the back end to combine and analyze the active power time series data and reactive power time series data of the building air conditioner, so as to capture the interactive correlation relationship and fusion feature representation information between the active power time series characteristics and the reactive power time series characteristics of the building air conditioner, so as to identify and detect the air conditioning load mode type (such as switch two-state type, finite state type, continuous change type and fluctuation type). This building air conditioning load monitoring method based on deep learning can effectively overcome the defects existing in the traditional NILM method, and realize the identification of air conditioning load mode type by digging out the time series fusion semantics between the reactive power and active power of the building air conditioner in a more intelligent way, so as to adapt to the changing characteristics of different types of air conditioning load modes, and provide strong support for optimizing the operation of the air conditioning system, and help achieve energy conservation and emission reduction goals.

[0028] Specifically, Figure 1 and Figure 2 As shown, the building air conditioning load monitoring method based on deep learning is characterized by comprising: S1, using a smart meter to sample the monitored building air conditioner to obtain a time queue of active power and a time queue of reactive power; S2, using a singular spectrum decomposition module to perform time series encoding on the time queue of active power and the time queue of reactive power to obtain multiple active power time series subsequences and multiple reactive power time series subsequences; S3, respectively performing time series feature extraction on the multiple active power time series subsequences and the multiple reactive power time series subsequences to obtain active power time series multi-scale coding features and reactive power time series multi-scale coding features; S4, performing feature alignment and fusion processing based on semantic information field modulation on the active power time series multi-scale coding features and the reactive power time series multi-scale coding features to obtain active power-reactive power time series fine-grained alignment and fusion features; S5, based on the active power-reactive power time series fine-grained alignment and fusion features, determining the load monitoring result, and the load monitoring result is used to represent the air conditioning load mode type label.

[0029] Exemplarily, in step S1, the air conditioner of the monitored building is sampled by a smart meter to obtain a time series of active power and a time series of reactive power. It should be understood that the smart meter can collect time series data of active power and reactive power of the air conditioning system in the monitored building at high frequency intervals (for example, every 15 seconds), and these data reflect the power output of the air conditioning equipment over time.

[0030] Exemplarily, in step S2, a singular spectrum decomposition module is used to perform time series encoding on the time queue of active power and the time queue of reactive power to obtain multiple active power time series subsequences and multiple reactive power time series subsequences. It should be understood that singular spectrum decomposition (Singular Spectrum Analysis, SSA) is a powerful time series analysis technology, which can be used to decompose, predict and reconstruct time series data. SSA does not rely on a specific data generation model, and is therefore applicable to various types of time series, including nonlinear and non-stationary series. In the power system, SSA can be used to process the time series data of active power and reactive power to extract meaningful subsequences, help understand and analyze the system dynamic behavior in the monitored building air conditioning, reduce noise interference, and provide a basis for subsequent identification and monitoring of air conditioning load mode types.

[0031] In one embodiment, a process of using a singular spectrum analysis (SSA) module to perform time series encoding on a time series of active power and a time series of reactive power to obtain multiple active power time series subsequences and multiple reactive power time series subsequences includes: selecting a window length L (also called an embedding dimension), which should be large enough to capture the periodicity and trend in the time series, but not too large to introduce too much noise. For each time series, a trajectory matrix is ​​constructed according to the selected window length L. For example, for the active power time series {X1, X2, ..., XN}, a trajectory matrix of size L×K (where K=N-L+1) is constructed, each column of the matrix is ​​a window of the time series, and the data of adjacent windows differ by only one time point. Singular value decomposition is performed on the constructed trajectory matrix. This step decomposes the original time series into several orthogonal components, each of which carries a part of the information of the original time series and is sorted according to the proportion of variance they explain. In this way, it is possible to identify which components correspond to noise and which correspond to the main modes of the signal (such as trends, seasonal fluctuations, etc.). According to actual needs or analysis results, the components corresponding to certain specific singular vectors can be selected for reconstruction. For example, the first few most important components can be selected to reconstruct the time series, thereby removing noise and extracting the main dynamic features; or the fluctuations in different frequency ranges can be separated as separate time series subsequences. Finally, the data processed by the above steps can be regarded as active power time series subsequences and reactive power time series subsequences after SSA encoding. These subsequences not only reduce the noise interference in the original data, but also retain important periodic and non-periodic components, which will help to further understand and analyze the operating status of the building air-conditioning system.

[0032] Exemplarily, in step S3, time series feature extraction is performed on the multiple active power time series subsequences and the multiple reactive power time series subsequences to obtain active power time series multi-scale coding features and reactive power time series multi-scale coding features. Considering that the multiple time series subsequence data on active power and reactive power after singular spectrum decomposition have time series correlation and mutual influence in the time dimension, this correlation feature is crucial for identifying complex air conditioning load patterns, because these patterns often have significant time dependence and periodicity.

