Non-intrusive air conditioner load identification method based on load enhancement feature imaging

Abstract

The invention discloses a non-intrusive air conditioner load identification method based on load enhancement feature imaging, which comprises the following steps: acquiring current sequence data of operation equipment and a corresponding load type according to known historical operation data of the air conditioner equipment; performing feature extraction on the current sequence data by adopting a time convolution network with enhanced internal information, and outputting one-dimensional current feature data; converting the one-dimensional current feature data into two-dimensional current feature data fusing deep and shallow features by using an improved feature matrix learning method; performing network model training on the two-dimensional convolutional neural network according to the two-dimensional current feature data and the corresponding load type to generate a load identification model; performing feature extraction on the real-time current sequence data of the to-be-identified equipment to obtain real-time two-dimensional current feature data; and inputting the real-time two-dimensional current data into a load identification model for load type identification, and outputting the load type of the to-be-identified equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of non-intrusive load identification, and more specifically, to a non-intrusive air-conditioning load identification method based on load enhancement feature graphics. Background Art

[0002] With the challenge of growing global energy consumption, non-intrusive load monitoring (NILM) technology has become a key solution. NILM technology plays a key role in energy management and control, especially in the industrial sector, through low-intrusive and cost-effective means. Traditional load monitoring methods usually require expensive system modifications and equipment installation, while NILM technology avoids these, greatly simplifying the monitoring process and reducing costs. This makes energy management in industrial environments more feasible and effective. NILM technology provides users with real-time insights into energy consumption patterns and peak hours, enabling better energy usage planning and management. This optimization supports the readjustment of energy consumption, which is particularly important in the industrial, commercial and residential sectors. By leveraging real-time data, users can intelligently control energy use, reduce waste and promote sustainable development. In industrial environments, such as manufacturing plants or factories, air conditioning is a significant heat load due to its high power consumption, long operating time and large temperature inertia. Understanding the load profile of air conditioning is crucial to achieve more efficient operation, as air conditioning accounts for about 35% of the building load.

[0003] Generally speaking, NILM algorithms can be divided into two categories: event-based methods and event-free methods. The former attempts to decompose appliances by detecting and clustering individual transitions in the signal; while the latter matches each sample in the total signal with the consumption of an appliance. In recent years, deep learning has been considered as the advocate of both approaches, especially using deep neural networks (DNNs) to achieve remarkable research results. Some well-known models include sequence-to-sequence (Seq2Seq) learning, sequence-to-point (Seq2Point) learning, recurrent neural network (RNN), window-gated recurrent unit (WGRU), and denoising autoencoder (DAE). The method described here belongs to the latter type of event-free detection methods.

[0004] Among event-based detection methods, finding effective features for appliance classification remains a major research challenge in NILM. Various techniques are used to convert time series data into images for analysis. Among these methods, methods such as Gramian angle field (GAF), Markov transition field (MTF), and encoding voltage and current signals into GAF representations are common. Common methods include: converting consumption time series signals into polar coordinates, and then using GAF sum field and difference field transformation to create a two-dimensional image; introducing an adaptive weighted recursive block for feature representation in event-driven NILM, which can convert the activation current of each cycle into a weighted cycle graph and learn related hyperparameters at the same time; proposing a hybrid prediction method based on time series image encoding technology MTF and convolutional neural network, aiming to improve the regression graph method that requires preset and weighted value constraints; and enhancing the uniqueness of load features by encoding the VI trajectory color, which improves the generalization performance of the model. However, these load feature construction methods all have the problem of information loss. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention provides a non-intrusive air-conditioning load identification method based on load enhancement feature graphics.

[0006] According to one aspect of the present invention, a non-intrusive air conditioning load identification method based on load enhancement feature graphics is provided, comprising:

[0007] According to the historical operation data of the known air-conditioning equipment, the current sequence data of the operating equipment and its corresponding load type are obtained;

[0008] A temporal convolutional network with enhanced internal information is used to extract features from current sequence data and output one-dimensional current feature data;

[0009] The one-dimensional current feature data is converted into two-dimensional current feature data integrating deep and shallow features by using an improved feature matrix learning method;

[0010] The two-dimensional convolutional neural network is trained according to the two-dimensional current characteristic data and the corresponding load type to generate a load identification model;

[0011] Extract features from the real-time current sequence data of the device to be identified, and obtain real-time two-dimensional current feature data;

[0012] The real-time two-dimensional current data is input into the load identification model for load type identification, and the load type of the device to be identified is output.

[0013] Optionally, a temporal convolutional network that enhances internal information is used to extract features from the current sequence data, and output one-dimensional current feature data, including:

[0014] A time convolution network is used to extract features from the current sequence data and output the initial one-dimensional current feature data;

[0015] The attention weight of each feature in the initial one-dimensional current feature data is calculated through the self-attention mechanism, and the one-dimensional current feature data is output.

