CN115693918A

CN115693918A - Comprehensive intelligent power utilization system and method for building

Info

Publication number: CN115693918A
Application number: CN202211099055.3A
Authority: CN
Inventors: 许建成; 唐平亚
Original assignee: Zhejiang Xinyou Electromechanical Equipment Installation Co ltd
Current assignee: Zhejiang Xinyou Electromechanical Equipment Installation Co ltd
Priority date: 2022-09-07
Filing date: 2022-09-07
Publication date: 2023-02-03
Anticipated expiration: 2042-09-07
Also published as: CN115693918B

Abstract

The method comprises the steps of extracting implicit relevant features of state information of all electric equipment in a building to be monitored at the current time point by using a first convolutional neural network model as a feature extractor, extracting global semantic relevant features of type text description of all the electric equipment in all the electric equipment by using a context encoder comprising an embedded layer, fusing the features by using a graph neural network, obtaining a classification result for indicating whether state combination of all the electric equipment in the building to be monitored is normal or not by using a state semantic relevant feature matrix obtained after fusion through a classifier, and generating an electric abnormality prompt for abnormal state combination of all the electric equipment in the building to be monitored in response to the classification result. Therefore, the abnormity of the state combination of all the electric equipment in the building can be accurately analyzed and judged, and the early warning prompt is generated for the abnormal condition.

Description

Comprehensive intelligent power utilization system and method for building

Technical Field

The application relates to the technical field of intelligent power grids, in particular to a building comprehensive intelligent power utilization system and a method thereof.

Background

Due to the rapid development of social economy, people have higher and higher requirements on the comfort of the building environment, so that the building energy consumption is higher and higher, how to improve the comfort of the building environment without improving the building energy consumption can even achieve the energy-saving effect, and the energy-saving building becomes a research target of green buildings.

The intelligent power utilization technology is an important component for building a strong intelligent power grid, and the core of the intelligent power utilization technology is to realize intelligent service, meet diversified requirements of users, realize stable, reliable, economic and safe power supply and construct a novel power supply and utilization relation of real-time interaction of power flow, information flow and service flow between the power grid and the users. The energy utilization mode of the user is changed, energy conservation and emission reduction are promoted, and the proportion of clean electric energy in terminal energy consumption is improved. At present, china is still relatively lagged behind in the aspect of power utilization technology, most of the power utilization technologies adopt traditional switches, and the intelligent power utilization technology is less in application, which is an important aspect of energy consumption waste.

The emerging internet of things technology can organically combine multi-energy optimization and intelligent power utilization, so that the purposes of scientific power utilization and open source throttling are achieved, and the wide application of the internet of things technology in buildings becomes the inevitable trend of building intellectualization and energy conservation.

Therefore, an optimized comprehensive intelligent power utilization scheme for the building is expected.

Disclosure of Invention

The present application is proposed to solve the above-mentioned technical problems. The embodiment of the application provides a building comprehensive intelligent power utilization system and a method thereof, a first convolutional neural network model is used as a feature extractor to extract implicit relevant features of state information of all power utilization equipment in a building to be monitored at the current time point, a context encoder comprising an embedded layer is used to extract global semantic relevant features of type text description of all the power utilization equipment in all the power utilization equipment, then feature fusion is carried out through a graph neural network, finally a classification result used for indicating whether state combinations of all the power utilization equipment in the building to be monitored are normal or not is obtained through a classifier according to a state semantic relevant feature matrix obtained after fusion, and a power utilization abnormity prompt is generated for abnormal state combinations of all the power utilization equipment in the building to be monitored in response to the classification result. Therefore, the abnormity of the state combination of all the electric equipment in the building can be accurately analyzed and judged, and the early warning prompt is generated for the abnormal condition.

According to an aspect of the present application, there is provided a building integrated intelligent power utilization system, including:

the state information acquisition unit is used for acquiring the state information of all the electric equipment in the building to be monitored at the current time point;

the state association unit is used for arranging the state information of all the electric equipment in the building to be monitored at the current time point into a state input vector, and then calculating the product between the state input vector and the transposed vector thereof to obtain a state association matrix;

the state association feature extraction unit is used for obtaining a state association feature matrix by taking the state association matrix as a first convolution neural network model of a feature extractor after training;

the equipment description information acquisition unit is used for acquiring the type text description of each piece of electric equipment in all pieces of electric equipment;

the device description semantic coding unit is used for obtaining a plurality of device text description feature vectors by respectively training the type text description of each electric device in all the electric devices through a context coder which is finished and contains an embedded layer;

the matrix construction unit is used for arranging the device text description feature vectors into a device text description feature matrix according to the device sample dimension;

the graph neural network unit is used for enabling the device text description feature matrix and the state association feature matrix to pass through a trained graph neural network to obtain a state semantic association feature matrix, wherein the graph neural network encodes the device text description feature matrix and the state association feature matrix through learnable neural network parameters to obtain device text description association feature representation containing irregular state topological information;

the electric equipment state monitoring result generating unit is used for obtaining a classification result by the trained classifier of the state semantic association feature matrix, and the classification result is used for indicating whether the state combination of all electric equipment in the building to be monitored is normal or not; and

and the power utilization control result generation unit is used for responding to the abnormal state combination of all the power utilization equipment in the building to be monitored in the classification result and generating a power utilization abnormity prompt.

In the above building integrated intelligent power utilization system, the state association feature extraction unit is further configured to: each layer of the trained first convolutional neural network model as a feature extractor is respectively carried out in the forward transmission of the layer: performing convolution processing on input data to obtain a convolution characteristic diagram; performing mean pooling on the convolution feature map based on a local feature matrix to obtain a pooled feature map; and performing nonlinear activation on the pooled feature map to obtain an activated feature map; and the output of the last layer of the trained first convolution neural network model serving as the feature extractor is the state correlation feature matrix, and the input of the first layer of the trained first convolution neural network model serving as the feature extractor is the state correlation matrix.

