Electrical equipment non-intrusive load identification method and system, medium and equipment

Abstract

The invention belongs to the field of non-intrusive load identification, and provides a non-intrusive load identification method and system for electrical equipment, a medium and equipment, and the technical scheme is as follows: an event in which the state of the equipment changes is judged through event detection, transient and steady voltage, current and power characteristics of the equipment are extracted, and an equipment state information base is constructed; the state transition probabilities of different devices are counted under natural operation of the devices, a state transition probability matrix is obtained, meanwhile, in order to more accurately obtain the states of the devices, according to transient current and voltage, steady-state current and voltage and power characteristics, similarity between new time and data in a state information base is calculated, a similarity information matrix is constructed, and the state transition probability matrix is calculated. And updating the state transition probability matrix, constructing a neural network training set based on a composite feature vector composed of the equipment state information base, the state transition probability matrix and the similarity information matrix, and training a classification neural network to obtain an equipment category. And complementarity among different characteristics is utilized, so that the load identification capability of the office equipment is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of non-intrusive load identification, and in particular relates to a method, system, medium and equipment for non-intrusive load identification of electrical equipment. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Smart electricity consumption is one of the important links of smart grid and the core content of realizing advanced measurement system and interactive service system. Load monitoring is one of the key technologies to realize smart electricity consumption. Non-intrusive load disaggregation (NILD) uses the total load information at the power entrance to monitor the status and decompose the energy consumption of the user's internal electrical equipment. NILD is conducive to friendly interaction between users and power grid companies, and has guiding significance for peak load shaving and valley filling of power grids and green and energy-saving electricity consumption of users.

[0004] NILM aims to solve the problem of how to obtain the name, operation period, energy consumption and other information of the electrical equipment in the area when only the total voltage, current and other power monitoring data of the area are available. In actual operation, this problem can be concretized as knowing the total voltage, current and other data at the current and past moments, and solving the problem of the name of the equipment in operation at the current moment. One of the commonly used methods is to judge the load status based on event detection, but this method currently has the following problems: (1) During the device identification process, many devices may have many states, and threshold detection may result in misidentification. A device may be misidentified as being turned on, but in fact it is not turned on, but is only in the intermediate state of the electrical device. (2) The limitation of using a single device feature for load identification leads to poor identification accuracy. Summary of the invention

[0005] In order to solve at least one technical problem existing in the above-mentioned background technology, the present invention provides a method, system, medium and equipment for non-intrusive load identification of electrical equipment, which proposes a load identification method based on the fusion of transient and steady-state electrical information and state transition probability matrix, and utilizes the complementarity between different features to improve the load identification capability of office equipment.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: A first aspect of the present invention provides a method for non-intrusive load identification of electrical equipment, comprising the following steps: Acquire high-frequency voltage and current data of electrical equipment that has undergone equipment state changes; Based on the electrical high-frequency voltage and current data, extract the transient and steady-state voltage and current characteristics of each electrical device, calculate the power characteristics of each device based on the transient and steady-state voltage and current characteristics, and build a device status information library by combining the transient and steady-state voltage and current characteristics and power characteristics of each device; Based on the device state information database, construct the state transition probability matrix of the same device and the state transition probability matrix between different devices; Based on the device state information database, the similarity between the state to be judged and each state feature of each device is calculated in turn to form a similarity information matrix; The load identification network training data set is constructed by combining the composite feature vector composed of the equipment state information database, the state transition probability matrix and the similarity information matrix; Training the load identification network based on the load identification network training data set to obtain a trained load identification network; The class of the equipment is obtained by identifying the electrical equipment to be identified based on the trained load identification network.

[0007] Furthermore, the change in the state of the electrical equipment is determined by an event detection method that combines a heuristic threshold triggering method with a CUSUM-based threshold triggering method.

[0008] Furthermore, the transient voltage and current characteristics include the peak-to-peak value, absolute average value and effective value of the transient waveform; the steady-state voltage and current characteristics include the steady-state current effective value, fundamental wave, third harmonic and fifth harmonic of the steady-state waveform.

[0009] Furthermore, the process of constructing the state transition probability matrix of the same device includes: According to the different steady-state characteristics of the equipment, the equipment is classified, and the state transition probability of each equipment is calculated according to the transient and steady-state voltage and current characteristics and power characteristics of each electrical equipment. The state transition probability matrix of each equipment is calculated according to the state transition probability of each equipment; The power consumption mode of the load equipment in the actual environment is analyzed, and the state transition probability matrix of each device is updated to obtain the optimized state transition probability matrix.

