TWI802245B - Power consumption analysis system and power consumption analysis method based on non-intrusive appliance load monitoring - Google Patents

Abstract

A power consumption analysis system and a power consumption analysis method based on non-intrusive appliance load monitoring are provided. The power consumption analysis method includes: obtaining training data through a data extraction process; performing pre-processing on the training data; training a deep learning model with the pre-processed training data, and using the deep learning model that meets a training completion condition as a power consumption analysis model; obtaining total electricity consumption data of a target field with the data extraction process, and inputting the total electricity consumption data into the electricity consumption analysis model to predict operating powers of target electrical appliances; obtaining operating data of the target electrical appliances and determining whether the operating powers are within operating power ranges to determine operating states of the target electrical appliances.

Description

Power consumption analysis system and power consumption analysis method based on non-intrusive equipment load monitoring

The present invention relates to an analysis system and an analysis method, in particular to an electricity analysis system and an electricity analysis method based on non-invasive equipment load monitoring.

Nonintrusive appliance load monitoring (NIALM) system is a set of systems that can help users understand the detailed power consumption of the home. It analyzes the voltage and total current changes of a certain circuit in the home, and from these changes The status of individual appliances can be identified in the system, so the power consumption of each appliance can be recorded through a single meter for user reference.

The low-cost non-intrusive equipment load monitoring system is an indispensable technology for a future smart home or smart building. In recent years, the vigorous development of computer, communication and storage technologies has made it feasible to collect a large number of commonly used non-intrusive load monitoring databases.

The existing NIALM electrical analysis models generally have insufficient analytical performance, especially the recognition of electrical fingerprints and the F-Score of the model, and the training of the model is easily affected by multi-dimensional data.

The technical problem to be solved by the present invention is to provide a power consumption analysis system and power consumption analysis method based on non-intrusive equipment load monitoring for the deficiencies of the prior art.

In order to solve the above-mentioned technical problems, one of the technical solutions adopted by the present invention is to provide a method for power consumption analysis based on non-intrusive equipment load monitoring, which includes: using a data acquisition program to obtain multiple A plurality of electric power consumption data of the target electric appliance and a total power consumption historical data are used as a training data; a pre-processing procedure is performed on the training data, including synchronizing the sampling time of these electrical power consumption data and the total power consumption historical data And carry out a maximum and minimum regularization (MinMaxScaler) processing to the training data; train a deep learning model with the pre-processed training data, and use the deep learning model that reaches the training completion condition as a power analysis model, wherein, The deep learning model at least includes an input layer, a feature extraction dimension reduction layer, a coding layer and an electrical power prediction layer; a total electricity consumption history data of the target field is obtained by the data acquisition program, and input to the Using an electrical analysis model to predict multiple operating powers of the target electrical appliances respectively; obtaining multiple pieces of operating data of the target electrical appliances from an electrical appliance operating data database, wherein the pieces of operating data include those corresponding to the target electrical appliances a plurality of operating power ranges; and judging whether the operating powers are within the operating power ranges, so as to determine the operating conditions of the target electrical appliances.

In order to solve the above-mentioned technical problems, another technical solution adopted by the present invention is to provide a power analysis system based on non-intrusive equipment load monitoring, which includes multiple target electrical appliances and computing devices. The target electrical appliances are set in the target field, and the target electrical appliances are connected to the general power consumption circuit. A computing device, including a memory and a processor, wherein the processor is configured to: execute a data acquisition program to respectively obtain a plurality of electric power consumption data and a total power consumption history data of multiple electric appliances in a target field as A training data; a pre-processing procedure is performed on the training data, which includes synchronizing the sampling time of the electrical power consumption data of these electrical appliances and the total power consumption historical data and performing a MinMaxScaler processing on the training data; A deep learning model is trained with the pre-processed training data, and the deep learning model that meets the training completion condition is used as an electrical analysis model, wherein the deep learning model includes at least an input layer and a feature extraction dimension reduction layer, a coding layer, and an electrical power prediction layer; a total electricity consumption history data of the target field is obtained by the data acquisition program data, and input the power analysis model to predict multiple operating powers of these electrical appliances respectively; obtain multiple operating data of these electrical appliances from an electrical appliance operating data database, wherein, the operating data includes corresponding Multiple operating power ranges of the electrical appliances; and judging whether the operating powers are within the operating power ranges, so as to judge the operating conditions of the electrical appliances, and generate an analysis result of power consumption.