In the above building integrated intelligent power utilization system, the device description semantic coding unit includes:

the word segmentation subunit is used for performing word segmentation processing on the type text description of each electric device in all the electric devices so as to convert the type text description of each electric device in all the electric devices into a word sequence consisting of a plurality of words; an embedding encoding subunit, configured to map, using an embedding layer of the context encoder, each word in the sequence of words to a word vector to obtain a sequence of word vectors; and a context encoding subunit, configured to perform global context-based semantic encoding on the sequence of word vectors using a converter of the context encoder to obtain the plurality of device text description feature vectors.

In the above building integrated intelligent power system, the power device state monitoring result generating unit is further configured to: processing the state semantic relevance feature matrix using the classifier to generate the classification result with a formula:

wherein

Representing the projection of the state semantic relevance feature matrix as a vector,

is a weight matrix of the fully-connected layer,

a bias matrix representing the fully connected layers.

In the above building integrated intelligent power utilization system, the building integrated intelligent power utilization system further includes a training module for performing joint training on the first convolution neural network model as the feature extractor, the context encoder including the embedding layer, the graph neural network, and the classifier.

In the above building integrated intelligent power utilization system, the training module includes: the system comprises a training data acquisition unit, a data processing unit and a data processing unit, wherein the training data acquisition unit is used for acquiring training data, and the training data comprises state information of all electric equipment in the building to be monitored at a preset time point and type text descriptions of all electric equipment in the electric equipment; the training state association unit is used for arranging the state information of all the electric equipment in the building to be monitored in the training data at a preset time point into a training state input vector, and then calculating the product between the training state input vector and the transposed vector thereof to obtain a training state association matrix; the training state associated feature extraction unit is used for enabling the training state associated matrix to pass through the first convolution neural network model serving as the feature extractor to obtain a training state associated feature matrix; a training device description semantic encoding unit, configured to pass type text descriptions of all the electrical devices in the training data through the context encoder including the embedded layer, respectively, to obtain a plurality of training device text description feature vectors; the training matrix constructing unit is used for arranging the plurality of training equipment text description feature vectors into a training equipment text description feature matrix according to the equipment sample dimension; the training graph neural network unit is used for enabling the training equipment text description feature matrix and the training state associated feature matrix to pass through the graph neural network so as to obtain a training state semantic associated feature matrix; the matrix expansion unit is used for expanding the training state semantic association feature matrix into classification feature vectors according to row vectors; the classification vector correction unit is used for correcting the classification feature vector by using the weight matrix of the classifier before and after each iteration update to obtain a corrected classification feature vector; the classification loss unit is used for enabling the corrected classification characteristic vector to pass through the classifier so as to obtain a classification loss function value; and a parameter updating unit, configured to jointly train the first convolutional neural network model as the feature extractor, the context encoder including the embedding layer, the graph neural network, and the classifier through back propagation of gradient descent with the classification loss function value as a loss function value.

In the above building integrated intelligent power utilization system, the classification vector correction unit is further configured to: correcting the classified feature vector by using a weight matrix before and after each iteration update of the classifier according to the following formula to obtain the corrected classified feature vector; wherein the formula is:

wherein

A feature vector representing the classification of the feature vector,

and

respectively representing the weight matrix of the classifier before and after each iteration of updating,

which represents the zero norm of the vector,

it is indicated that the sum is by position,

the difference in terms of position is indicated,

it is meant that the matrix multiplication is performed,

an exponential operation of a vector representing a calculation of a natural exponential function value raised to a power of a feature value of each position in the vector is represented.

According to another aspect of the present application, there is provided a building integrated intelligent power utilization method, including:

acquiring state information of all electric equipment in a building to be monitored at the current time point;

after arranging the state information of all the electric equipment in the building to be monitored at the current time point into state input vectors, calculating the product between the state input vectors and the transposed vectors thereof to obtain a state incidence matrix;

obtaining a state correlation characteristic matrix by using the trained first convolution neural network model serving as a characteristic extractor;

obtaining type text description of each electric device in all the electric devices;

respectively training the type text description of each electric device in all the electric devices by a context encoder containing an embedded layer to obtain a plurality of device text description feature vectors;

arranging the plurality of device text description feature vectors into a device text description feature matrix according to the device sample dimension;

obtaining a state semantic association feature matrix by passing the device text description feature matrix and the state association feature matrix through a trained graph neural network, wherein the graph neural network encodes the device text description feature matrix and the state association feature matrix through learnable neural network parameters to obtain a device text description association feature representation containing irregular state topology information;

obtaining a classification result by the trained classifier of the state semantic association feature matrix, wherein the classification result is used for indicating whether the state combination of all the electric equipment in the building to be monitored is normal or not; and

and generating a power utilization abnormity prompt in response to the fact that the classification result is that the state combinations of all the power utilization equipment in the building to be monitored are abnormal.

Compared with the prior art, the building comprehensive intelligent power utilization system and the method thereof have the advantages that the first convolutional neural network model is used as the feature extractor to extract implicit associated features of state information of all power utilization equipment in a building to be monitored at the current time point, the context encoder comprising the embedded layer is used to extract global semantic associated features of type text description of all the power utilization equipment in all the power utilization equipment, then the feature fusion is carried out through the graph neural network, finally, the obtained state semantic associated feature matrix after the fusion is used for obtaining a classification result for indicating whether state combination of all the power utilization equipment in the building to be monitored is normal or not through the classifier, and the power utilization abnormity prompt is generated for abnormal state combination of all the power utilization equipment in the building to be monitored in response to the classification result. Therefore, the abnormity of the state combinations of all the electric equipment in the building can be accurately analyzed and judged, and the early warning prompt is generated for the abnormal condition.

Drawings

The above and other objects, features and advantages of the present application will become more apparent by describing in more detail embodiments of the present application with reference to the attached drawings. The accompanying drawings are included to provide a further understanding of the embodiments of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings, like reference numbers generally indicate like parts or steps.

Fig. 1 illustrates an application scenario diagram of a building integrated intelligent power utilization system according to an embodiment of the present application.

Fig. 2 illustrates a block diagram of a building integrated intelligent electricity utilization system according to an embodiment of the present application.

Fig. 3 illustrates a block diagram of the device description semantic coding unit in the building integrated intelligent power utilization system according to an embodiment of the present application.