[0016] Optionally, the temporal convolution network includes two dilated convolution operations, each of which uses a ReLU activation function and a Dropout operation, and further includes a residual connection after the second convolution operation, wherein the initial one-dimensional current feature data O n The expression is:

[0017]

[0018] Where d is the expansion coefficient. By injecting holes into the sequence, we can perform convolution operations with the time points at the previous moment. i is the weight of the convolution kernel, I [n-di] Refers to the data points of the input sequence at the adjusted interval position;

[0019] One-dimensional current characteristic data O n * The expression is:

[0020] O n * =O n +Attention

[0021] in, D k is a matrix that stabilizes the training gradient, softmax is an activation function with normalization function, Q, K, and V are the query, key, and value in the self-attention mechanism respectively.

[0022] Optionally, the one-dimensional current feature data is converted into two-dimensional current feature data integrating deep and shallow features using an improved feature matrix learning method, including:

[0023] The one-dimensional current feature data is convolved and pooled twice to obtain a shallow feature map and a deep feature map respectively;

[0024] According to the shallow feature map and the deep feature map, the improved feature matrix learning method is used to calculate the two-dimensional feature matrix, and the dimension of the two-dimensional feature matrix is ​​32*25*25.

[0025] Optionally, the operations of the shallow feature map F1 and the deep feature map F2 are:

[0026] F1 = MaxPool2d(ReLU(Conv2d(O n * )))

[0027] F2=MaxPool2d(ReLU(Conv2d(F1)))

[0028] In the formula, O n * is one-dimensional current feature data; the number of input channels of the shallow feature map is a1=1, the number of output channels is a2=64, the pixel padding value is padding_F1=1, the convolution kernel size is k1=2, and the stride is stride_F1=1; the number of input channels of the deep feature map is a2=64, the number of output channels is a3=64, the pixel padding value is padding_F2=1, the convolution kernel size of the deep feature map is k2=3, and the stride is stride_F2=1; where h and w represent the height and width of the feature map, respectively, and are calculated using the following formula:

[0029]

[0030] Where padding refers to the number of zero-valued pixels added around the feature map, kernel_size refers to the convolution kernel size, and stride refers to the step size of the convolution kernel sliding on the input feature map.

[0031] Optionally, the expression of the improved feature matrix learning method is:

[0032]

[0033] In the formula, O n * and W represent the input one-dimensional current characteristic data and network weight respectively, and IFM is the two-dimensional current characteristic data.

[0034] Optionally, the two-dimensional convolutional neural network includes: a convolution layer, a batch normalization layer, a ReLU activation function layer, a maximum pooling layer and an adaptive average pooling layer.

[0035] According to another aspect of the present invention, a non-intrusive air conditioning load identification device based on load enhancement feature graphics is provided, comprising:

[0036] An acquisition module, used to acquire the current sequence data of the operating equipment and its corresponding load type according to the historical operating data of the known air-conditioning equipment;

[0037] A first extraction module is used to extract features from the current sequence data using a time convolution network that enhances internal information, and output one-dimensional current feature data;

[0038] A conversion module, used for converting one-dimensional current feature data into two-dimensional current feature data integrating deep and shallow features by using an improved feature matrix learning method;

[0039] A training module is used to train a two-dimensional convolutional neural network model according to two-dimensional current characteristic data and corresponding load types to generate a load identification model;

[0040] The second extraction module is used to extract features from the real-time current sequence data of the device to be identified, and obtain real-time two-dimensional current feature data;

[0041] The identification module is used to input real-time two-dimensional current data into the load identification model for load type identification and output the load type of the device to be identified.

[0042] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the storage medium stores a computer program, and the computer program is used to execute the method described in any one of the above aspects of the present invention.

[0043] According to another aspect of the present invention, an electronic device is provided, comprising: a processor; a memory for storing instructions executable by the processor; the processor is configured to read the executable instructions from the memory and execute the instructions to implement the method described in any one of the above aspects of the present invention.

[0044] Therefore, the present invention obtains the current data of the known equipment during operation based on its operation data; secondly, a time convolution network with enhanced internal information is used to extract the features of the one-dimensional current data; then an improved feature matrix learning method is used to convert the one-dimensional features into two-dimensional features that fuse deep and shallow features; finally, the two-dimensional feature data is input into a two-dimensional convolutional neural network for network model training; for unknown equipment, feature mapping is performed based on its current data and the trained classification network is used to identify the air-conditioning load, thereby completing non-intrusive air-conditioning load identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] A more complete understanding of exemplary embodiments of the present invention may be obtained by referring to the following drawings:

[0046] Figure 1 is a flow chart of a non-intrusive air conditioning load identification method based on load enhancement feature graphics provided by an exemplary embodiment of the present invention;

[0047] Figure 2 is another flow chart of a non-intrusive air conditioning load identification method based on load enhancement feature graphics provided by an exemplary embodiment of the present invention;

[0048] Figure 3 is a schematic diagram of a feature image without feature enhancement provided by an exemplary embodiment of the present invention;

[0049] Figure 4is a schematic diagram of a feature image enhanced by the feature of the present invention provided by an exemplary embodiment of the present invention;

[0050] Figure 5 is a structural schematic diagram of a non-intrusive air conditioning load identification device based on load enhancement feature graphics provided by an exemplary embodiment of the present invention;

[0051] Figure 6 This is a structure of an electronic device provided by an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0052] Below, the exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments of the present invention, and it should be understood that the present invention is not limited to the exemplary embodiments described here.