One of the beneficial effects of the present invention is that the present invention proposes a NIALM electrical analysis system and method based on a deep learning model, which has better electrical analysis performance and better electrical fingerprints than the existing NIALM electrical analysis model The recognition degree is not easily affected by data and multi-dimensional data in terms of model training, and the F-Score of the obtained power analysis model is better.

In order to further understand the features and technical content of the present invention, please refer to the following detailed description and drawings related to the present invention. However, the provided drawings are only for reference and description, and are not intended to limit the present invention.

1: Electrolytic analysis system

10: target field

11:Digital power meter

12: Computing device

100, 102, 104: target electrical appliances

106: total power circuit

120: Processor

122: memory

124: Network interface

126: Input and output interface

128: busbar

3: Deep learning model

30: Input layer

32: Feature extraction dimensionality reduction layer

34: Coding layer

36: flat layer

38: Electric Power Prediction Layer

320, 380, 382: fully connected layers

341, 342, 343, 344, 346: one-dimensional convolution layer

345, 347, 381: discard layers

D1: Computer readable instructions

D2: Training data

D3: Pre-processing procedure

D4: Deep Learning Model

D5: Historical data of total electricity consumption

D6: Home appliance operation data database

FIG. 1 is a flow chart of an electrolysis method according to an embodiment of the present invention.

Fig. 2 is a block diagram of a NIALM-based electrolysis system according to the present invention.

Fig. 3 is a schematic diagram of a deep learning model according to an embodiment of the present invention.

The following is a description of the implementation of the "electricity analysis system and method based on non-invasive equipment load monitoring" disclosed by the present invention through specific specific examples. Those skilled in the art can understand from the content disclosed in this specification Advantages and effects of the present invention. The present invention can be implemented or applied through other different specific embodiments, and various modifications and changes can be made to the details in this specification based on different viewpoints and applications without departing from the concept of the present invention. In addition, the drawings of the present invention are only for simple illustration, and are not drawn according to the actual size, which is stated in advance. the following The embodiments will further describe the relevant technical content of the present invention in detail, but the disclosed content is not intended to limit the protection scope of the present invention. In addition, the term "or" used herein may include any one or a combination of more of the associated listed items depending on the actual situation.

FIG. 1 is a flow chart of an electrolysis method according to an embodiment of the present invention. Referring to FIG. 1 , the first embodiment of the present invention provides a non-intrusive appliance load monitoring (NIALM)-based power consumption analysis method, which generally includes data pre-processing, feature extraction, and power disassembly, etc. step.

Among them, the data pre-processing is to normalize the load data to be disassembled or identified according to the environmental voltage characteristics or regional information. Feature extraction is to extract feature data that can be used to identify, classify or dismantle loads, such as load real work, reactive work, current harmonics, switching transient characteristics, etc. Electric energy dismantling is to establish a power consumption analysis model based on the total load characteristics and the power consumption of the target electrical appliances by executing algorithms, so as to analyze the power consumption of each electrical appliance at different time points.

Please refer further to FIG. 2 , which is a block diagram of a NIALM-based electrolysis system according to the present invention. In an embodiment of the present invention, the power consumption analysis method shown in FIG. 1 is applicable to the power consumption analysis system 1 shown in FIG. The target electrical appliances 100 , 102 and 104 are arranged in the target field 10 and connected to the general power consumption circuit 106 .