FIG. 4 illustrates a block diagram of the training module in the building integrated intelligent electricity utilization system according to an embodiment of the present application.

Fig. 5 illustrates a flow chart of a building integrated intelligent electricity utilization method according to an embodiment of the application.

Fig. 6 illustrates an architecture diagram of a building integrated intelligent electricity utilization method according to an embodiment of the present application.

Detailed Description

Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be understood that the described embodiments are only some embodiments of the present application and not all embodiments of the present application, and that the present application is not limited by the example embodiments described herein.

Overview of a scene

Accordingly, it is considered that when the technology of internet of things is used for organically combining the multi-energy optimization and the intelligent power utilization in the building, the anomaly analysis detection needs to be performed on the state combination of all the power utilization equipment in the building. However, at present, there are hidden state associations between some electric devices in a building, for example, curtains and lights are generally turned on and off at the same time, so it is necessary to sufficiently dig out the implicit state associations between the electric devices to determine the states of the electric devices in the building, and further generate a prompt signal when the abnormal states are determined. Therefore, after receiving the prompt signal, the user can adjust the state of the corresponding electric equipment to save energy.

Based on this, in the technical scheme of the application, the state implicit associated features of all the electric devices in the building and the global semantic associated features described by the type text of each electric device in all the electric devices are respectively extracted by using a deep neural network model, and the features are further fused by using a graph neural network, so that the classification effect of the state associated features of each electric device is strengthened and optimized based on the global type text description features of each electric device, and the accuracy of detection and judgment on the state anomaly of all the electric devices in the building is further improved.

Specifically, in the technical scheme of the application, firstly, state information of all electric equipment in a building to be monitored at the current time point is obtained. Here, if the electric device is in the on state, the value of the corresponding position is 1, and if the electric device is in the off state, the value of the corresponding position is 0.

And then arranging the state information of all the electric equipment in the building to be monitored at the current time point into a state input vector. It should be understood that, in order to fully dig out the implicit relationship to determine the abnormal state of the electric devices in the building when feature mining is performed, the product of the state input vector and the transposed vector thereof is further calculated to obtain a state association matrix, so as to construct the state association information between the electric devices, considering that there may be an implicit association feature between the states of the electric devices in the building, for example, curtains and lights are generally turned on and off at the same time. And then, carrying out deep correlation feature mining on the state correlation matrix by using a first convolution neural network model which has excellent performance in the aspect of implicit feature extraction and serves as a feature extractor to extract implicit correlation features of all electric equipment in the building to be monitored so as to obtain the state correlation feature matrix.

Further, considering that each of all the electric devices in the building has a different type, it is obvious that the accuracy of classification can be improved by using the type features to strengthen the relevance features of each electric device. Therefore, in the technical solution of the present application, semantic features described in the type text of each of the all electrical devices are selected and utilized to enhance and optimize the state relevance of each of the electrical devices. Specifically, type text descriptions of all the electric devices in all the electric devices are obtained, and the type text descriptions of all the electric devices in all the electric devices are respectively encoded through a context encoder comprising an embedded layer, so that global-based high-dimensional semantic features of the types of all the electric devices in all the electric devices are extracted to be more suitable for representing essential type relevance features among all the electric devices, and therefore a plurality of device text description feature vectors are obtained. In this way, the device text description feature vectors corresponding to the electric devices are arranged into a device text description feature matrix according to the device sample dimension, so as to integrate the relevance features of the electric devices based on the global electric device type.

It should be understood that each row vector in the device text description feature matrix corresponds to a type feature of each electrical device (node), and the state association feature matrix represents a high-dimensional implicit feature representation of state association (i.e., node-to-node association) between two corresponding electrical devices. Therefore, in the technical solution of the present application, the device text description feature matrix and the state association feature matrix construct a graph data structure. Therefore, in the technical solution of the present application, a graph neural network model is used as a feature encoder to encode the device text description feature matrix and the state association feature matrix through learnable neural network parameters to obtain a device text description association feature representation containing irregular state topology information.

And then, obtaining a classification result for indicating whether the state combinations of all the electric equipment in the building to be monitored are normal or not by the trained classifier of the state semantic association feature matrix. Accordingly, in a specific example of the present application, in response to the classification result being that the combination of the states of all the electric devices in the building to be monitored is abnormal, an electric abnormality prompt is generated.

Particularly, in the technical solution of the present application, for the state semantic relevance feature matrix obtained by passing the device text description feature matrix and the state relevance feature matrix through a graph neural network, since the state semantic relevance feature vector of the state semantic relevance feature matrix perpendicular to the sample dimension represents the state topological text description feature of a single electric device, the relevance of the state topological text description feature among multiple devices may be low, and thus, in the training process of the classifier, the difficulty in adapting between the parameters of the weight matrix of the classifier and the classification feature vector is increased.

Based on this, in the technical solution of the present application, while adjusting parameters of the weight matrix of the classifier in the training process, the feature vector obtained after the state semantic association feature matrix is expanded is, for example, recorded as

And performing classifier iterative scene-dependent optimization, which is expressed as:

and

is the weight matrix of the classifier before and after each iteration update,

representing the zero norm of the vector.

Here, the iterative scene-dependent optimization of the classifier optimizes the class probability representation of the classification feature vector by using the metric of the scene point correlation before and after the parameters of the weight matrix are updated during the iteration of the classifier as a correction factor, and describes the correlation of the classification feature vector by making support of the distribution similarity of the classification scene of the classifier, so as to improve the adaptability between the parameters of the weight matrix of the classifier and the classification feature vector from the perspective of the classification feature vector, and thus, by adjusting the parameters of the weight matrix of the classifier and the parameters of the classification feature vector at the same time, the training speed of the classifier and the accuracy of the classification result of the classification feature vector can be improved. Therefore, the abnormity of the state combination of all the electric equipment in the building can be analyzed and judged, and the early warning prompt is generated for the abnormal condition, so that the user can adjust the state of the corresponding electric equipment after receiving the prompt signal to realize energy conservation.