[0053] It should be noted that the relative arrangement of components and steps, the numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present invention unless specifically stated otherwise.

[0054] Those skilled in the art can understand that the terms "first" and "second" in the embodiments of the present invention are only used to distinguish different steps, devices or modules, etc., and neither represent any specific technical meaning nor indicate the necessary logical order between them.

[0055] It should also be understood that, in the embodiments of the present invention, “plurality” may refer to two or more than two, and “at least one” may refer to one, two or more than two.

[0056] It should also be understood that any component, data or structure mentioned in the embodiments of the present invention can generally be understood as one or more, unless explicitly limited or otherwise indicated in the context.

[0057] In addition, the term "and / or" in the present invention is only a description of the association relationship of the associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in the present invention generally indicates that the associated objects before and after are in an "or" relationship.

[0058] It should also be understood that the description of the various embodiments of the present invention focuses on the differences between the various embodiments, and the same or similar aspects thereof can be referenced to each other, and for the sake of brevity, they will not be described one by one.

[0059] At the same time, it should be understood that for the convenience of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship.

[0060] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the invention, its application, or uses.

[0061] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered part of the specification.

[0062] It should be noted that like reference numerals and letters refer to similar items in the following figures, and therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0063] Embodiments of the present invention can be applied to electronic devices such as terminal devices, computer systems, servers, etc., which can operate with many other general or special computing system environments or configurations. Examples of well-known terminal devices, computing systems, environments and / or configurations suitable for use with electronic devices such as terminal devices, computer systems, servers, etc. include, but are not limited to: personal computer systems, server computer systems, thin clients, thick clients, handheld or laptop devices, microprocessor-based systems, set-top boxes, programmable consumer electronics, network personal computers, small computer systems, large computer systems, and distributed cloud computing technology environments including any of the above systems, etc.

[0064] Electronic devices such as terminal devices, computer systems, servers, etc. can be described in the general context of computer system executable instructions (such as program modules) executed by computer systems. Generally, program modules can include routines, programs, object programs, components, logic, data structures, etc., which perform specific tasks or implement specific abstract data types. Computer systems / servers can be implemented in a distributed cloud computing environment, where tasks are performed by remote processing devices linked through a communication network. In a distributed cloud computing environment, program modules can be located on local or remote computing system storage media including storage devices.

[0065] Exemplary Methods

[0066] Figure 1 FIG. 1 is a flow chart of a non-intrusive air conditioning load identification method based on load enhancement feature graphics provided by an exemplary embodiment of the present invention. This embodiment can be applied to electronic devices, such as Figure 1 As shown, the non-intrusive air conditioning load identification method 100 based on load enhancement feature graphics includes the following steps:

[0067] Step 101, obtaining the current sequence data of the operating equipment and its corresponding load type according to the historical operating data of the known air-conditioning equipment;

[0068] Step 102, using a temporal convolutional network that enhances internal information to extract features from the current sequence data, and outputting one-dimensional current feature data;

[0069] Step 103, using an improved feature matrix learning method to convert the one-dimensional current feature data into two-dimensional current feature data that integrates deep and shallow features;

[0070] Step 104, training a two-dimensional convolutional neural network model according to the two-dimensional current characteristic data and the corresponding load type to generate a load identification model;

[0071] Step 105, extracting features from the real-time current sequence data of the device to be identified, and obtaining real-time two-dimensional current feature data;

[0072] Step 106: input the real-time two-dimensional current data into the load identification model to identify the load type, and output the load type of the device to be identified.

[0073] Specifically, in the context of growing global energy consumption, traditional load monitoring methods have problems with system transformation and increased costs, and are unable to provide real-time insights into energy consumption patterns and peak hours without interfering with system operation. At the same time, existing air-conditioning load feature extraction methods have the problem of information loss when identifying load features, especially for event-driven load monitoring tasks, where effective air-conditioning load feature extraction is particularly critical. Therefore, the present invention aims to provide a new method based on non-intrusive load identification technology to achieve low-interference monitoring of the system and provide more accurate and representative load features, thereby achieving real-time insights and effective management of energy consumption.

[0074] In order to solve the above technical problems, the technical solution proposed by the present invention includes the following steps: first, based on the operating data of known equipment, the current data of the equipment during operation is obtained; secondly, the features of the one-dimensional current data are extracted by using a temporal convolutional network that enhances internal information; then, the one-dimensional features are converted into two-dimensional features that integrate deep and shallow features by using an improved feature matrix learning method; finally, the two-dimensional feature data is input into a two-dimensional convolutional neural network for network model training; for unknown equipment, feature mapping is performed based on its current data and the trained classification network is used to identify the air conditioning load, thereby completing non-intrusive air conditioning load identification. Specifically, the following steps are included:

[0075] Step D01: Based on the operating data of the known equipment, obtain the time t s Current data, assuming the sampling frequency is f s , then the number of sampling points in this time period N = t s ·f s , the current sequence data during operation I=[I1,I2,…,In ,…,I N ], 1≤n≤N.