In some embodiments, the power consumption analysis system 1 further includes a digital power meter 11, which is connected between the utility power and the total power consumption circuit 106 in the target field 10, so that the target electrical appliances 100, 102 and 104 can be connected by the utility power During power supply operation, in addition to capturing the power consumption of the total power consumption circuit 106, various power parameters are also captured. In addition, the digital power meter 11 can be, for example, an existing smart meter, which can transmit the captured power and power parameters to (or directly connect to) the computing device 12 through the network for analysis of power consumption.

2, the computing device 12 may include a processor 120, a memory 122, a network The interface 124 and the input-output interface 126 , and the above components can communicate through the bus 128 . However, the above-mentioned embodiments are just examples, and the present invention is not limited to using the bus bar 128 for communication.

The processor 120 is electrically coupled to the memory 122 and is configured to access computer-readable instructions D1 from the memory 122 to control components in the computing device 12 to perform functions of the computing device 12 .

The memory 122 may include any storage device for storing data, such as a hard disk, a solid state drive or other storage devices for storing data, but is not limited thereto. The memory 122 is configured to store at least a plurality of computer-readable instructions D1, training data D2, pre-processing programs D3, deep learning models D4, total electricity consumption history data D5, and home appliance operation data database D6. The memory 122 may also include random access memory (random access memory; RAM), read only memory (read only memory; ROM), and flash memory to store data or instructions under the control of the processor 120 .

In this embodiment, the network interface 124 can be configured to perform network access under the control of the processor 120 , and the network interface 124 can be, for example, a wired or wireless network card. For example, the network interface 124 can be connected to the network to access the power and power parameters captured by the digital power meter 11 .

The I/O interface 126 is operable by the user to communicate with the processor 120 for data input and output. The input-output interface 126 can be connected with input or output devices such as a keyboard, a mouse, and a monitor.

Therefore, taking the above architecture as an example, the processor 120 can execute an operating system, and use the RAM in the memory 122 as a temporary data storage medium to provide an appropriate operating environment for executing the power analysis method of the present invention. In more detail, the electroanalysis method can be implemented, for example, by using a computer program to control various components of the computing device 12 . The computer program can be stored in a non-transitory computer-readable recording medium, such as read-only memory, flash memory, floppy disk, hard disk, optical disk, flash drive, magnetic tape, database accessible by the network Or those who are familiar with the art can easily think of a computer-readable recording medium with the same function.

As shown in Figure 1, the electrolytic method may include the following steps:

Step S10: Execute the data acquisition procedure to obtain the electrical consumption data and the total electrical consumption history data of the target electrical appliances in the target field respectively as training data.

In this step, the data acquisition program utilizes the digital power meter 11 to acquire a plurality of data parameters of the total power consumption circuit 106 of the target field 10 within a period of time as the total power consumption historical data. These data parameters include: current first 10 odd-frequency harmonic characteristics, voltage form factor, current form factor, real power (P) and imaginary power (Q) and other power parameters with a total of 4 to 14 dimensions.

Among them, the effective data training period and data sampling frequency used in the embodiment of the present invention and the control group can refer to the following table 1:

As shown in the above table, the control group is to sample one piece of electricity consumption data every 15 minutes or every minute during the effective data training period of about 20 days, and can include real power (P) and imaginary power (Q). The difference from the control group is that in the embodiment of the present invention, during the effective data training period of about 4 days, one piece of electricity consumption data is sampled every 10 seconds, and can include the aforementioned current first 10 order odd-frequency harmonic characteristics, voltage form factor, current waveform factor, real power (P) and imaginary power (Q), etc.

Step S11: Execute a pre-processing procedure on the training data.

The pre-processing procedure includes synchronizing the sampling time of the electricity consumption data of these electrical appliances and the total electricity consumption history data, and performing a MinMaxScaler processing on the training data D2.

Wherein, a sampling sequence every X seconds of each piece of electric power consumption data of these electric appliances is synchronized with the sampling time of the total electric power consumption historical data, wherein, X is a positive integer and is in the range of 5 to 15 Inside. In a preferred embodiment of the present invention, for example, the sampling sequence every 10 seconds in the power consumption data of the target electrical appliance can be synchronized with the sampling time of the total power consumption history data.