[0033] In one embodiment, the multiple active power time series subsequences and the multiple reactive power time series subsequences are respectively subjected to time series feature extraction to obtain active power time series multi-scale coding features and reactive power time series multi-scale coding features, including: inputting the multiple active power time series subsequences and the multiple reactive power time series subsequences into a sequence encoder based on a bidirectional LSTM model to obtain active power time series multi-scale coding feature vectors as the active power time series multi-scale coding features and reactive power time series multi-scale coding feature vectors as the reactive power time series multi-scale coding features. It should be understood that through the processing of the sequence encoder based on the bidirectional LSTM model, the time series multi-scale long-term dependent correlation relationship and feature information of the active power and reactive power of the monitored building air conditioner in the time dimension can be captured respectively, which is very useful for understanding the long-term operating laws of the air conditioning system, such as the impact of seasonal changes on the air conditioning usage mode, thereby helping to better describe the diversity of air conditioning load modes, including but not limited to the switch two-state type, finite state type, continuous change type and fluctuation type, and providing a basis for the subsequent identification of air conditioning load mode types.

[0034] Exemplarily, in step S4, the active power time series multi-scale coding features and the reactive power time series multi-scale coding features are subjected to feature alignment fusion processing based on semantic information field modulation to obtain active power-reactive power time series fine-grained alignment fusion features. It should be understood that in building air-conditioning systems, there is a complex interaction between active power time series semantics and reactive power time series semantics. For example, under different operating modes, the change trends of these two powers may be both related and independent. Therefore, it is necessary to capture the subtle differences in these changes and the intrinsic connections between them more carefully, so as to provide support for accurate identification of air-conditioning load modes. However, traditional feature fusion methods may result in information loss or inaccurate representation, especially when processing data from different sources or different modalities. Moreover, the operation of the air-conditioning system is affected not only by internal mechanisms but also by external environmental factors. This means that there is a significant gap between low-level perceptual features directly obtained from power data and high-level semantic concepts. At the same time, air-conditioning systems in different buildings may also have different configurations and usage habits. Therefore, in the technical solution of the present application, the active power time series multi-scale coding features and the reactive power time series multi-scale coding features are further subjected to feature alignment and fusion processing based on semantic information field modulation to obtain active power-reactive power time series fine-grained alignment and fusion features.

[0035] In one embodiment, Figure 3As shown, the active power time series multi-scale coding features and the reactive power time series multi-scale coding features are subjected to feature alignment and fusion processing based on semantic information field modulation to obtain active power-reactive power time series fine-grained alignment and fusion features, including: S41, semantic information field calculation is performed on the active power time series multi-scale coding features and the reactive power time series multi-scale coding features to obtain the fine-grained semantic information field between the active power-reactive power time series coding features; S42, based on the fine-grained semantic information field between the active power-reactive power time series coding features, feature fusion based on field mapping fine-grained alignment is performed on the active power time series multi-scale coding features and the reactive power time series multi-scale coding features to obtain the active power-reactive power time series fine-grained alignment and fusion features.

[0036] In one embodiment, Figure 4 As shown, the semantic information field calculation is performed on the active power time series multi-scale coding features and the reactive power time series multi-scale coding features to obtain a fine-grained semantic information field between the active power-reactive power time series coding features, including: S411, dimensionally modulating the active power time series multi-scale coding feature vector and the reactive power time series multi-scale coding feature vector to obtain a modulated active power time series multi-scale coding feature vector and a modulated reactive power time series multi-scale coding feature vector; S412, performing fine-grained association coding and semantic information field coding on the modulated active power time series multi-scale coding feature vector and the modulated reactive power time series multi-scale coding feature vector to obtain a fine-grained semantic information field between the active power-reactive power time series coding features.

[0037] In one embodiment, in step S411, the active power time series multi-scale coded feature vector and the reactive power time series multi-scale coded feature vector are dimensionally modulated to obtain the modulated active power time series multi-scale coded feature vector and the modulated reactive power time series multi-scale coded feature vector, including: inputting the active power time series multi-scale coded feature vector and the reactive power time series multi-scale coded feature vector into a dimensional modulation module based on point convolution to obtain the modulated active power time series multi-scale coded feature vector and the modulated reactive power time series multi-scale coded feature vector, wherein the modulated active power time series multi-scale coded feature vector and the modulated reactive power time series multi-scale coded feature vector have the same feature dimension. Specifically, the process can be expressed by a formula:

[0038] v ′1 =LeakyReLU{Conv 1×1 (v1)}

[0039] v ′2 =LeakyReLU{Conv 1×1(v2)}

[0040] Wherein, v1 and v2 are the active power time series multi-scale coded feature vector and the reactive power time series multi-scale coded feature vector, Conv 1×1 (·) is the point convolution processing, Leaky ReLU (·) is the LeakyReLU activation function, v ′1 and v ′2 They are respectively the multi-scale coded feature vector of active power time series after modulation and the multi-scale coded feature vector of reactive power time series after modulation.