[0076] Step D02: Using a temporal convolutional network that enhances internal information to extract features of the one-dimensional current data I to obtain output features O n * ;

[0077] A non-intrusive air conditioning load identification method based on load enhancement feature graphics is characterized in that the method adopts a temporal convolutional network that enhances internal information. The main steps include:

[0078] Step S01, the entire current sequence I is sent to the time series convolution network for feature extraction. The time series convolution network includes two dilated convolution operations, each convolution is followed by a ReLU activation function and a Dropout operation, and a residual connection is included after the second convolution operation. Assume that the number of input channels is num_inputs, the number of output channels of each layer is num_channels, the size of the convolution kernel is k, the discard rate of the convolution layer is dropout, the length of the input sequence is input_length, and the size of the batches sent to the network is batch_size. At this time, the dimension size of I sent to the network is batch_size*num_inputs*input_length, and the dilated convolution is performed on the sequence I, where any point I in the sequence n , the output O with the dimension of batch_size*num_channels*input_length is calculated by the following formula n :

[0079]

[0080] Where d is the expansion coefficient. By injecting holes into the sequence, we can perform convolution operations with the time points at the previous moment. i is the weight of the convolution kernel, I [n-di] Refers to the data points of the input sequence at the adjusted interval position. When d = 1, the dilated convolution is equivalent to the ordinary causal convolution. When d is 2, the input that the filter can accept is expanded from 1*3 to 1*5, and so on, compared with the ordinary causal convolution, thereby effectively expanding the receptive range of the convolution kernel and looking back on historical information to extract richer features.

[0081] Step S02: The feature O obtained in the temporal convolutional network nThe attention weight of each feature is calculated through the self-attention mechanism. By focusing on the differences in features, the internal information of the feature is enhanced. The size parameter of the self-attention is max_length. The output feature O is enhanced by the internal information obtained through the self-attention mechanism. n * , O n * The dimension sizes of and Attention are both batch_size*num_channels*max_length.

[0082] The attention mechanism is applied to the last layer of the temporal convolutional network to linearly map the input sequence X. The self-attention mechanism is defined as follows: Q = W q X, K = W k X、V=W v X. Among them, Q, K and V are the abbreviations of query, key and value respectively. K and V are one-to-one corresponding. The similarity is calculated by dividing the dot product of matrix Q and K by D k To calculate, in order to prevent the result value from being too large, and then use the softmax function for normalization, and finally apply the weighted summation method to get the final output, as shown in the following formula: D k is a matrix that stabilizes the training gradient, and softmax is an activation function with normalization function. The final feature with attention weight is O n * =O n +Attention.

[0083] Step D03: Use the improved feature matrix learning method to convert the one-dimensional feature into a two-dimensional feature IFM that fuses deep and shallow features.

[0084] Step S01, perform two convolution and pooling operations on the one-dimensional features to obtain a shallow feature map F1 and a deep feature map F2 respectively. The specific operations of the feature map are as follows:

[0085] F1 = MaxPool2d(ReLU(Conv2d(O n * )))

[0086] F2=MaxPool2d(ReLU(Conv2d(F1)))

[0087] Among them, MaxPool2d represents the maximum pooling operation, ReLU refers to the nonlinear activation operation, and Conv2d refers to the two-dimensional convolution operation. The number of input channels of the shallow feature map is a1, the number of output channels is a2, the pixel padding value is padding_F1, the convolution kernel size is k1, and the stride is stride_F1; the number of input channels of the deep feature map is a2, the number of output channels is a3, the pixel padding value is padding_F2, the convolution kernel size of the deep feature map is k2, and the stride is stride_F2. That is, the dimension of F1 is batch_size*a2*h1*w1, and the dimension of F2 is batch_size*a3*h2*w2, where h and w represent the height and width of the feature map respectively, calculated by the following formula: Where padding refers to the number of zero-valued pixels added around the feature map, kernel_size refers to the convolution kernel size, and stride refers to the step size of the convolution kernel sliding on the input feature map.

[0088] Step S02, using an improved feature matrix learning method to calculate the feature matrix IFM, the dimension of which is batch_size*(h1*w1)*(h2*w2).

[0089] The calculation formula of the improved feature matrix learning method is as follows:

[0090] Among them, n * and W represent the input feature map data and network weights respectively.

[0091] Step D04: Input the two-dimensional feature data into a pre-designed two-dimensional convolutional neural network to train the network model. In the network training, the category number of the equipment is used as the load label for iterative training.

[0092] The two-dimensional convolutional neural network consists of a convolutional layer, a batch normalization layer, a ReLU activation function layer, a maximum pooling layer, and an adaptive average pooling layer.

[0093] Step D05: For unknown devices, feature mapping is performed based on their current data through a pre-trained convolutional neural network, and these features are used to identify the air-conditioning load, thereby completing non-intrusive air-conditioning load identification.

