Non-intrusive load monitoring method based on V—I trajectory and neural network

A non-intrusive load monitoring method based on a V-I trajectory and a neural network, comprising:

• step 1, collecting voltage, current and power data of a load of an electric household side in real time through a household electricity meter installed in a home or a building, and performing filtering on the collected voltage, current and power data;
• step 2, applying a bilateral sliding window algorithm to the filtered power data, comprising:
• in two adjacent windows with a same number of sampling data points of the filtered power data, removing a maximum value and a minimum value of data in each window, respectively, calculating an average value of the data in each window, comparing a difference between average values of the data in the two adjacent windows with a preset threshold, and determining that the switching event occurs in response to determining that the difference exceeds the preset threshold;
• step 3, continuously detecting the power data of the load, after the power data maintains a same steady state for at least 3 consecutive times, obtaining voltage, current, and power data of the load according to data in a steady state before and after the switching event;
• step 4, obtaining a V-I trajectory through the voltage and current data in the steady state obtained in step 3, and converting the V-I trajectory into an RGB image with a size of 2N*2N, where N is a positive integer, wherein the RGB image is used to reflect shape characteristics of the V-I trajectory, a phase difference of the voltage and current data, and a characteristic information of an active power, power characteristics are represented by an image pixel value, and wherein said converting the V-I trajectory into the RGB image with the size of 2N*2N comprises:
• step 4.1, defining an initial value of each pixel to (0, 0, 0);
• step 4.2, obtaining, according to the voltage and current data at the steady state of the load obtained in step 3, a maximum absolute value Umax of the voltage data and a maximum absolute value Imax of the current data, for high-power loads with an active power greater than or equal to 510 W, setting Umax to 400V and Imax to 20 A;
• and for low-power loads with the active power less than 510 W, dynamically calculating Umax and Imax;
• step 4.3, calculating Δu=Umax/N and Δi=Imax/N:
• step 4.4, calculating
• Xj
• =
• N
• +
• int
• ⁡
• (
• Uj
• Δ
• ⁢
• u
• )
• ,
• Yj=N+int (Ij/Δi), as RGB pixel coordinates to be defined, for each sampling point (Uj, Ij), where 0<js sample, and ‘sample’ refers to a number of sampling points collected within one complete cycle of a voltage or current waveform, based on a nominal line frequency;
• and
• step 4.5, defining a pixel value correspondingly according to an active power value of the load;
• dividing the V-I trajectory into three stages, and then setting pixels in each stage to three different colors, comprising:
• If 0<j<sample/3,
• then a pixel value (Xj, Yj) is defined as (color_value, 0, 0);
• If sample/3<j<2*sample/3,
• then the pixel value (Xj, Yj) is defined as (0, color_value, 0);
• If j>2*sample/3,
• then the pixel value (Xj, Yj) is defined as (0, 0, color_value);
• wherein color_value represents the defined pixel value;
• and
• step 5, normalizing the RGB image obtained in step 4, and inputting the normalized RGB image into a monitoring network implemented using a convolutional neural network (CNN) or an AlexNet model, determining, by the monitoring network, a load monitoring result, and transmitting the load monitoring result via an STM32F7 microcontroller to a user terminal for notifying a user of a current operational status of the load in the home or building,
• wherein the convolutional neural network comprises two convolutional layers, two max pooling layers and three fully connected layers, and is trained using historical operation data of an electrical device and the RGB image constructed based on V-I trajectory features;
• and
• wherein the convolutional neural network is configured to run directly on a microcontroller unit (MCU) having computational resources greater than an STM32F7 microcontroller, and wherein the AlexNet model enables the non-intrusive load monitoring method to be executed on a computer or server.