[0041] It should be understood that since the time series data of active power and reactive power may come from different physical processes, the feature vectors generated by each of them after being encoded by the bidirectional LSTM model may have different dimensions or structures. Directly fusing these feature vectors may result in information loss or inaccurate representation. Through the point convolution operation, the dimensions of the two feature vectors can be adjusted so that they can be compared and combined in the same space in subsequent processing, thereby ensuring the comparability and consistency between the features. Pointwise convolution, also known as 1x1 convolution, is used here to perform linear transformation, which can change the number of channels of the input feature map without affecting its spatial resolution. For this scenario, point convolution is used to remap the encoded feature vectors of active power and reactive power into a new feature space, where each dimension represents a new, higher-level abstract feature. This transformation can help capture more complex interactions between the original features, rather than just simple superposition or splicing. After point convolution, the generated modulated feature vector not only retains the original time dependency and periodicity information, but also introduces cross-domain information interaction, that is, the interaction between active power and reactive power. This helps reveal the potential correlation between the two, such as the phenomenon that the change trends of the two powers are both related and independent under different operating modes. Therefore, the modulated feature vector can better reflect the working state of the actual air-conditioning system and its dynamic change characteristics, providing a richer and more accurate feature representation for subsequent load pattern identification.

[0042] In one embodiment, in step S412, the modulated active power time series multi-scale coding feature vector and the modulated reactive power time series multi-scale coding feature vector are fine-grained associated and encoded with semantic information fields to obtain the fine-grained semantic information field between the active power and reactive power time series coding features, including: fine-grained associated encoding of the modulated active power time series multi-scale coding feature vector and the modulated reactive power time series multi-scale coding feature vector to obtain the active power-reactive power time series fine-grained associated feature matrix; the active power-reactive power time series fine-grained associated feature matrix is ​​input into the semantic information field coding network based on the convolution kernel to obtain the fine-grained semantic information field between the active power-reactive power time series coding features. Specifically, the process can be expressed as follows:

[0043]

[0044] Among them, v ′1 and v ′2 are respectively the multi-scale coded feature vector of active power time series after modulation and the multi-scale coded feature vector of reactive power time series after modulation, L is the length of the multi-scale coded feature vector of reactive power time series after modulation, M x is the active power-reactive power time series fine-grained correlation feature matrix, Conv 3×3 (·) is a convolutional layer with a convolution kernel of 3×3, Ω is the fine-grained semantic information field between active power-reactive power temporal encoding features, is matrix multiplication.

[0045] That is, through fine-grained correlation coding, the system can capture the local dependencies of the modulated active power and reactive power features at each time point, revealing the short-term behavior patterns in the air-conditioning system, such as startup, shutdown or other transient events. At the same time, the generated active power-reactive power time series fine-grained correlation feature matrix not only retains the time series properties of the original features, but also introduces cross-domain information interaction and enhances the feature expression capability. This enhanced feature representation can better reflect the dynamic behavior of the air-conditioning system, including but not limited to the switch two-state type, finite state type, continuous change type and fluctuating load mode. In addition, fine-grained correlation coding ensures that even the features extracted in different time series contexts can correspond accurately, which improves the quality and expression capability of the fused feature representation, and helps to more accurately describe the actual working state of the air-conditioning system.

[0046] Next, this fine-grained correlation feature matrix is ​​input into the semantic information field encoding network based on the convolution kernel. By performing convolution operations on the feature matrix, the semantic correlation relationship and hidden connection between the two features can be captured at a higher level. The convolution operation helps the model learn the spatial structure and pattern between the features and construct a space rich in semantic information. This enables the model to have a deeper understanding of the true meaning of the data, such as distinguishing similar but different load modes (such as two-state switch type and fluctuating type). The learning of the convolution layer allows the model to automatically discover and strengthen the subtle semantic relationships between active power and reactive power features, which may not be obvious to human analysts. In addition, the constructed semantic information field helps to alleviate domain differences, promote the effective transfer of knowledge and the rapid adjustment of the model. Even when faced with air-conditioning systems with different configurations and usage habits in different buildings, the trained model can better adapt to the new environment and provide accurate load pattern recognition results. Finally, a highly comprehensive and semantically rich active power-reactive power time series fine-grained aligned fusion feature representation is obtained.

[0047] In one embodiment, Figure 5 As shown, based on the fine-grained semantic information field between the active power-reactive power time series coding features, the active power time series multi-scale coding features and the reactive power time series multi-scale coding features are subjected to feature fusion based on field mapping fine-grained alignment to obtain the active power-reactive power time series fine-grained aligned fusion features, including: S421, mapping the modulated active power time series multi-scale coding feature vector and the modulated reactive power time series multi-scale coding feature vector to the fine-grained semantic information field between the active power-reactive power time series coding features respectively to obtain fine-grained aligned active power time series multi-scale coding feature vector and fine-grained aligned reactive power time series multi-scale coding feature vector; S422, calculating the position-weighted sum between the fine-grained aligned active power time series multi-scale coding feature vector and the fine-grained aligned reactive power time series multi-scale coding feature vector to obtain the active power-reactive power time series fine-grained aligned fusion feature vector as the active power-reactive power time series fine-grained aligned fusion feature.