[0094] The present invention discloses a non-intrusive air conditioning load identification method based on load enhancement feature graphics. The following is a detailed description with reference to the accompanying drawings: Figure 2It is a flow chart of the method described in the present invention. The experimental environment of this example is a 64-bit Windows 10 operating system, the CPU is Intel(R) Core(TM) i9-12900H, the graphics card is NVIDIA GeForce RTX 3060, the memory is 40G, and the Matlab software version is R2021b. The specific implementation process is as follows:

[0095] Step D01: Based on the operating data of the known equipment, obtain the current data for 2592000s (1 month). Assuming the sampling frequency is 0.2Hz, the number of sampling points in this time period is N=t s ·f s =51840, the current sequence data during operation I = [I1, I2, ..., I n ,…,I N ], 1≤n≤51840.

[0096] Step D02: Using a temporal convolutional network that enhances internal information to extract features of the one-dimensional current data I to obtain output features O n * ;

[0097] A non-intrusive air conditioning load identification method based on load enhancement feature graphics is characterized in that the method adopts a temporal convolutional network that enhances internal information. The main steps include:

[0098] Step S01, the entire current sequence I is sent to the time series convolution network for feature extraction. The time series convolution network includes two dilated convolution operations, each convolution is followed by a ReLU activation function and a Dropout operation, and a residual connection is included after the second convolution operation. The number of input channels is num_inputs = 1, the number of output channels of each layer is num_channels, the size of the convolution kernel is k = 2, the discard rate of the convolution layer is dropout = 0.2, the input sequence length is input_length = 200, and the size of the batches sent to the network is batch_size = 32. At this time, the dimension size of I sent to the network is 32*1*200, and a dilated convolution is performed on the sequence I, where any point I in the sequence n , the output O with dimension 32*8*200 is calculated by the following formula n :

[0099]

[0100] Where d is the expansion coefficient. By injecting holes into the sequence, we can perform convolution operations with the time points at the previous moment. i is the weight of the convolution kernel, I [n-di]Refers to the data points of the input sequence at adjusted interval positions.

[0101] Step S02: The feature O obtained in the temporal convolutional network n The attention weight of each feature is calculated through the self-attention mechanism. By focusing on the differences in features, the internal information of the feature is enhanced. The size parameter of the self-attention is max_length = 200. The output feature O is enhanced by the internal information obtained through the self-attention mechanism. n * , O n * The dimensions of and Attention are both 32*8*200.

[0102] The attention mechanism is applied to the last layer of the temporal convolutional network to linearly map the input sequence X. The self-attention mechanism is defined as follows: Q = W q X, K = W k X、V=W v X. Among them, Q, K and V are the abbreviations of query, key and value respectively. K and V are one-to-one corresponding. The similarity is calculated by dividing the dot product of matrix Q and matrix K by D k To calculate, in order to prevent the result value from being too large, and then use the softmax function for normalization, and finally apply the weighted summation method to get the final output, as shown in the following formula: D k is a matrix that stabilizes the training gradient, and softmax is an activation function with normalization function. The final feature with attention weight is O n * =O n +Attention.

[0103] Step D03: Use the improved feature matrix learning method to convert the one-dimensional feature into a two-dimensional feature IFM that fuses deep and shallow features.

[0104] Step S01, perform two convolution and pooling operations on the one-dimensional features to obtain a shallow feature map F1 and a deep feature map F2. The specific operation of the feature map is as follows: F1 = MaxPool2d (ReLU (Conv2d (O n *))), F2 = MaxPool2d (ReLU (Conv2d (F1))). Among them, MaxPool2d represents the maximum pooling operation, ReLU refers to the nonlinear activation operation, and Conv2d refers to the two-dimensional convolution operation. The number of input channels of the shallow feature map is a1 = 1, the number of output channels is a2 = 64, the pixel padding value is padding_F1 = 1, the convolution kernel size is k1 = 2, and the stride is stride_F1 = 1; the number of input channels of the deep feature map is a2 = 64, the number of output channels is a3 = 64, the pixel padding value is padding_F2 = 1, the convolution kernel size of the deep feature map is k2 = 3, and the stride is stride_F2 = 1. Among them, h and w represent the height and width of the feature map, respectively, and are calculated by the following formula: Where padding refers to the number of zero-valued pixels added around the feature map, kernel_size refers to the convolution kernel size, and stride refers to the step size of the convolution kernel sliding on the input feature map. That is, the dimension of F1 is 32*64*25*25, and the dimension of F2 is 32*64*25*25.

[0105] Step S02, using an improved feature matrix learning method to calculate a feature matrix IFM, the dimension of which is 32*25*25.

[0106] The calculation formula of the improved feature matrix learning method is as follows: Among them, n * and W represent the input feature map data and network weights respectively.

[0107] Step D04: Input the two-dimensional feature data into a pre-designed two-dimensional convolutional neural network to train the network model. In the network training, the category number of the equipment is used as the load label for iterative training.

[0108] The two-dimensional convolutional neural network consists of a convolution layer, a batch normalization layer, a ReLU activation function layer, a maximum pooling layer, and an adaptive average pooling layer. The hyperparameters of the network model are set as shown in Table 1.