[0010] Furthermore, the process of constructing the state transition probability matrix between different devices includes: The state transition probability between different devices is calculated by counting the transfer frequencies between different devices; By counting the previous state transition probabilities between different states of each device, the state transition probability matrix from each state of each device to each state of other devices is obtained; The state transition probability matrix between all devices is constructed by combining each state of each device to each state of other devices; The specific probability of each device state is determined based on the similarity matrix, and the state transition probability matrix between all devices is updated to obtain the state transition probability matrix between all devices in the actual situation.

[0011] Furthermore, when constructing the similarity information matrix, the similarity of each state feature is calculated using the Euclidean distance formula.

[0012] Furthermore, the method also includes recording the time, weather and season information while obtaining the transient and steady-state current, voltage and power characteristics of the equipment, and normalizing the collected time, weather and season data and adding them to the load identification network training data set.

[0013] A second aspect of the present invention provides a non-intrusive load identification system for electrical equipment, comprising: A data acquisition module, used to obtain high-frequency voltage and current data of electrical equipment that has undergone equipment state changes; A state information library construction module is used to extract the transient and steady-state voltage and current characteristics of each electrical device based on the electrical high-frequency voltage and current data, calculate the power characteristics of each device according to the transient and steady-state voltage and current characteristics, and construct a device state information library by combining the transient and steady-state voltage and current characteristics and power characteristics of each device; A state transition probability matrix construction module is used to construct a state transition probability matrix of the same device and a state transition probability matrix between different devices based on a device state information library; A similarity information matrix construction module is used to calculate the similarity between the state to be judged and each state feature of each state of each device in turn based on the device state information library to form a similarity information matrix; A load identification network training data set construction module is used to construct a load identification network training data set by combining a composite feature vector composed of a device state information base, a state transition probability matrix between the same device and different devices, and a similarity information matrix; A load identification network training module, used for training the load identification network based on the load identification network training data set to obtain a trained load identification network; The load identification module is used to identify the electrical equipment to be identified based on the trained load identification network to obtain the type of equipment.

[0014] A third aspect of the present invention provides a computer-readable storage medium.

[0015] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps in the above-mentioned method for non-intrusive load identification of electrical equipment.

[0016] A fourth aspect of the present invention provides a computer device.

[0017] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the above-mentioned method for non-intrusive load identification of electrical equipment are implemented.

[0018] Compared with the prior art, the present invention has the following beneficial effects: 1. In order to address the limitation of using a single device feature for load identification, which leads to poor identification accuracy, the present invention determines the event of device state change through event detection, extracts the transient and steady-state voltage, current and power characteristics of the device, builds a device state information library, and counts the state transition probabilities of different devices under natural operation of the device to obtain a state transition probability matrix. By utilizing the complementarity between different features, the load identification capability of office equipment is improved.

[0019] 2. The present invention overcomes the problem of misidentification in threshold detection during event detection. Through the state transition probability matrix in the state information library, the start and stop states of office equipment can be more accurately determined, and the decomposition efficiency is also improved to a certain extent.

[0020] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0022] Figure 1 It is a flow chart of a non-intrusive load identification method for electrical equipment provided by an embodiment of the present invention; Figure 2 is a flow chart of event detection completed by a threshold triggering method provided by an embodiment of the present invention; Figure 3 This is a flow chart of a CUSUM determining a device switching task provided by an embodiment of the present invention; Figure 4 is a schematic diagram of a sliding window processing high-frequency current sequence provided by an embodiment of the present invention; Figure 5 Schematic diagram of the effective value of low-frequency current processed by a moving window provided in an embodiment of the present invention; Figure 6 It is the operation process of the finite state load provided by the embodiment of the present invention; Figure 7 This is the printer operation process provided by the embodiment of the present invention; Figure 8This is the actual working state of the printer provided by the embodiment of the present invention. DETAILED DESCRIPTION