[0048] In one embodiment, in step S421, the modulated active power time series multi-scale coding feature vector and the modulated reactive power time series multi-scale coding feature vector are respectively mapped to the fine-grained semantic information field between the active power-reactive power time series coding features to obtain fine-grained aligned active power time series multi-scale coding feature vector and fine-grained aligned reactive power time series multi-scale coding feature vector, including: multiplying the modulated active power time series multi-scale coding feature vector and the modulated reactive power time series multi-scale coding feature vector with the fine-grained semantic information field between the active power-reactive power time series coding features to obtain the fine-grained aligned active power time series multi-scale coding feature vector and the fine-grained aligned reactive power time series multi-scale coding feature vector. Specifically, the process can be expressed by the formula:

[0049]

[0050]

[0051] Among them, v ′1 and v ′2 are the multi-scale coding feature vectors of active power time series after modulation and the multi-scale coding feature vectors of reactive power time series after modulation, Ω is the fine-grained semantic information field between active power and reactive power time series coding features, is matrix multiplication, v 1t and v 2t They are respectively the fine-grained aligned active power time series multi-scale coded feature vector and the fine-grained aligned reactive power time series multi-scale coded feature vector.

[0052] That is, the present application uses a fine-grained semantic information field matrix as a medium, which contains deep interactive relationships extracted from the active power-reactive power time series fine-grained correlation feature matrix through convolution operations. When the present application multiplies the modulated active power and reactive power feature vectors with this semantic information field matrix, it actually maps the original features to a new, semantically rich space. This mapping enables the active power and reactive power features at each time point to find their best matching position in the semantic space, thereby achieving fine-grained alignment between features. Through such an alignment process, even time series features extracted in different contexts (for example, due to changes in the operating mode of the air-conditioning system or the influence of external environmental factors) can achieve consistency and correspondence at the semantic level. This not only improves the quality of feature representation, but also enhances the model's ability to understand complex dynamic behaviors. For example, during the startup, shutdown, and other transient events of an air conditioning system, the trends of active and reactive power may be both correlated and independent; through alignment, these subtle differences can be captured more accurately, which in turn helps to more finely distinguish different load mode types, such as two-state switch type, finite state type, continuous change type, and fluctuation type.

[0053] In one embodiment, in step S422, the position-weighted sum between the fine-grained aligned active power time series multi-scale coded feature vector and the fine-grained aligned reactive power time series multi-scale coded feature vector is calculated to obtain an active power-reactive power time series fine-grained aligned fusion feature vector as the active power-reactive power time series fine-grained aligned fusion feature. Specifically, the process can be expressed as follows:

[0054] v c =αv 1t +βv 2t

[0055] Among them, v 1t and v 2t are respectively the fine-grained aligned active power time series multi-scale coded feature vector and the fine-grained aligned reactive power time series multi-scale coded feature vector, α and β are weight hyperparameters, v c The active power-reactive power time series is finely granularly aligned and fused with feature vectors.

[0056] That is, the present application introduces a weighting mechanism to fuse the active power and reactive power feature vectors that have been fine-grained aligned. The weight at each position reflects the importance or contribution of the two power features at that position, which allows the model to dynamically adjust the balance between the two according to the actual situation. For example, in some cases, the change in active power may reflect the operating status of the air-conditioning system better than the reactive power; in other scenarios, the situation may be the opposite. Therefore, the process of weighted sum by position can flexibly capture this change, ensuring that the final fused features retain the original information and highlight the key features. The feature vector fused in this way - the active power-reactive power time series fine-grained alignment fusion feature vector, not only integrates the time series characteristics of active power and reactive power, but also incorporates the complex interaction between them. This fusion feature is very important for describing the real working state of the air-conditioning system because it helps to better understand the actual operating mode of the equipment and its dynamic change characteristics. For example, it can more accurately distinguish similar but different load modes (such as switch two-state type and fluctuation type), and provide richer and highly explanatory input data for subsequent classification tasks.

[0057] In summary, through the feature alignment fusion processing based on semantic information field modulation, the active power time series multi-scale coding feature vector and the reactive power time series multi-scale coding feature vector can be fine-grainedly aligned to ensure that even the features extracted in different time series contexts can accurately correspond, thereby improving the quality and expressiveness of the fused active power-reactive power time series feature representation. This is very important for describing the dynamic behavior of the air-conditioning system, as it helps to better understand the actual working state of the equipment. Moreover, through the feature alignment fusion processing based on semantic information field modulation, the semantic association relationship and hidden connection between the active power time series multi-scale coding features and the reactive power time series multi-scale coding features can be captured at a higher level, so that the model can have a deeper understanding of the true meaning of the data, such as distinguishing similar but different load modes (such as two-state switch type and fluctuating type). In addition, by constructing a space rich in semantic information, fine-grained alignment and deep interaction between the active power time series multi-scale encoding feature vector and the reactive power time series multi-scale encoding feature vector are achieved. The provided semantic information field helps to alleviate domain differences, promote the effective transfer of knowledge and the rapid adjustment of models, so as to capture and strengthen the subtle semantic relationship between the active power time series multi-scale encoding features and the reactive power time series multi-scale encoding features, thereby generating a highly comprehensive and semantically rich active power-reactive power time series fine-grained alignment fusion feature vector as the active power-reactive power time series fine-grained alignment fusion feature representation, which provides a basis for subsequent building air-conditioning load mode type identification and classification.