[0109] Table 1 Hyperparameter settings

[0110]

[0111] Step D05: For unknown devices, feature mapping is performed based on their current data through a pre-trained convolutional neural network, and these features are used to identify the air-conditioning load, thereby completing non-intrusive air-conditioning load identification. Figure 3 and Figure 4 This is the graphical analysis result of the load enhancement feature obtained in this example.

[0112] The present invention provides a non-invasive air conditioning load identification method based on load enhancement feature graphics, which has the following significant effects and characteristics: 1) Non-invasive monitoring: By using existing current data, no hardware modification or interruption of the existing system is required, and non-invasive monitoring of the air conditioning system is realized. This not only reduces the cost of system monitoring, but also avoids possible interference with the normal operation of the system. 2) Enhanced feature extraction: The current feature data of the air conditioning load is effectively extracted and enhanced through the time convolution network (TCN) and feature matrix learning methods. These methods optimize the representation of features, so that the features not only retain the time dependence of the original data, but also enhance the correlation and discrimination between features. 3) Graphical feature representation: The improved feature matrix learning method in the present invention can convert the extracted one-dimensional current data features into a two-dimensional image form. This graphical representation method provides richer and more intuitive data input for the use of deep learning algorithms, especially when using convolutional neural networks (CNN) for further feature analysis and classification. 4) Improve the generalization ability of the system: The present invention demonstrates excellent model generalization ability by performing feature mapping and classification on the current data of unknown devices. This makes the method not limited to a specific type of air conditioning equipment, but can also be extended to load monitoring and identification of other similar equipment.

[0113] Therefore, the present invention obtains the current data of the known equipment during operation based on its operation data; secondly, a time convolution network with enhanced internal information is used to extract the features of the one-dimensional current data; then an improved feature matrix learning method is used to convert the one-dimensional features into two-dimensional features that fuse deep and shallow features; finally, the two-dimensional feature data is input into a two-dimensional convolutional neural network for network model training; for unknown equipment, feature mapping is performed based on its current data and the trained classification network is used to identify the air-conditioning load, thereby completing non-intrusive air-conditioning load identification.

[0114] Exemplary Devices

[0115] Figure 5 FIG. 1 is a schematic diagram of a non-intrusive air conditioning load identification device based on load enhancement feature graphics provided by an exemplary embodiment of the present invention. Figure 5 As shown, the device 500 includes:

[0116] An acquisition module, used to acquire the current sequence data of the operating equipment and its corresponding load type according to the historical operating data of the known air-conditioning equipment;

[0117] A first extraction module is used to extract features from the current sequence data using a time convolution network that enhances internal information, and output one-dimensional current feature data;

[0118] A conversion module, used for converting one-dimensional current feature data into two-dimensional current feature data integrating deep and shallow features by using an improved feature matrix learning method;

[0119] A training module is used to train a two-dimensional convolutional neural network model according to two-dimensional current characteristic data and corresponding load types to generate a load identification model;

[0120] The second extraction module is used to extract features from the real-time current sequence data of the device to be identified, and obtain real-time two-dimensional current feature data;

[0121] The identification module is used to input real-time two-dimensional current data into the load identification model for load type identification and output the load type of the device to be identified.

[0122] Optionally, the first extraction module includes:

[0123] The extraction submodule is used to extract features from the current sequence data using a time convolutional network and output initial one-dimensional current feature data;

[0124] The first calculation submodule is used to calculate the attention weight of each feature in the initial one-dimensional current feature data through a self-attention mechanism, and output the one-dimensional current feature data.

[0125] Optionally, the temporal convolution network includes two dilated convolution operations, each of which uses a ReLU activation function and a Dropout operation, and further includes a residual connection after the second convolution operation, wherein the initial one-dimensional current feature data O n The expression is:

[0126]

[0127] Where d is the expansion coefficient. By injecting holes into the sequence, we can perform convolution operations with the time points at the previous moment. i is the weight of the convolution kernel, I [n-di] Refers to the data points of the input sequence at the adjusted interval position;

[0128] One-dimensional current characteristic data O n * The expression is:

[0129] O n * =O n +Attention

[0130] in, D kis a matrix that stabilizes the training gradient, softmax is an activation function with normalization function, Q, K, and V are the query, key, and value in the self-attention mechanism respectively.

[0131] Optionally, the conversion module comprises:

[0132] The convolution submodule is used to perform two convolution and pooling operations on the one-dimensional current feature data to obtain a shallow feature map and a deep feature map respectively;

[0133] The second calculation submodule is used to calculate a two-dimensional feature matrix based on the shallow feature map and the deep feature map using an improved feature matrix learning method. The dimension of the two-dimensional feature matrix is ​​32*25*25.