[0023] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0024] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0025] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0026] As mentioned in the background technology, the current technical problem of judging the load state based on event detection is proposed in the present invention. A non-invasive load detection method that integrates transient and steady-state electrical characteristics with a state transition probability matrix is ​​proposed. The method first determines the event of equipment state change through event detection, then collects the high-frequency voltage and current information of the equipment and extracts the transient and steady-state voltage, current and power characteristics of the equipment, constructs an equipment state information library, and statistically analyzes the state transition probabilities of different equipment under the natural operation of the equipment to obtain a state transition probability matrix. At the same time, in order to more accurately judge the state corresponding to the specific equipment, according to the transient current and voltage peak-to-peak value, absolute average value, effective value and steady-state current and voltage effective value, fundamental wave, third harmonic, fifth harmonic and power characteristics, by calculating the similarity between the new time and the data in the state information library, a similarity information matrix is ​​constructed, the timestamp, weather, season are collected and normalized, and the state transition probability matrix is ​​converted into one-dimensional data. The transient and steady-state voltage, current, power characteristics and the normalized timestamp, season, weather and the flattened and reduced-dimensional state transition probability matrix are used as the input of the classification neural network, and the output is the equipment category, and the classification neural network is trained.

[0027] Embodiment 1 like Figure 1 As shown, this embodiment provides a non-intrusive load identification method for electrical equipment, comprising the following steps: Step 1: Determine the device status change event based on an event detection method that combines heuristic threshold triggering and CUSUM-based threshold triggering; specifically, include: Step 101: Figure 2 As shown, the time detection process based on heuristic threshold triggering includes: Step 1011, calculate the effective value of the current in each cycle in the sliding window. The calculation formula of the effective value of the current under high-frequency sampling is shown in formula (1): (1), In the formula, is the effective value of the current in the current cycle, is the number of sampling points for each power frequency cycle, For the The instantaneous value of the current at each sampling point.

[0028] Step 1012: Calculate the difference between the effective value of the current cycle at the latest moment and the effective value of the current cycle at the previous moment as shown in formula (2): (2), In the formula is the effective value of the periodic current at the current moment, is the effective value of the current cycle at the previous moment, It is the difference between the effective value of the current cycle at the current moment and the effective value of the current cycle at the previous moment.

[0029] Step 1013: Setting the abnormal detection threshold of the sliding window current effective value I th Since the operating conditions of power equipment in the power grid will affect current fluctuations, and the dispatch of the power system or other reasons will cause periodic or random changes in the power grid, in order to avoid false detection caused by grid current fluctuations, the anomaly detection threshold I th It cannot be set too low, otherwise it will cause the event detection algorithm to report frequently, affecting the stable operation of the non-intrusive load monitoring system. I th It cannot be set too high, otherwise the switching event will not be detected normally. I th The minimum value setting refers to the grid fluctuation range setting under no-load and load conditions in the experimental environment. I th The maximum value range setting should not be greater than the effective value of the current when the minimum power equipment used in the experiment is switched on.

[0030] Step 1014: Determine the monitoring window Whether the abnormality detection conditions are met, if The absolute value of I th and If it is greater than 0, it is considered that a device is connected. The absolute value of I th and If it is less than 0, it is considered that a device is shut down.

[0031] Step 1015: After the abnormal detection conditions are met, the data in the monitoring window is further analyzed and processed, and the required data is sent to the host computer or database for subsequent processing.

[0032] Step 102: Figure 3 As shown, the process of the threshold-triggered event detection method based on CUSUM includes: Step 1021, determine the target value: first, it is necessary to determine the expected value, which can be a known reference value or a historical average value. In this embodiment, the effective value of the current at the previous moment is selected as the expected value.

[0033] Step 1022, calculate the difference value: for each cycle of the effective value of the current, calculate the difference between it and the expected value. If the difference value is positive, the task of detecting whether there is a device connected will be executed. If the difference value is negative, the task of detecting whether there is a device exit will be executed.

[0034] Step 1023, cumulative difference value: add the difference value calculated in each sliding process to the accumulated difference value in the previous sliding process to obtain a cumulative sum, which represents the cumulative deviation degree of the entire sequence.

[0035] Step 1024, setting a detection threshold: The set threshold is used to determine when the accumulation exceeds the threshold.

[0036] Step 1025, event triggering: when the cumulative sum exceeds the set threshold, it is considered that equipment switching has occurred.

[0037] In this embodiment, the threshold trigger method is combined with the CUSUM method to improve the accuracy of event detection according to the characteristics of electrical equipment in practical applications. CUSUM is a statistical method for detecting change points in a data sequence. Like the threshold trigger method, it has the ability to monitor and detect data changes in real time. It monitors the change or drift of the data by calculating the cumulative sum of the data sequence, and triggers an event when the cumulative sum exceeds a preset threshold. Therefore, it can sensitively monitor small current changes and capture subtle trend changes.