Based on this, this application has proposed a building synthesizes intelligent power consumption system, and it includes: the state information acquisition unit is used for acquiring the state information of all the electric equipment in the building to be monitored at the current time point;

the state association unit is used for arranging the state information of all the electric equipment in the building to be monitored at the current time point into a state input vector, and then calculating the product between the state input vector and the transposed vector of the state input vector to obtain a state association matrix; the state association feature extraction unit is used for obtaining a state association feature matrix by taking the state association matrix as a first convolution neural network model of a feature extractor after training; the equipment description information acquisition unit is used for acquiring the type text description of each piece of electric equipment in all pieces of electric equipment; the equipment description semantic coding unit is used for obtaining a plurality of equipment text description feature vectors by respectively training the type text description of each piece of electric equipment in all pieces of electric equipment through a context coder which contains an embedded layer; the matrix construction unit is used for arranging the device text description feature vectors into a device text description feature matrix according to the device sample dimension;

the graph neural network unit is used for enabling the device text description feature matrix and the state association feature matrix to pass through a trained graph neural network to obtain a state semantic association feature matrix, wherein the graph neural network encodes the device text description feature matrix and the state association feature matrix through learnable neural network parameters to obtain device text description association feature representation containing irregular state topological information; the electric equipment state monitoring result generating unit is used for obtaining a classification result by the trained classifier of the state semantic association feature matrix, and the classification result is used for indicating whether the state combination of all electric equipment in the building to be monitored is normal or not; and the power utilization control result generation unit is used for responding to the abnormal state combination of all the power utilization equipment in the building to be monitored in response to the classification result and generating a power utilization abnormity prompt.

Fig. 1 illustrates an application scenario diagram of a building integrated intelligent power utilization system according to an embodiment of the present application. As shown in fig. 1, in the application scenario, state information (e.g., M as illustrated in fig. 1) of all electric devices (e.g., T as illustrated in fig. 1) in a building to be monitored at a current time point and a type text description (e.g., C as illustrated in fig. 1) of each of the electric devices are obtained; then, the acquired state information of all the electric devices at the current time point and the type text description of each electric device are input into a server (for example, S shown in fig. 1) deployed with a building integrated intelligent electric system, wherein the server processes the state information of all the electric devices at the current time point and the type text description of each electric device by using a building integrated intelligent electric algorithm to generate a classification result for indicating whether the state combination of all the electric devices in the building to be monitored is normal or not, and an electric abnormality prompt is generated in response to the classification result that the state combination of all the electric devices in the building to be monitored is abnormal.

Having described the general principles of the present application, various non-limiting embodiments of the present application will now be described with reference to the accompanying drawings.

Exemplary System

Fig. 2 illustrates a block diagram of a building integrated intelligent electricity utilization system according to an embodiment of the present application. As shown in fig. 2, the building integrated intelligent power utilization system 100 according to the embodiment of the present application includes: the state information acquisition unit 110 is configured to acquire state information of all electric devices in the building to be monitored at a current time point; the state association unit 120 is configured to arrange state information of all electric devices in the building to be monitored at a current time point into a state input vector, and then calculate a product between the state input vector and a transposed vector thereof to obtain a state association matrix; a state association feature extraction unit 130, configured to obtain a state association feature matrix through a trained first convolution neural network model serving as a feature extractor; the device description information acquisition unit 140 is configured to acquire a type text description of each of the all electrical devices; the device description semantic encoding unit 150 is configured to obtain a plurality of device text description feature vectors by using a context encoder which is trained and contains an embedded layer to describe type text descriptions of each of the all electrical devices; a matrix constructing unit 160, configured to arrange the multiple device text description feature vectors into a device text description feature matrix according to the device sample dimension; the graph neural network unit 170 is configured to pass the device text description feature matrix and the state associated feature matrix through a trained graph neural network to obtain a state semantic associated feature matrix, where the graph neural network encodes the device text description feature matrix and the state associated feature matrix through learnable neural network parameters to obtain a device text description associated feature representation containing irregular state topology information; the electric equipment state monitoring result generating unit 180 is used for obtaining a classification result by the trained classifier of the state semantic association feature matrix, wherein the classification result is used for indicating whether the state combination of all electric equipment in the building to be monitored is normal or not; and the power utilization control result generating unit 190 is used for responding to the abnormal state combination of all the power utilization equipment in the building to be monitored in response to the classification result, and generating a power utilization abnormity prompt.

Specifically, in the embodiment of the present application, the state information collecting unit 110 is configured to obtain state information of all electric devices in a building to be monitored at a current time point. As can be seen from the foregoing, it is considered that when the technology of internet of things is utilized for organically combining multi-energy optimization and intelligent electricity utilization in a building, it is necessary to perform anomaly analysis detection on the state combinations of all the electric devices in the building. However, at present, there are hidden state associations between some electric devices in a building, for example, curtains and lights are generally turned on and off at the same time, so it is necessary to sufficiently dig out the implicit state associations between the electric devices to determine the states of the electric devices in the building, and further generate a prompt signal when the abnormal states are determined. Therefore, after receiving the prompt signal, the user can adjust the state of the corresponding electric equipment to save energy.

Based on this, in the technical scheme of the application, the state implicit associated features of all the electric devices in the building and the global semantic associated features described by the type text of each of the electric devices in the all the electric devices are respectively extracted by using a deep neural network model, and the features are further fused by using a graph neural network, so that the classification effect of the state associated features of each electric device is enhanced and optimized based on the global type text description features of each electric device, and the accuracy of the state anomaly detection and judgment of all the electric devices in the building is further improved.

Specifically, in this embodiment of the application, the state association unit 120 is configured to arrange state information of all electric devices in the building to be monitored at the current time point into a state input vector, and then calculate a product between the state input vector and a transposed vector thereof to obtain a state association matrix. That is, after state information of all electric devices in the building to be monitored at the current time point is obtained, the state information of all the electric devices in the building to be monitored at the current time point is arranged as a state input vector. It should be understood that the subsequent computer processing is facilitated by constructing the state information of all the electric devices in the building to be monitored at the current time point into a vector form.