[0134] Optionally, the operations of the shallow feature map F1 and the deep feature map F2 are:

[0135] F1 = MaxPool2d(ReLU(Conv2d(O n * )))

[0136] F2=MaxPool2d(ReLU(Conv2d(F1)))

[0137] In the formula, O n * is one-dimensional current feature data; the number of input channels of the shallow feature map is a1=1, the number of output channels is a2=64, the pixel padding value is padding_F1=1, the convolution kernel size is k1=2, and the stride is stride_F1=1; the number of input channels of the deep feature map is a2=64, the number of output channels is a3=64, the pixel padding value is padding_F2=1, the convolution kernel size of the deep feature map is k2=3, and the stride is stride_F2=1; where h and w represent the height and width of the feature map, respectively, and are calculated using the following formula:

[0138]

[0139] Where padding refers to the number of zero-valued pixels added around the feature map, kernel_size refers to the convolution kernel size, and stride refers to the step size of the convolution kernel sliding on the input feature map.

[0140] Optionally, the expression of the improved feature matrix learning method is:

[0141]

[0142] In the formula, O n *and W represent the input one-dimensional current characteristic data and network weight respectively, and IFM is the two-dimensional current characteristic data.

[0143] Optionally, the two-dimensional convolutional neural network includes: a convolution layer, a batch normalization layer, a ReLU activation function layer, a maximum pooling layer and an adaptive average pooling layer.

[0144] Exemplary Electronic Devices

[0145] Figure 6 This is a structure of an electronic device provided by an exemplary embodiment of the present invention. Figure 6 As shown, the electronic device 60 includes one or more processors 61 and a memory 62 .

[0146] The processor 61 may be a central processing unit (CPU) or other forms of processing units having data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions.

[0147] The memory 62 may include one or more computer program products, and the computer program product may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory (cache), etc. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 61 may run the program instructions to implement the methods of the software programs of the various embodiments of the present invention described above and / or other desired functions. In one example, the electronic device may also include: an input device 63 and an output device 64, which are interconnected via a bus system and / or other forms of connection mechanisms (not shown).

[0148] In addition, the input device 63 may also include, for example, a keyboard, a mouse, etc.

[0149] The output device 64 can output various information to the outside, and can include, for example, a display, a speaker, a printer, a communication network and a remote output device connected thereto.

[0150] Of course, to simplify, Figure 6 Only some of the components related to the present invention in the electronic device are shown, and components such as a bus, an input / output interface, etc. are omitted. In addition, the electronic device may further include any other appropriate components according to specific application conditions.

[0151] Exemplary computer program products and computer-readable storage media

[0152] In addition to the above-mentioned methods and devices, an embodiment of the present invention may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to execute the steps of the method according to various embodiments of the present invention described in the above-mentioned "Exemplary Method" section of this specification.

[0153] The computer program product may be written in any combination of one or more programming languages ​​to write program code for performing the operations of the embodiments of the present invention, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0154] In addition, an embodiment of the present invention may also be a computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, enable the processor to execute the steps of the method according to various embodiments of the present invention described in the above “Exemplary Method” section of this specification.

[0155] The computer readable storage medium can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can include, for example, but is not limited to, a system, system or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0156] The basic principle of the present invention is described above in conjunction with specific embodiments. However, it should be pointed out that the advantages, strengths, effects, etc. mentioned in the present invention are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. must be possessed by each embodiment of the present invention. In addition, the specific details disclosed above are only for the purpose of illustration and facilitation of understanding, rather than limitation, and the above details do not limit the present invention to being implemented by adopting the above specific details.

[0157] Each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system embodiment, since it basically corresponds to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0158] The block diagrams of the devices, systems, equipment, and systems involved in the present invention are only illustrative examples and are not intended to require or imply that they must be connected, arranged, and configured in the manner shown in the block diagram. As will be appreciated by those skilled in the art, these devices, systems, equipment, and systems can be connected, arranged, and configured in any manner. Words such as "including", "comprising", "having", etc. are open words, referring to "including but not limited to", and can be used interchangeably with them. The words "or" and "and" used here refer to the words "and / or" and can be used interchangeably with them, unless the context clearly indicates otherwise. The word "such as" used here refers to the phrase "such as but not limited to", and can be used interchangeably with it.

[0159] The method and system of the present invention may be implemented in many ways. For example, the method and system of the present invention may be implemented by software, hardware, firmware or any combination of software, hardware, firmware. The above order of steps for the method is only for illustration, and the steps of the method of the present invention are not limited to the order specifically described above, unless otherwise specifically stated. In addition, in some embodiments, the present invention may also be implemented as a program recorded in a recording medium, which includes machine-readable instructions for implementing the method according to the present invention. Thus, the present invention also covers a recording medium storing a program for executing the method according to the present invention.

[0160] It should also be noted that in the system, device and method of the present invention, each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be regarded as equivalent schemes of the present invention. The above description of the disclosed aspects is provided to enable any technician in the field to make or use the present invention. Various modifications to these aspects are very obvious to those skilled in the art, and the general principles defined here can be applied to other aspects without departing from the scope of the present invention. Therefore, the present invention is not intended to be limited to the aspects shown here, but in accordance with the widest range consistent with the principles and novel features disclosed here.