[0058] Exemplarily, in step S5, based on the active power-reactive power time series fine-grained alignment fusion feature, the load monitoring result is determined, and the load monitoring result is used to represent the air conditioning load mode type label. In one embodiment, based on the active power-reactive power time series fine-grained alignment fusion feature, the load monitoring result is determined, and the load monitoring result is used to represent the air conditioning load mode type label, including: inputting the active power-reactive power time series fine-grained alignment fusion feature vector into a load monitoring module based on a classifier to obtain the load monitoring result. That is to say, the fine-grained feature alignment fusion representation information between the active power time series multi-scale correlation feature and the reactive power time series multi-scale correlation feature of the monitored building air conditioner is used to perform classification processing, so as to identify and monitor the building air conditioning load mode type. It is worth mentioning that the results of building air conditioning load monitoring reveal the diversity of air conditioning load modes, which can be divided into four categories according to their power change characteristics: switch two-state type, finite state type, continuous change type and fluctuation type. This deep learning-based building air-conditioning load monitoring method can effectively overcome the defects of the traditional NILM method. It can realize the identification of air-conditioning load mode types by mining the time series fusion semantics between building air-conditioning reactive power and active power in a more intelligent way, so as to adapt to the changing characteristics of different types of air-conditioning load modes, provide strong support for optimizing the operation of the air-conditioning system, and help achieve energy conservation and emission reduction goals.

[0059] In one embodiment, the classifier-based load monitoring module includes a fully connected layer (DenseLayer) and a Softmax layer, wherein the fully connected layer (Dense Layer) maps the active power-reactive power time series fine-grained alignment fusion feature vector to a fixed-dimensional space in preparation for subsequent classification. After the fully connected layer, a Softmax layer is added. The Softmax function can convert any real value output by the previous layer into a probability distribution, in which each value represents the probability that the sample belongs to the corresponding category. For K category problems, the Softmax layer will generate a probability vector of length K, and the sum of all elements is equal to 1.

[0060] Here, when the active power time series multi-scale coding feature vector and the reactive power time series multi-scale coding feature vector respectively represent the time series multi-scale periodic coding features of the active power and reactive power of the monitored building air conditioner, after performing feature alignment interaction based on the semantic information field, the active power-reactive power time series fine-grained alignment fusion feature vector will also have an alignment interaction distribution space structure difference due to the large difference in the source data time series multi-scale correspondence offset, affecting the convergence consistency of the classifier and thus affecting the accuracy of the load monitoring results obtained by the load monitoring module.

[0061] Therefore, in one example, the active power-reactive power time series fine-grained alignment fusion feature vector is optimized, and the optimization process includes:

[0062] The absolute value sum of all eigenvalues ​​of the active power-reactive power time series fine-grained alignment fusion feature vector and the square root of the square sum are calculated to obtain the first active power-reactive power time series fine-grained alignment fusion space structure value and the second active power-reactive power time series fine-grained alignment fusion space structure value, that is:

[0063]

[0064] Among them, v i represents the i-th eigenvalue of the active power-reactive power time series fine-grained alignment fusion feature vector, w1 represents the first active power-reactive power time series fine-grained alignment fusion space structure value, and w2 represents the second active power-reactive power time series fine-grained alignment fusion space structure value;

[0065] Determine the number n of eigenvalues ​​of all eigenvalues ​​of the active power-reactive power time series fine-grained alignment fusion feature vector, that is, the length of the active power-reactive power time series fine-grained alignment fusion feature vector;

[0066] For each eigenvalue of the active power-reactive power time series fine-grained alignment fusion eigenvector, the first active power-reactive power time series fine-grained alignment fusion spatial structure value minus the product of the eigenvalue and the number of eigenvalues ​​is calculated to obtain the first active power-reactive power time series fine-grained alignment fusion long-range dependency value x i =w1-v i ×n, where w1 represents the first active power-reactive power time series fine-grained alignment fusion spatial structure value, v i represents the i-th eigenvalue of the active power-reactive power time series fine-grained aligned fusion feature vector, n represents the number of eigenvalues ​​of all eigenvalues ​​of the active power-reactive power time series fine-grained aligned fusion feature vector, x i It represents the first active power-reactive power time series fine-grained alignment fusion long-range dependency value;