On the other hand, the historical data of total electricity consumption includes the aforementioned characteristics of the first 10 odd-frequency harmonics of current, voltage form factor, current form factor, real power (P) and imaginary power (Q), etc., which correspond to multiple dimensions ( For example, 4 to 14 dimensions), and the maximum and minimum normalization process is for each of these dimensions, the data of the corresponding power parameter is normalized to between 0 and 1 according to its maximum value and minimum value, and the resulting The training data D2 can effectively improve the power analysis performance of the trained model.

Step S12: Train the deep learning model with the pre-processed training data, and use the deep learning model that meets the training completion condition as the power analysis model.

Referring to FIG. 3 , it is a schematic diagram of a deep learning model according to an embodiment of the present invention. As shown in FIG. 3 , the deep learning model 3 is an improved Seq2point model, which at least includes an input layer 30 , a feature extraction dimensionality reduction layer 32 , an encoding layer 34 and an electrical power prediction layer 38 .

The existing Seq2point model is a regression (Regression) calculation method on the CNN network. The algorithm assumes that under the window W of time t, the power of individual electrical appliances at the time point t+w/2 will be equal to the total load Y _{t: t+w} has a strong correlation.

In contrast, in the embodiment of the present invention, the input layer 30 is modified to add multi-dimensional data support, and the feature extraction dimensionality reduction layer 32 is added at the same time to ensure that the subsequent encoding layer 34 can process the dimensionality-reduced data Perform feature extraction. Wherein, the encoding layer 34 may include one-dimensional convolutional layers 341, 342, 343, 344, 346 and Dropout layers 345, 347. Wherein, the activation functions of the one-dimensional convolutional layers 341 , 342 , 343 , 344 , and 346 use the ReLU function, wherein the numbers corresponding to the input and output of each layer represent the dimension of the layer.

The deep learning model 3 also includes a flatten layer 36 connected between the coding layer 34 and the electrical power prediction layer 38 . The feature extraction dimension reduction layer 32 is a fully connected layer 320 , and the output dimension of the fully connected layer 320 can be, for example, 2, which is smaller than the input dimension of 12 .

The appliance power prediction layer 38 includes fully connected layers 380 , 382 and a dropout layer 381 , wherein the fully connected layer 382 is used as an output layer to output the appliance power predicted by the deep learning model 3 .

Next, under the above framework, the deep learning model 3 can be trained. First, random values can be used to initialize the weights and parameters in all filters of the deep learning model 3, and then, the pre-processed training data is used as input, and the probability corresponding to each class is obtained through forward propagation. In the initial stage, since the weights and parameters of the filter are set randomly, the probability of each class obtained is also random.

Then calculate the total error of the output layer (that is, the fully connected layer 382). In order to reduce this error, backpropagation can be used to calculate the gradient of each weight in the network, and the gradient descent method is used to update the values of all filters to minimize the output error. However, the above-mentioned optimization method is only an example, and the present invention is not limited to using the gradient descent method to find the optimal solution.

Afterwards, the weight of the filter can be updated according to the error. If the total output error decreases after the weight is updated, it means that the deep learning model 3 has learned to distinguish specific data by adjusting the weight. Then, the above steps can be repeated until the total output error is minimum, which means that the training completion condition is reached.

It should be noted that in the deep learning model 3, the parameters such as the number of filters, the size of the network and the structure of the network have been determined before training, and will not be changed during the training process, and only the parameters updated during the training process are The value of the filter and the weight of the network.

Step S13: Obtain the total power consumption history data of the target field through the data acquisition program, and input the power consumption analysis model to respectively predict multiple operating powers of the target electrical appliances.

Similarly, the aforementioned data acquisition method can be used to extract the total power consumption history data D5 of the target field 10 within a predetermined period of time to be analyzed through the digital power meter 11 . It should be noted that the retrieved power consumption data and power parameters need to correspond to the data type used in the training data D2. For example, assuming that the data parameters in the total power consumption historical data in the training data D2 use power parameters such as voltage form factor, current form factor, real power (P) and imaginary power (Q), then the total power consumption historical data D5 needs to be These electric parameters are also included.