It should be understood that, in order to fully dig out the implicit relationship to determine the abnormal state of the electric devices in the building when feature mining is performed, the product of the state input vector and the transposed vector thereof is further calculated to obtain a state association matrix, so as to construct the state association information between the electric devices, considering that there may be an implicit association feature between the states of the electric devices in the building, for example, curtains and lights are generally turned on and off at the same time.

Specifically, in this embodiment of the present application, the state-associated feature extraction unit 130 is configured to train the state-associated matrix through a trained first convolution neural network model serving as a feature extractor to obtain a state-associated feature matrix. Namely, deep correlation feature mining is carried out on the state correlation matrix by using a first convolution neural network model which has excellent performance in the aspect of implicit feature extraction and serves as a feature extractor, so as to extract implicit correlation features of all electric equipment in the building to be monitored, and therefore the state correlation feature matrix is obtained.

More specifically, in the embodiment of the present application, the layers of the trained first convolutional neural network model as the feature extractor are respectively performed in the forward pass of the layer: performing convolution processing on input data to obtain a convolution characteristic diagram; performing mean pooling based on a local feature matrix on the convolution feature map to obtain a pooled feature map; and performing nonlinear activation on the pooled feature map to obtain an activated feature map; and the output of the last layer of the trained first convolution neural network model serving as the feature extractor is the state correlation feature matrix, and the input of the first layer of the trained first convolution neural network model serving as the feature extractor is the state correlation matrix.

That is, the state association matrix is first input into the first layer of the trained first convolutional neural network model as the feature extractor, then each layer of the trained first convolutional neural network model as the feature extractor performs convolution processing, mean pooling processing and nonlinear activation processing on input data in forward transfer of the layer, and finally the output of the last layer of the trained first convolutional neural network model as the feature extractor is the state association feature matrix.

Specifically, in this embodiment of the present application, the device description information acquisition unit 140 and the device description semantic encoding unit 150 are configured to acquire a type text description of each of the all electrical devices, and obtain a plurality of device text description feature vectors by respectively training the type text descriptions of each of the all electrical devices through a context encoder including an embedded layer. It should be understood that considering that each of all the electric devices in the building has different types, the use of the type feature to strengthen the relevance feature of each electric device obviously can improve the classification accuracy. Therefore, in the technical solution of the present application, semantic features described in text by the type of each of the all electrical devices are selected to enhance and optimize the state relevance of each electrical device.

In the technical scheme of the application, the type text descriptions of all the electric devices in all the electric devices are obtained and are respectively encoded in a context encoder comprising an embedded layer, so that global high-dimensional semantic features of the types of all the electric devices in all the electric devices are extracted to be more suitable for representing essential type correlation features among all the electric devices, and therefore a plurality of device text description feature vectors are obtained.

More specifically, in this embodiment of the present application, fig. 3 illustrates a block diagram of the device description semantic coding unit in the building integrated intelligent power utilization system according to this embodiment of the present application, and as shown in fig. 3, the device description semantic coding unit 150 includes: a word segmentation subunit 210, configured to perform word segmentation on the type text description of each of the all electrical devices to convert the type text description of each of the all electrical devices into a word sequence composed of multiple words; an embedding encoding subunit 220, configured to map, using an embedding layer of the context encoder, each word in the sequence of words to a word vector to obtain a sequence of word vectors; and a context encoding subunit 230, configured to perform global context-based semantic encoding on the sequence of word vectors using a converter of the context encoder to obtain the plurality of device text description feature vectors.

In particular, the word segmentation refers to a process of segmenting a Chinese character sequence into a single word, that is, recombining continuous character sequences into a word sequence according to a certain specification. Specifically, in a specific example of the present application, an understanding-based word segmentation method may be selected, wherein the understanding-based word segmentation method achieves the effect of recognizing words by letting a computer simulate human understanding of sentences. In another specific example of the present application, a statistical-based word segmentation method may also be selected, where the statistical-based word segmentation method is to use a statistical machine learning model to learn a rule of word segmentation on the premise of giving a large amount of already segmented texts, so as to implement segmentation on unknown texts.

It should be understood that, in the technical solution of the present application, the context encoder performs global context semantic-based encoding on the sequence of word vectors using a Bert model based on a converter, and in particular, the Bert model performs global context semantic-based encoding on each word vector in the sequence of word vectors with a global semantic context of the sequence of word vectors based on an intrinsic mask structure of the converter to obtain the plurality of device text description feature vectors.

Specifically, in this embodiment of the application, the matrix constructing unit 160 is configured to arrange the device text description feature vectors into a device text description feature matrix according to the device sample dimension. That is, after the multiple device text description feature vectors are obtained, the multiple device text description feature vectors corresponding to the electric devices are arranged into a device text description feature matrix according to the device sample dimension, so as to integrate the relevance features of the electric devices based on the global electric device type.

Specifically, in this embodiment of the present application, the graph neural network unit 170 is configured to pass the device text description feature matrix and the state association feature matrix through a trained graph neural network to obtain a state semantic association feature matrix, where the graph neural network encodes the device text description feature matrix and the state association feature matrix through learnable neural network parameters to obtain a device text description association feature representation containing irregular state topology information. It should be understood that each row vector in the device text description feature matrix corresponds to a type feature of each electrical device (node), and the state association feature matrix represents a high-dimensional implicit feature representation of state association (i.e., node-to-node association) between two corresponding electrical devices.

Therefore, in the technical solution of the present application, the device text description feature matrix and the state association feature matrix construct a graph data structure. Specifically, the device text description feature matrix and the state association feature matrix are encoded through learnable neural network parameters by using a graph neural network model as a feature encoder to obtain a device text description association feature representation containing irregular state topology information.

Specifically, in this embodiment of the present application, the electric device status monitoring result generating unit 180 and the electric control result generating unit 190 are configured to use the status semantic association feature matrix to obtain a classification result through a classifier that is trained to complete, the classification result is used to indicate whether the status combination of all electric devices in the building to be monitored is normal, and is used to generate an electric abnormality prompt in response to that the classification result is abnormal in the status combination of all electric devices in the building to be monitored.