[0038] Step 2: Extract high-frequency current and voltage waveform data of electrical equipment based on the sliding window; In this embodiment, the step of extracting high-frequency current and voltage waveform data of the electrical equipment based on the sliding window includes: Step 201: data collection; First, use a suitable sensor or current transmitter to collect the current signal to be monitored. The sensor will convert the current signal into a voltage signal or a digital signal for further processing and analysis.

[0039] Step 202: Set the window size; Before starting monitoring, determine the window size and sliding step size used to analyze the current signal. The window size determines the time range of the analyzed data after each update, such as 20 milliseconds or 0.1 seconds. The sliding step size determines the time interval for window updates. The selection of the two parameters depends on the characteristics of the monitored object and the actual situation on site.

[0040] Step 203, sliding window; After determining the window size and sliding step, the monitoring system will start sliding the window on the current signal. Each sliding window will cover the current data for a period of time, and subsequent analysis and processing will be carried out on this basis.

[0041] Step 204, calculation within the window: within each window, the current signal undergoes a series of preprocessing steps such as filtering and denoising to reduce interference or highlight the characteristics of the signal, and then the current signal within the window is subjected to feature analysis, such as extracting frequency domain features, time domain features (such as mean, variance), time-frequency features, etc. These features are helpful for subsequent abnormality detection and analysis.

[0042] Step 205: After feature extraction, an algorithm or model is used to perform abnormality detection on the current signal in the window. By comparing with a preset threshold or model, it is detected whether the current in the window meets the abnormality detection conditions. If it is detected that the current signal in the window meets the abnormality detection conditions, such as exceeding the preset threshold or not meeting the normal mode, corresponding measures are taken to ensure the normal operation of the system.

[0043] The real-time current sequence is monitored using a sliding window, where the monitoring window width of high-frequency data is 5 power frequency cycles and the update frequency is 50Hz. This method is used to monitor the loop current in order to capture the transient waveform when the device status changes. The schematic diagram of the sliding window processing current sequence is shown in Figure 4 As shown, Figure 4 middle t is the current moment of monitoring, For an update cycle, is the number of sampling points for each current cycle, and the arrow is the time update direction. The monitoring window width of low-frequency data is also 5, the unit is the current effective value of 1 cycle, and the update frequency is 10Hz. The schematic diagram of sliding window processing low-frequency current effective value is shown in Figure 5 As shown, Figure 5 middle For the current moment, For one update cycle, the effective current value cache of one cycle is obtained every 5 power frequency cycles.

[0044] Through the sliding window current monitoring method, a fixed-length window is set on the current signal, and it slides forward continuously over time to achieve continuous monitoring and real-time analysis of the current in the window. The sliding window current monitoring method has the advantages of strong real-time performance, high flexibility and high efficiency.

[0045] Step 3: Extract the transient and steady-state voltage and current characteristics of each device based on the collected high-frequency voltage and current data of the device, calculate the power characteristics of each device based on the transient and steady-state voltage and current information, and build a device status information library based on the transient and steady-state voltage and current characteristics and power characteristics of each device; Since the high-frequency sequence captured during the monitoring process contains more than just the target device, it is usually the superposition of the working current and voltage of multiple devices. Therefore, it is necessary to extract the transient separation waveform and the steady-state separation waveform respectively to extract the transient characteristics and steady-state characteristics, and divide the complete device switching process into pre-steady state, transient and post-steady state, where the transient separation waveform and steady state The separation formulas of the separated waveforms are shown in equations (3) and (4): (3), (4), in, It is the transient process waveform; is the pre-steady-state waveform; is the post-steady-state waveform; The number of sampling points for each cycle.

[0046] Each device has unique characteristic signals in the transient and steady-state processes when it is turned on and off. Therefore, it is particularly important to extract effective load characteristics from the characteristic curve of the device. According to the above waveform separation formulas (3) and (4), the waveform captured by event detection can be decomposed into transient separation waveforms and steady-state separation waveforms.