[0161] The above description has been given for the purpose of illustration and description. In addition, this description is not intended to limit the embodiments of the present invention to the forms disclosed herein. Although a number of example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions and sub-combinations thereof.

Claims

1. A non-intrusive air conditioning load identification method based on load enhancement feature graphics, characterized in that: include: According to the historical operation data of the known air-conditioning equipment, the current sequence data of the operating equipment and the corresponding load type are obtained; Using a time convolution network that enhances internal information to extract features from the current sequence data, and outputting one-dimensional current feature data; The one-dimensional current feature data is converted into two-dimensional current feature data integrating deep and shallow features by using an improved feature matrix learning method; Performing network model training on a two-dimensional convolutional neural network according to the two-dimensional current characteristic data and the corresponding load type to generate a load identification model; Extract features from the real-time current sequence data of the device to be identified, and obtain real-time two-dimensional current feature data; The real-time two-dimensional current data is input into the load identification model to perform load type identification, and the load type of the device to be identified is output.

2. The method according to claim 1, characterized in that: A temporal convolutional network that enhances internal information is used to extract features from the current sequence data, and output one-dimensional current feature data, including: Using a time convolution network to extract features from the current sequence data, and outputting initial one-dimensional current feature data; The attention weight of each feature in the initial one-dimensional current feature data is calculated through a self-attention mechanism, and the one-dimensional current feature data is output.

3. The method according to claim 2, characterized in that The temporal convolution network includes two dilated convolution operations, each of which uses a ReLU activation function and a Dropout operation, and also includes a residual connection after the second convolution operation, wherein the initial one-dimensional current feature data O n The expression is: Where d is the expansion coefficient. By injecting holes into the sequence, we can perform convolution operations with the time points at the previous moment. i is the weight of the convolution kernel, I [n-di] Refers to the data points of the input sequence at the adjusted interval position; The one-dimensional current characteristic data O n * The expression is: O n * =O n +Attention in, D k is a matrix that stabilizes the training gradient, softmax is an activation function with normalization function, Q, K, and V are the query, key, and value in the self-attention mechanism respectively.

4. The method according to claim 1, characterized in that: The one-dimensional current feature data is converted into two-dimensional current feature data integrating deep and shallow features by using an improved feature matrix learning method, including: The one-dimensional current characteristic data is subjected to two convolution and pooling operations to obtain a shallow feature map and a deep feature map respectively; According to the shallow feature map and the deep feature map, the two-dimensional feature matrix is ​​calculated using an improved feature matrix learning method, and the dimension of the two-dimensional feature matrix is ​​32*25*25.

5. The method according to claim 4, characterized in that The operations of the shallow feature map F1 and the deep feature map F2 are: F1=MaxPool2d(ReLU(Conv2d(O n * ))) F2=MaxPool2d(ReLU(Conv2d(F1))) In the formula, O n * is one-dimensional current feature data; the number of input channels of the shallow feature map is a1=1, the number of output channels is a2=64, the pixel padding value is padding_F1=1, the convolution kernel size is k1=2, and the stride is stride_F1=1; the number of input channels of the deep feature map is a2=64, the number of output channels is a3=64, the pixel padding value is padding_F2=1, the convolution kernel size of the deep feature map is k2=3, and the stride is stride_F2=1; where h and w represent the height and width of the feature map, respectively, and are calculated using the following formula: Where padding refers to the number of zero-valued pixels added around the feature map, kernel_size refers to the convolution kernel size, and stride refers to the step size of the convolution kernel sliding on the input feature map.

6. The method according to claim 5, characterized in that The expression of the improved feature matrix learning method is: In the formula, O n * and W represent the input one-dimensional current characteristic data and network weight respectively, and IFM is the two-dimensional current characteristic data.

7. The method according to claim 1, characterized in that The two-dimensional convolutional neural network includes: a convolution layer, a batch normalization layer, a ReLU activation function layer, a maximum pooling layer and an adaptive average pooling layer.

8. A non-intrusive air conditioning load identification device based on load enhancement feature graphics, characterized in that: include: An acquisition module, used for acquiring the current sequence data of the operating equipment and its corresponding load type according to the historical operating data of the known air-conditioning equipment; A first extraction module, configured to extract features from the current sequence data using a time convolution network that enhances internal information, and output one-dimensional current feature data; A conversion module, used for converting the one-dimensional current feature data into two-dimensional current feature data integrating deep and shallow features by using an improved feature matrix learning method; A training module, used to perform network model training on a two-dimensional convolutional neural network according to the two-dimensional current characteristic data and the corresponding load type, so as to generate a load identification model; The second extraction module is used to extract features from the real-time current sequence data of the device to be identified, and obtain real-time two-dimensional current feature data; The identification module is used to input the real-time two-dimensional current data into the load identification model to perform load type identification and output the load type of the device to be identified.

9. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and the computer program is used to execute the method according to any one of claims 1 to 7.

10. An electronic device, characterized in that: The electronic device comprises: processor; a memory for storing instructions executable by the processor; The processor is used to read the executable instructions from the memory and execute the instructions to implement the method described in any one of claims 1 to 7.

Images