Namely, the state semantic relevance feature matrix is used for obtaining a classification result for indicating whether the state combination of all the electric equipment in the building to be monitored is normal or not through a trained classifier. Accordingly, in a specific example of the present application, in response to the classification result being that the combination of the states of all the electric devices in the building to be monitored is abnormal, an electric abnormality prompt is generated.

More specifically, in the embodiment of the present application, the classifier is used to process the state semantic relevance feature matrix with the following formula to generate the classification result, where the formula is:

wherein

is a weight matrix of the fully-connected layer,

a bias matrix representing the fully connected layers.

Further, in the technical solution of the present application, the building integrated intelligent power utilization system further includes a training module 300 for performing joint training on the first convolutional neural network model as the feature extractor, the context encoder including the embedded layer, the graph neural network, and the classifier.

Fig. 4 illustrates a block diagram of the training module in the building integrated intelligent power utilization system according to the embodiment of the present application, and as shown in fig. 4, the training module 300 includes: the training data acquisition unit 301 is configured to acquire training data, where the training data includes state information of all electric devices in the building to be monitored at a predetermined time point and text descriptions of types of the electric devices in all the electric devices; a training state association unit 302, configured to arrange state information of all electrical devices in the building to be monitored at a predetermined time point in the training data into a training state input vector, and then calculate a product between the training state input vector and a transposed vector thereof to obtain a training state association matrix; a training state associated feature extraction unit 303, configured to pass the training state associated matrix through the first convolutional neural network model as a feature extractor to obtain a training state associated feature matrix; a training device description semantic encoding unit 304, configured to pass type text descriptions of all the electrical devices in the training data through the context encoder including the embedded layer, respectively, to obtain a plurality of training device text description feature vectors; a training matrix constructing unit 305, configured to arrange the multiple training device text description feature vectors into a training device text description feature matrix according to the device sample dimension; a training diagram neural network unit 306, configured to pass the training device text description feature matrix and the training state associated feature matrix through the diagram neural network to obtain a training state semantic associated feature matrix; a matrix expansion unit 307, configured to expand the training state semantic relevance feature matrix into classification feature vectors according to row vectors; a classification vector correction unit 308, configured to correct the classification feature vector using the weight matrix before and after each iteration update of the classifier to obtain a corrected classification feature vector; a classification loss unit 309, configured to pass the corrected classification feature vector through the classifier to obtain a classification loss function value; and a parameter updating unit 310, configured to jointly train the first convolutional neural network model as the feature extractor, the context encoder including the embedding layer, the graph neural network, and the classifier with the classification loss function value as a loss function value through back propagation of gradient descent.

Particularly, in the technical scheme of the application, for the state semantic associated feature matrix obtained by passing the device text description feature matrix and the state associated feature matrix through a graph neural network, since the state semantic associated feature vector of the state semantic associated feature matrix perpendicular to the sample dimension represents the state topological text description feature of a single electric device, the correlation of the state topological text description features among multiple devices may be low, and thus, in the training process of the classifier, the difficulty in adapting between the parameters of the weight matrix of the classifier and the classification feature vector is increased.

Performing scene-dependent optimization of classifier iteration, namely using a weight matrix of the classifier before and after each iteration update to correct the classification feature vector according to the following formula to obtain the corrected classification feature vector; wherein the formula is:

wherein

Representing the classified feature vector in a manner that the classified feature vector,

and

which represents the zero norm of the vector,

it is indicated that the sum is by position,

the difference in terms of position is indicated,

it is meant that the matrix multiplication is performed,

Here, the iterative scene-dependent optimization of the classifier optimizes class probability representation of the classification feature vector by using a metric of scene point correlation before and after updating of parameters of a weight matrix of the classifier during iteration of the classifier as a correction factor, performs correlation description on the classification feature vector by making support for distribution similarity of classification scenes of the classifier, and improves adaptability between the parameters of the weight matrix of the classifier and the classification feature vector from the perspective of the classification feature vector, so that training speed of the classifier and accuracy of classification results of the classification feature vector can be improved by adjusting the parameters of the weight matrix of the classifier and the parameters of the classification feature vector at the same time. Therefore, the abnormity of the state combinations of all the electric equipment in the building can be analyzed and judged, and the early warning prompt is generated for the abnormal condition, so that the user can adjust the states of the corresponding electric equipment after receiving the prompt signal to save energy.

In summary, the building integrated intelligent power consumption system 100 according to the embodiment of the present application is clarified, which extracts implicit associated features of state information of all power consumption devices in a building to be monitored at a current time point by using a first convolutional neural network model as a feature extractor, extracts global semantic associated features described in type texts of each power consumption device in all power consumption devices by using a context encoder including an embedded layer, then performs feature fusion by using a graph neural network, finally obtains a classification result used for indicating whether state combinations of all power consumption devices in the building to be monitored are normal by using a classifier according to a state semantic associated feature matrix obtained after fusion, and generates a power consumption abnormal prompt for abnormal state combinations of all power consumption devices in the building to be monitored in response to the classification result. Therefore, the abnormity of the state combination of all the electric equipment in the building can be accurately analyzed and judged, and the early warning prompt is generated for the abnormal condition.

As described above, the building integrated intelligent power utilization system 100 according to the embodiment of the present application may be implemented in various terminal devices, such as a server for building integrated intelligent power utilization, and the like. In one example, the building integrated intelligent power system 100 according to the embodiment of the present application may be integrated into a terminal device as one software module and/or hardware module. For example, the building integrated intelligent power utilization system 100 may be a software module in the operating system of the terminal device, or may be an application developed for the terminal device; of course, the building integrated intelligent power system 100 may also be one of the hardware modules of the terminal device.

Alternatively, in another example, the building integrated intelligent power system 100 and the terminal device may be separate devices, and the building integrated intelligent power system 100 may be connected to the terminal device through a wired and/or wireless network and transmit the interaction information according to an agreed data format.