Step S14: Obtain multiple pieces of operating data of the target electrical appliances from the home appliance operating data database, wherein the pieces of operating data include multiple operating power ranges respectively corresponding to the target electrical appliances. Specifically, in order to improve the power analysis performance, the power analysis method of the present invention takes the specification of the target electrical appliance as a priori condition to correct the output of the F-Score of the power analysis model.

Wherein, the home appliance operating data database D6 may include, for example, the highest operating power and the lowest operating power of the target electrical appliances 100 , 102 , 104 .

Step S15: Determine whether the operating powers are within the operating power ranges, so as to determine the operating conditions of the target electrical appliances.

For example, the step S16 can be entered by using the electroanalysis method: judging whether the predicted operating power (W) is greater than the minimum operating power (L) and smaller than the maximum operating power (H).

In response to the predicted operating power (W) being greater than the minimum operating power (L) and less than the maximum operating power (H), proceed to step S17 by electroanalytic method: determine that the state of the target electrical appliance is on, and its operating power is the predicted The operating power (W).

In response to the predicted operating power (W) being not greater than the minimum operating power (L) and less than the maximum operating power (H), proceed to step S18 by electroanalytic method: judging whether the maximum operating power (H) is less than or equal to the predicted operating power Power (W). If so, then enter step S19 with electrolytic method: judge The state of the target electrical appliance is on, and its operating power is the highest operating power (H). If not, proceed to step S20 by electroanalysis: determine that the state of the target electrical appliance is off, and its operating power is the minimum operating power (L).

Please refer to Table 2 below, which shows the comparison of the qualitative prediction performance (F-Score) of the operating state of household appliances between the power consumption analysis model provided by the present invention and the existing NIALM power consumption analysis model.

It should be noted that, in the above table, the power analysis model provided by the present invention uses P, Q, current harmonics, current form factor and voltage form factor in the training features, so its qualitative prediction performance (F-Score) It reaches 87.6%, which is obviously better than the existing NIALM electrolytic model in the table.

[Advantageous Effects of Embodiment]

One of the beneficial effects of the present invention is that the present invention proposes a NIALM electrical analysis system and method based on a deep learning model, which has better electrical analysis performance and better electrical fingerprints than the existing NIALM electrical analysis model Recognition, in model training, not easily affected Influenced by data and multi-dimensional data, and the F-Score of the obtained power analysis model is better.

The content disclosed above is only a preferred feasible embodiment of the present invention, and does not therefore limit the scope of the patent application of the present invention. Therefore, all equivalent technical changes made by using the description and drawings of the present invention are included in the application of the present invention. within the scope of the patent.

Claims (8)