[0067] Calculate the product of the square root of the number of eigenvalues ​​multiplied by the eigenvalue minus the second active power-reactive power time series fine-grained alignment fusion spatial structure value to obtain the second active power-reactive power time series fine-grained alignment fusion long-range dependency value Wherein, w2 represents the second active power-reactive power time series fine-grained alignment fusion spatial structure value, v irepresents the i-th eigenvalue of the active power-reactive power time series fine-grained alignment fusion feature vector, n represents the number of eigenvalues ​​of all eigenvalues ​​of the active power-reactive power time series fine-grained alignment fusion feature vector, y i Indicates the second active power-reactive power time series fine-grained alignment fusion long-range dependency value;

[0068] The exponential value calculated by taking the first active power-reactive power time series fine-grained alignment fusion long-range dependency value as the exponent of the natural constant and the inverse of the second active power-reactive power time series fine-grained alignment fusion long-range dependency value are weighted summed to obtain the optimized eigenvalue corresponding to each eigenvalue , where x i represents the first active power-reactive power time series fine-grained alignment fusion long-range dependency value, y i represents the second active power-reactive power time series fine-grained alignment fusion long-range dependency value, e represents a natural constant, γ and δ represent weighted hyperparameters, and f ′i Indicates the optimized eigenvalue corresponding to each eigenvalue;

[0069] The optimized eigenvalues ​​are combined into an optimized active power-reactive power time series fine-grained alignment fusion eigenvector.

[0070] That is, in view of the possible spatial structure loss in the high-dimensional space of the feature set of the active power-reactive power time series fine-grained aligned fusion feature vector, which causes the weight of the classifier to implicitly infer the spatial structure information based on the features, resulting in inconsistent convergence. A long-distance feature dependency relationship is established based on the spatial structure representation of the active power-reactive power time series fine-grained aligned fusion feature vector relative to the overall feature scale of the active power-reactive power time series fine-grained aligned fusion feature vector, so as to establish the feature local connectivity of the active power-reactive power time series fine-grained aligned fusion feature vector, and to capture the spatial ambiguous information of the object feature value through the unstructured feature value point prediction of the active power-reactive power time series fine-grained aligned fusion feature vector, thereby improving the spatial inductive deviation perception ability of the feature set of the active power-reactive power time series fine-grained aligned fusion feature vector, improving the convergence consistency of the classifier, and improving the accuracy of the load monitoring result obtained by the active power-reactive power time series fine-grained aligned fusion feature vector through the load monitoring module based on the classifier.

[0071] In summary, according to the embodiment of the present application, a building air-conditioning load monitoring method based on deep learning is explained. This building air-conditioning load monitoring method based on deep learning can effectively overcome the defects existing in the traditional NILM method, and realizes the identification of air-conditioning load mode types by mining the timing fusion semantics between the reactive power and active power of the building air-conditioning in a more intelligent way, so as to adapt to the changing characteristics of different types of air-conditioning load modes, provide strong support for optimizing the operation of the air-conditioning system, and help achieve the goals of energy conservation and emission reduction.

[0072] Figure 6 Schematic block diagram of a building air conditioning load monitoring system based on deep learning according to an embodiment of the present application. Figure 6 As shown, the building air conditioning load monitoring system 100 based on deep learning includes: a building air conditioning power sampling module 110, which is used to use a smart meter to sample the monitored building air conditioning to obtain a time queue of active power and a time queue of reactive power; a power time series encoding module 120, which is used to use a singular spectrum decomposition module to perform time series encoding on the time queue of active power and the time queue of reactive power to obtain multiple active power time series subsequences and multiple reactive power time series subsequences; a power time series feature extraction module 130, which is used to respectively encode the multiple active power time series subsequences and the multiple reactive power time series subsequences; The time series feature extraction is performed on the sequence to obtain the active power time series multi-scale coding features and the reactive power time series multi-scale coding features; the power feature alignment fusion processing module 140 is used to perform feature alignment fusion processing based on semantic information field modulation on the active power time series multi-scale coding features and the reactive power time series multi-scale coding features to obtain the active power-reactive power time series fine-grained alignment fusion features; the load monitoring result determination module 150 is used to determine the load monitoring result based on the active power-reactive power time series fine-grained alignment fusion features, and the load monitoring result is used to represent the air conditioning load mode type label.

[0073] In one embodiment, the power feature alignment fusion processing module includes: a semantic information field calculation unit, which is used to perform semantic information field calculation on the active power time series multi-scale coding features and the reactive power time series multi-scale coding features to obtain a fine-grained semantic information field between the active power-reactive power time series coding features; a fine-grained alignment fusion processing unit, which is used to perform feature fusion based on field mapping fine-grained alignment on the active power time series multi-scale coding features and the reactive power time series multi-scale coding features based on the fine-grained semantic information field between the active power-reactive power time series coding features to obtain the active power-reactive power time series fine-grained alignment fusion features.