Exemplary method

Fig. 5 illustrates a flow chart of a building integrated intelligent electricity utilization method according to an embodiment of the application. As shown in fig. 5, the building integrated intelligent electricity utilization method according to the embodiment of the present application includes: s110, acquiring state information of all electric equipment in the building to be monitored at the current time point; s120, after arranging the state information of all the electric equipment in the building to be monitored at the current time point into a state input vector, calculating the product between the state input vector and the transposed vector thereof to obtain a state incidence matrix; s130, obtaining a state association feature matrix by using the state association matrix as a first convolution neural network model of a feature extractor after training; s140, obtaining type text description of each electric device in all the electric devices; s150, obtaining a plurality of device text description feature vectors by respectively training the type text description of each electric device in all the electric devices through a context encoder which is finished by training and comprises an embedded layer; s160, arranging the plurality of device text description feature vectors into a device text description feature matrix according to the device sample dimension; s170, passing the device text description feature matrix and the state association feature matrix through a trained graph neural network to obtain a state semantic association feature matrix, wherein the graph neural network encodes the device text description feature matrix and the state association feature matrix through learnable neural network parameters to obtain a device text description association feature representation containing irregular state topology information; s180, obtaining a classification result by the trained classifier of the state semantic association feature matrix, wherein the classification result is used for indicating whether the state combination of all the electric equipment in the building to be monitored is normal or not; and S190, responding to the abnormal state combination of all the electric equipment in the building to be monitored as the classification result, and generating an electric abnormality prompt.

Fig. 6 illustrates an architecture diagram of a building integrated intelligent electricity utilization method according to an embodiment of the present application. As shown in fig. 6, first, in the network architecture of the building integrated intelligent electricity utilization method, state information of all electricity utilization devices in a building to be monitored at a current time point is obtained; then, after arranging the state information of all the electric equipment in the building to be monitored at the current time point into state input vectors, calculating the product between the state input vectors and the transposed vectors thereof to obtain a state incidence matrix; then, obtaining a state association feature matrix by using the trained first convolution neural network model as a feature extractor; then, obtaining type text description of each electric device in all the electric devices; secondly, respectively training the type text description of each electric device in all the electric devices by a context encoder which contains an embedded layer to obtain a plurality of device text description feature vectors; then, arranging the plurality of device text description feature vectors into a device text description feature matrix according to the device sample dimension; then, the device text description feature matrix and the state association feature matrix are subjected to a trained graph neural network to obtain a state semantic association feature matrix, wherein the graph neural network encodes the device text description feature matrix and the state association feature matrix through learnable neural network parameters to obtain a device text description association feature representation containing irregular state topology information; then, obtaining a classification result by the trained classifier of the state semantic association characteristic matrix, wherein the classification result is used for indicating whether the state combination of all the electric equipment in the building to be monitored is normal or not; and finally, responding to the abnormal state combination of all the electric equipment in the building to be monitored as the classification result, and generating an electric utilization abnormity prompt.

In an embodiment of the application, in the above method for building integrated intelligent power utilization, the training the state association matrix through a first convolutional neural network model as a feature extractor to obtain a state association feature matrix further includes: each layer of the trained first convolutional neural network model as a feature extractor is respectively carried out in the forward transmission of the layer: performing convolution processing on input data to obtain a convolution characteristic diagram; performing mean pooling based on a local feature matrix on the convolution feature map to obtain a pooled feature map; and performing nonlinear activation on the pooled feature map to obtain an activated feature map; and the output of the last layer of the trained first convolution neural network model serving as the feature extractor is the state correlation feature matrix, and the input of the first layer of the trained first convolution neural network model serving as the feature extractor is the state correlation matrix.

In an embodiment of the application, in the above method for building integrated intelligent power utilization, the obtaining a plurality of device text description feature vectors by respectively training the type text description of each of the all power utilization devices with a context encoder including an embedded layer includes: performing word segmentation processing on the type text description of each electric device in all the electric devices so as to convert the type text description of each electric device in all the electric devices into a word sequence consisting of a plurality of words; mapping each word in the sequence of words to a word vector using an embedding layer of the context encoder to obtain a sequence of word vectors; and globally context-based semantic encoding the sequence of word vectors using a converter of the context encoder to obtain the plurality of device text description feature vectors.

In an embodiment of the application, in the above method for building integrated intelligent power utilization, the training the state semantic association feature matrix by using a trained classifier to obtain a classification result, where the classification result is used to indicate whether a state combination of all power utilization devices in a building to be monitored is normal, further including: processing the state semantic relevance feature matrix using the classifier to generate the classification result with a formula:

wherein

is a weight matrix of the fully-connected layer,

a bias matrix representing the fully connected layers.

In an embodiment of the present application, in the above building integrated intelligent power utilization method, the building integrated intelligent power utilization method further includes: jointly training the first convolutional neural network model as a feature extractor, the context encoder comprising an embedding layer, the graph neural network, and the classifier.

In an embodiment of the application, in the above building integrated intelligent power utilization method, the jointly training the first convolutional neural network model as a feature extractor, the context encoder including an embedded layer, the graph neural network, and the classifier includes: acquiring training data, wherein the training data comprises state information of all electric equipment in the building to be monitored at a preset time point and type text description of each electric equipment in all electric equipment; after state information of all electric equipment in the building to be monitored at a preset time point in the training data is arranged into a training state input vector, calculating a product between the training state input vector and a transposed vector thereof to obtain a training state incidence matrix; passing the training state association matrix through the first convolution neural network model as a feature extractor to obtain a training state association feature matrix; respectively enabling the type text description of each electric device in all the electric devices in the training data to pass through the context encoder containing the embedded layer to obtain a plurality of training device text description feature vectors; arranging the plurality of training equipment text description feature vectors into a training equipment text description feature matrix according to the equipment sample dimension; enabling the training equipment text description feature matrix and the training state association feature matrix to pass through the graph neural network to obtain a training state semantic association feature matrix; expanding the training state semantic association feature matrix into classification feature vectors according to row vectors; correcting the classified feature vector by using the weight matrix of the classifier before and after each iteration update to obtain a corrected classified feature vector; passing the corrected classification feature vector through the classifier to obtain a classification loss function value; and jointly training the first convolutional neural network model as a feature extractor, the context encoder comprising an embedded layer, the graph neural network, and the classifier with the classification loss function value as a loss function value and by backpropagation of gradient descent.