A power consumption analysis method based on non-intrusive equipment load monitoring, which includes: using a data acquisition program to respectively obtain a plurality of electric power consumption data of multiple target electric appliances in a target field and a total power consumption history data as a Training data; a pre-processing procedure is performed on the training data, which includes: synchronizing the sampling time of these electrical power consumption data and the total power consumption historical data; and performing a maximum and minimum normalization (MinMaxScaler) process on the training data ; train a deep learning model with the pre-processed training data, and use the deep learning model that reaches the training completion condition as an electrical analysis model, wherein the deep learning model includes at least an input layer, a feature extraction and reduction Dimensional layer, a coding layer and an electrical power prediction layer; use the data acquisition program to obtain the total power consumption history data of the target field, and input the power consumption analysis model to respectively predict multiple operations of the target electrical appliances power; respectively obtain multiple pieces of operation data of the target electrical appliances from an electrical appliance operation data database, wherein the pieces of operation data include multiple operation power ranges respectively corresponding to the target appliances; and determine whether the operation power is within Within the operating power ranges, the operating conditions of the target electrical appliances are judged to generate a power consumption analysis result, wherein the deep learning model further includes a flatten layer connected to the coding layer and the electrical appliance power prediction Between layers, the feature extraction dimensionality reduction layer is a first fully connected layer, and an output dimension of the first fully connected layer is smaller than an input dimension; wherein, the encoding layer includes multiple one-dimensional convolutional layers and multiple a first dropout layer; and Wherein, the electrical power prediction layer includes a second fully connected layer, a second dropout layer and a third fully connected layer. The electricity consumption analysis method as described in claim 1, wherein the data acquisition program includes using a digital power meter to extract a plurality of data parameters of a total power consumption circuit in the target field as the total power consumption historical data , these data parameters include: current first 10 order odd frequency harmonic characteristics, voltage form factor, current form factor, real power and imaginary power. The power consumption analysis method as described in claim 2, wherein the total power consumption historical data includes multiple dimension data corresponding to multiple dimensions, and the maximum and minimum normalization processing is for each of these dimensions, and the corresponding The dimension data of is normalized between 0 and 1 according to the maximum and minimum values in the dimension data. The power consumption analysis method as described in claim item 1, wherein the step of synchronizing the sampling time of these electrical power consumption data and the total power consumption historical data includes: A sampling sequence every X seconds is synchronized with the sampling time of the total power consumption historical data, wherein X is a positive integer ranging from 5 to 15. A power analysis system based on non-intrusive equipment load monitoring, which includes: a plurality of target electrical appliances, set in a target field, and these target electrical appliances are connected to a general power consumption circuit; a digital power meter, connected to In the total power consumption circuit; a computing device connected to the digital power meter, the computing device includes a memory and a processor, wherein the processor is configured to: execute a data acquisition program to respectively obtain a target A plurality of electric power consumption data and a total power consumption history data of multiple electric appliances in the field are used as a training data; a preprocessing procedure is performed on the training data, which includes: synchronizing the power consumption data of these electric appliances with the total power consumption data The sampling time of electrical historical data; and Performing a maximum and minimum normalization (MinMaxScaler) process on the training data; training a deep learning model with the pre-processed training data, and using the deep learning model that meets the training completion condition as a power analysis model, wherein the The deep learning model at least includes an input layer, a feature extraction dimensionality reduction layer, a coding layer, and an electrical power prediction layer; the total power consumption history data of the target field is obtained by the data acquisition program, and the user input An electroanalytic model is used to predict multiple operating powers of these electrical appliances respectively; multiple pieces of operating data of these electrical appliances are respectively obtained from an electrical appliance operating data database, wherein the pieces of operating data include multiple operating powers corresponding to these electrical appliances power range; and judging whether the operating powers are within the operating power ranges to determine the operating conditions of the electrical appliances to generate a power analysis result, wherein the deep learning model also includes a flat (flatten) layer, Connected between the coding layer and the electrical power prediction layer, the feature extraction dimension reduction layer includes a first fully connected layer, and an output dimension of the first fully connected layer is smaller than an input dimension; wherein, the coding layer It includes a plurality of one-dimensional convolution layers and a plurality of first dropout layers; and wherein, the electrical power prediction layer includes a second fully connected layer, a second dropout layer and a third fully connected layer. The power consumption analysis system as described in claim 5, further includes a digital power meter connected to the total power consumption circuit, and the data acquisition program includes configuring the digital power meter to obtain multiple data of the total power consumption circuit A data parameter is used as the total electricity consumption history data, and these data parameters include: current first 10 order odd-frequency harmonic characteristics, voltage form factor, current form factor, real power and reactive power. The power consumption analysis system as described in claim 6, wherein the total power consumption history data includes multiple dimension data corresponding to multiple dimensions, and the maximum and minimum normalization processing is for each of these dimensions, and the corresponding The dimension data of is normalized between 0 and 1 according to the maximum and minimum values in the dimension data. The power consumption analysis system as described in claim item 5, wherein the step of synchronizing the sampling time of these electrical power consumption data and the total power consumption historical data includes the A sampling sequence every X seconds is synchronized with the sampling time of the total electricity consumption historical data, where X is a positive integer ranging from 5 to 15.

Images