[0074] In one embodiment, the power time series feature extraction module is used to: input the multiple active power time series subsequences and the multiple reactive power time series subsequences into a sequence encoder based on a bidirectional LSTM model respectively to obtain an active power time series multi-scale coding feature vector as the active power time series multi-scale coding feature and a reactive power time series multi-scale coding feature vector as the reactive power time series multi-scale coding feature.

[0075] Here, those skilled in the art can understand that the specific operations of the various modules and units in the above-mentioned building air conditioning load monitoring system based on deep learning have been referred to above. Figures 1 to 6 The description of the deep learning-based building air-conditioning load monitoring method has been introduced in detail, and therefore, its repeated description will be omitted.

[0076] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0077] It should be understood that the specific examples in this article are only intended to help those skilled in the art better understand the embodiments of the present application, rather than to limit the scope of the embodiments of the present application.

[0078] It should also be understood that in the various embodiments of the present application, the size of the serial number of each process does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0079] It should also be understood that the various implementation methods described in this specification can be implemented individually or in combination, and the embodiments of the present application are not limited to this.

[0080] Unless otherwise stated, all technical and scientific terms used in the embodiments of the present application are the same as the meanings generally understood by those skilled in the art of the technical field of the present application. The terms used in the present application are only for the purpose of describing specific embodiments and are not intended to limit the scope of the present application. The term "and / or" used in the present application includes any and all combinations of one or more related listed items. The singular forms of "a kind of", "above" and "the" used in the embodiments of the present application and the appended claims are also intended to include majority forms, unless the context clearly indicates other meanings. In addition, the terms "first", "second" etc. are only used for description purposes and cannot be understood as indicating or suggesting relative importance.

[0081] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0082] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0083] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0084] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0085] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A building air conditioning load monitoring method based on deep learning, characterized in that: include: Using smart meters to sample the air conditioners in the monitored buildings to obtain the time queues of active power and reactive power; Using a singular spectrum decomposition module, time series encoding is performed on the time queue of active power and the time queue of reactive power to obtain a plurality of active power time series subsequences and a plurality of reactive power time series subsequences; Performing time series feature extraction on the multiple active power time series subsequences and the multiple reactive power time series subsequences respectively to obtain active power time series multi-scale coding features and reactive power time series multi-scale coding features; The active power time series multi-scale coding features and the reactive power time series multi-scale coding features are subjected to feature alignment and fusion processing based on semantic information field modulation to obtain active power-reactive power time series fine-grained alignment and fusion features, including: performing semantic information field calculation on the active power time series multi-scale coding features and the reactive power time series multi-scale coding features to obtain a fine-grained semantic information field between active power-reactive power time series coding features; based on the fine-grained semantic information field between the active power-reactive power time series coding features, performing feature fusion based on field mapping fine-grained alignment on the active power time series multi-scale coding features and the reactive power time series multi-scale coding features to obtain the active power-reactive power time series fine-grained alignment and fusion features; Based on the active power-reactive power time series fine-grained alignment fusion feature, a load monitoring result is determined, and the load monitoring result is used to represent an air conditioning load mode type label.

2. The building air conditioning load monitoring method based on deep learning according to claim 1 is characterized in that: The multiple active power time series subsequences and the multiple reactive power time series subsequences are respectively subjected to time series feature extraction to obtain active power time series multi-scale coding features and reactive power time series multi-scale coding features, including: inputting the multiple active power time series subsequences and the multiple reactive power time series subsequences into a sequence encoder based on a bidirectional LSTM model to obtain active power time series multi-scale coding feature vectors as the active power time series multi-scale coding features and reactive power time series multi-scale coding feature vectors as the reactive power time series multi-scale coding features.

3. The building air conditioning load monitoring method based on deep learning according to claim 2 is characterized in that: The semantic information field calculation is performed on the active power time series multi-scale coding features and the reactive power time series multi-scale coding features to obtain a fine-grained semantic information field between the active power and reactive power time series coding features, including: Dimensionally modulating the active power time series multi-scale coded feature vector and the reactive power time series multi-scale coded feature vector to obtain a modulated active power time series multi-scale coded feature vector and a modulated reactive power time series multi-scale coded feature vector; Fine-grained association coding and semantic information field coding are performed on the modulated active power time series multi-scale coding feature vector and the modulated reactive power time series multi-scale coding feature vector to obtain a fine-grained semantic information field between the active power-reactive power time series coding features.

4. The building air conditioning load monitoring method based on deep learning according to claim 3 is characterized in that: Dimensionally modulating the active power time series multi-scale coded feature vector and the reactive power time series multi-scale coded feature vector to obtain a modulated active power time series multi-scale coded feature vector and a modulated reactive power time series multi-scale coded feature vector, including: The active power time series multi-scale coded feature vector and the reactive power time series multi-scale coded feature vector are input into a dimensional modulation module based on point convolution to obtain the modulated active power time series multi-scale coded feature vector and the modulated reactive power time series multi-scale coded feature vector, wherein the modulated active power time series multi-scale coded feature vector and the modulated reactive power time series multi-scale coded feature vector have the same feature dimension.