In an embodiment of the application, in the above method for building integrated intelligent power utilization, the correcting the classification feature vector by using the weight matrix of the classifier before and after updating in each iteration to obtain a corrected classification feature vector further includes: correcting the classified feature vector by using a weight matrix before and after each iteration update of the classifier according to the following formula to obtain the corrected classified feature vector; wherein the formula is:

wherein

and

which represents the zero norm of the vector,

it is indicated that the sum is by position,

the difference in terms of position is indicated,

it is meant that the matrix multiplication is performed,

Here, it can be understood by those skilled in the art that the specific functions and operations in the above-described building integrated intelligent electricity usage method have been described in detail in the above description of the building integrated intelligent electricity usage system with reference to fig. 1 to 4, and thus, a repetitive description thereof will be omitted.

Claims

1. The utility model provides a building is synthesized intelligence and is used electric system which characterized in that includes:

2. The building integrated intelligent power utilization system according to claim 1, wherein the state association feature extraction unit is further configured to: each layer of the trained first convolutional neural network model as a feature extractor is respectively carried out in the forward transmission of the layer:

performing convolution processing on input data to obtain a convolution characteristic diagram;

performing mean pooling based on a local feature matrix on the convolution feature map to obtain a pooled feature map; and

performing nonlinear activation on the pooled feature map to obtain an activated feature map;

and the output of the last layer of the trained first convolution neural network model serving as the feature extractor is the state correlation feature matrix, and the input of the first layer of the trained first convolution neural network model serving as the feature extractor is the state correlation matrix.

3. The building integrated intelligent power utilization system of claim 2, wherein the device description semantic coding unit comprises:

the word segmentation subunit is used for performing word segmentation processing on the type text description of each electric device in all the electric devices so as to convert the type text description of each electric device in all the electric devices into a word sequence consisting of a plurality of words;

an embedding encoding subunit, configured to map, using an embedding layer of the context encoder, each word in the sequence of words to a word vector to obtain a sequence of word vectors; and

a context encoding subunit configured to perform global context-based semantic encoding on the sequence of word vectors using a converter of the context encoder to obtain the plurality of device text description feature vectors.

4. The building integrated intelligent power utilization system according to claim 3, wherein the power utilization equipment state monitoring result generating unit is further configured to: processing the state semantic relevance feature matrix using the classifier to generate the classification result with a formula:

wherein

is a weight matrix of the fully-connected layer,

a bias matrix representing the fully connected layers.

5. The building integrated intelligent power system according to claim 1, further comprising a training module for jointly training the first convolutional neural network model as a feature extractor, the context encoder including an embedded layer, the graph neural network, and the classifier.

6. The building integrated intelligent power utilization system of claim 5, wherein the training module comprises:

the system comprises a training data acquisition unit, a data processing unit and a data processing unit, wherein the training data acquisition unit is used for acquiring training data, and the training data comprises state information of all electric equipment in the building to be monitored at a preset time point and type text descriptions of all electric equipment in the electric equipment;

the training state association unit is used for arranging the state information of all the electric equipment in the building to be monitored in the training data at a preset time point into a training state input vector, and then calculating the product between the training state input vector and the transposed vector thereof to obtain a training state association matrix;

a training state associated feature extraction unit, configured to pass the training state associated matrix through the first convolutional neural network model serving as the feature extractor to obtain a training state associated feature matrix;

a training device description semantic encoding unit, configured to pass type text descriptions of all the electrical devices in the training data through the context encoder including the embedded layer, respectively, to obtain a plurality of training device text description feature vectors;

the training matrix constructing unit is used for arranging the plurality of training equipment text description feature vectors into a training equipment text description feature matrix according to the equipment sample dimension;

the training graph neural network unit is used for enabling the training equipment text description characteristic matrix and the training state associated characteristic matrix to pass through the graph neural network to obtain a training state semantic associated characteristic matrix;

the matrix expansion unit is used for expanding the training state semantic association feature matrix into classification feature vectors according to row vectors;

the classification vector correction unit is used for correcting the classification feature vector by using the weight matrix of the classifier before and after each iteration update to obtain a corrected classification feature vector;

the classification loss unit is used for enabling the corrected classification characteristic vector to pass through the classifier so as to obtain a classification loss function value; and

a parameter updating unit, configured to jointly train the first convolutional neural network model as the feature extractor, the context encoder including the embedding layer, the graph neural network, and the classifier through back propagation of gradient descent with the classification loss function value as a loss function value.

7. The building integrated intelligent power utilization system of claim 6, wherein the classification vector correction unit is further configured to: correcting the classified feature vector by using a weight matrix before and after each iteration update of the classifier according to the following formula to obtain the corrected classified feature vector;

wherein the formula is:

wherein

and

which represents the zero norm of the vector,

it is indicated that the sum is by position,

the difference in terms of position is indicated,

it is meant that the matrix multiplication is performed,

an exponential operation is represented on the vector of values,the exponential operation of the vector means calculating a natural exponent function value raised to the eigenvalue of each position in the vector.

8. A building comprehensive intelligent power utilization method is characterized by comprising the following steps:

obtaining a state association characteristic matrix by taking the state association matrix as a first convolution neural network model of a characteristic extractor after training;

9. The building integrated intelligent power utilization method according to claim 8, wherein the training of the state correlation matrix through a first convolution neural network model serving as a feature extractor is performed to obtain a state correlation feature matrix, and further comprising: each layer of the trained first convolutional neural network model as a feature extractor is respectively carried out in the forward transmission of the layer:

10. The building integrated intelligent power utilization method according to claim 9, wherein the obtaining a plurality of device text description feature vectors by respectively training the type text descriptions of the electric devices in all the electric devices and using a context coder including an embedded layer comprises:

performing word segmentation processing on the type text description of each electric device in all the electric devices so as to convert the type text description of each electric device in all the electric devices into a word sequence consisting of a plurality of words;

mapping each word in the sequence of words to a word vector using an embedding layer of the context encoder to obtain a sequence of word vectors; and

globally context-based semantic encoding the sequence of word vectors using a converter of the context encoder to obtain the plurality of device text description feature vectors.