5. The building air conditioning load monitoring method based on deep learning according to claim 4 is characterized in that: Fine-grained association coding and semantic information field coding are performed on the modulated active power time series multi-scale coding feature vector and the modulated reactive power time series multi-scale coding feature vector to obtain a fine-grained semantic information field between the active power-reactive power time series coding features, including: Performing fine-grained correlation coding on the modulated active power time series multi-scale coded feature vector and the modulated reactive power time series multi-scale coded feature vector to obtain an active power-reactive power time series fine-grained correlation feature matrix; The active power-reactive power time series fine-grained correlation feature matrix is ​​input into a convolution kernel-based semantic information field encoding network to obtain a fine-grained semantic information field between the active power-reactive power time series encoding features.

6. The building air conditioning load monitoring method based on deep learning according to claim 5 is characterized in that: Based on the fine-grained semantic information field between the active power-reactive power time series coding features, the active power time series multi-scale coding features and the reactive power time series multi-scale coding features are subjected to feature fusion based on field mapping fine-grained alignment to obtain the active power-reactive power time series fine-grained alignment fusion features, including: Mapping the modulated active power time series multi-scale coding feature vector and the modulated reactive power time series multi-scale coding feature vector to the fine-grained semantic information field between the active power-reactive power time series coding features to obtain a fine-grained aligned active power time series multi-scale coding feature vector and a fine-grained aligned reactive power time series multi-scale coding feature vector; The position-weighted sum between the fine-grained aligned active power time series multi-scale encoded feature vector and the fine-grained aligned reactive power time series multi-scale encoded feature vector is calculated to obtain an active power-reactive power time series fine-grained aligned fusion feature vector as the active power-reactive power time series fine-grained aligned fusion feature.

7. The building air conditioning load monitoring method based on deep learning according to claim 6 is characterized in that: The modulated active power time series multi-scale coding feature vector and the modulated reactive power time series multi-scale coding feature vector are respectively mapped to the fine-grained semantic information field between the active power-reactive power time series coding features to obtain fine-grained aligned active power time series multi-scale coding feature vector and fine-grained aligned reactive power time series multi-scale coding feature vector, including: multiplying the modulated active power time series multi-scale coding feature vector and the modulated reactive power time series multi-scale coding feature vector with the fine-grained semantic information field between the active power-reactive power time series coding features to obtain the fine-grained aligned active power time series multi-scale coding feature vector and the fine-grained aligned reactive power time series multi-scale coding feature vector.

8. The building air conditioning load monitoring method based on deep learning according to claim 7 is characterized in that: Based on the active power-reactive power time series fine-grained alignment fusion feature, a load monitoring result is determined, and the load monitoring result is used to represent the air conditioning load mode type label, including: inputting the active power-reactive power time series fine-grained alignment fusion feature vector into a classifier-based load monitoring module to obtain the load monitoring result.

9. A building air conditioning load monitoring system based on deep learning, characterized in that: include: The building air conditioning power sampling module is used to use the smart meter to sample the monitored building air conditioning to obtain the time queue of active power and the time queue of reactive power; A power time series encoding module, used for performing time series encoding on the active power time series and the reactive power time series using a singular spectrum decomposition module to obtain a plurality of active power time series subsequences and a plurality of reactive power time series subsequences; A power time series feature extraction module, used to extract time series features from the multiple active power time series subsequences and the multiple reactive power time series subsequences respectively to obtain active power time series multi-scale coding features and reactive power time series multi-scale coding features; A power feature alignment and fusion processing module is used to perform feature alignment and fusion processing based on semantic information field modulation on the active power time series multi-scale coding features and the reactive power time series multi-scale coding features to obtain active power-reactive power time series fine-grained alignment and fusion features, wherein the power feature alignment and fusion processing module includes: a semantic information field calculation unit, which is used to perform semantic information field calculation on the active power time series multi-scale coding features and the reactive power time series multi-scale coding features to obtain a fine-grained semantic information field between active power-reactive power time series coding features; a fine-grained alignment and fusion processing unit, which is used to perform feature fusion based on field mapping fine-grained alignment on the active power time series multi-scale coding features and the reactive power time series multi-scale coding features based on the fine-grained semantic information field between the active power-reactive power time series coding features to obtain the active power-reactive power time series fine-grained alignment and fusion features; The load monitoring result determination module is used to determine the load monitoring result based on the active power-reactive power time series fine-grained alignment fusion feature, and the load monitoring result is used to represent the air conditioning load mode type label.

10. The building air conditioning load monitoring system based on deep learning according to claim 9, characterized in that: The power time series feature extraction module is used to: input the multiple active power time series subsequences and the multiple reactive power time series subsequences into a sequence encoder based on a bidirectional LSTM model respectively to obtain an active power time series multi-scale coding feature vector as the active power time series multi-scale coding feature and a reactive power time series multi-scale coding feature vector as the reactive power time series multi-scale coding feature.

Images