[0047] In order to fully reflect the waveform characteristics when the load is switched, the present embodiment selects seven characteristics, namely, peak-to-peak value, absolute average value, effective value, steady-state current, effective value of voltage, fundamental wave, third harmonic and fifth harmonic, of which three characteristics are used as transient characteristics for equipment identification, and four characteristics are used as steady-state characteristics for equipment identification.

[0048] The characteristic symbols and meanings of the selected current characteristic quantities are shown in Table 1: Table 1 Types, symbols, names and meanings of current characteristic quantities

[0049] The characteristic symbols and meanings of the selected voltage characteristic quantities are shown in Table 2: Table 2 Types, symbols, names and meanings of voltage characteristic quantities

[0050] In this embodiment, active power and reactive power characteristics of each device are calculated based on transient current characteristics, steady-state current characteristics, voltage transient characteristics, and voltage steady-state characteristics of each device; The power characteristics used are the active power and reactive power of the device in steady-state operation. Since the state information library requires the sampling signal to be high-frequency voltage and current data, it is necessary to obtain the active power and reactive power of the device in steady-state operation based on the high-frequency discrete sampling data. FFT is used to extract the power characteristics of the device. The number of sampling points of the voltage and current signals in one steady-state cycle of this embodiment is . Let the FFT The frequency domain signal is Then its modulus and phase angle are The effective values ​​and phase angles of each voltage and current are: (5), (6), (7), Where: and Respectively fundamental voltage and current effective value; and They are Subharmonic voltage and current effective value; and They are Subharmonic voltage and current phase angle; and They are voltage and current signals respectively.

[0051] The time domain voltage and current signals can be expressed as: (8), (9), Where: is the angular velocity.

[0052] From this, we can conclude that the average active power and reactive power are: (10), (11), Considering that the equipment's operating time, working weather, and season may have an impact on the equipment's transient and steady-state characteristics, the equipment's transient and steady-state current, voltage, and power characteristics are collected while recording the time, weather, and season at the time; the collected time, weather, and season data are normalized to facilitate subsequent use as input for the classification neural network.

[0053] The normalization of time includes: (1) converting the 24 hours of a day into minutes, which is 1440 minutes in total. (2) converting the current time into minutes. (3) dividing the current time in minutes by the total number of minutes 1440 to obtain a decimal between 0 and 1, which is the normalized time.

[0054] Seasonal normalization includes mapping the four seasons (spring, summer, autumn, and winter) to values ​​between 0 and 1. For the four seasons, the intervals are evenly distributed between 0 and 1. The numerical mapping can be achieved by simple linear interpolation, mapping spring to 0, summer to 0.33, autumn to 0.66, and winter to 1.

[0055] Normalization of weather includes: first, the weather conditions that will be encountered are arranged in order: sunny, cloudy, light rain, heavy rain, snow, fog, wind. There are a total of 7 different weather states, which can be evenly distributed between 0 and 1. The specific values ​​are: Sunny day: 0 / (7-1)=0; cloudy day: 1 / (7-1)≈0.167; light rain: 2 / (7-1)≈0.333; heavy rain: 3 / (7-1)≈0.500; snowy day: 4 / (7-1)≈0.667; foggy day: 5 / (7-1)≈0.833; windy day: 6 / (7-1)=1.

[0056] The transient and steady-state voltage and current characteristics of different devices are further extracted from the high-frequency voltage and current data of the acquisition equipment. The transient current and voltage peak-to-peak value, absolute average value, and effective value are used as the transient characteristics of each device, and the steady-state current and voltage effective value, fundamental wave, third harmonic, and fifth harmonic are used as the steady-state characteristics of each device. The power characteristics of each device are calculated based on the transient and steady-state voltage and current information. The state information library of each type of equipment (open, closed, intermediate state, etc.) is composed of state characteristics such as transient and steady-state current and voltage characteristics and power characteristics.

[0057] Step 4: Based on the device state information library, construct the state transition probability matrix of the same device and the state transition probability matrix between different devices; specifically including: Step 401: Construction of the same device state transition probability matrix Taking household appliances as an example, according to the different steady-state characteristics of the equipment, the equipment types can be divided into on / off type equipment, finite state type equipment, and continuously variable state type equipment. Among them, the first two types of equipment have no essential difference, both have limited working states, and the load characteristics of each state are relatively stable, which can be used to describe most household appliances. The operation process of the finite state load is as follows Figure 6 shown.

[0058] Assume that the state set of electrical appliances is , represents the number of electrical appliance operating states, then its operating process can be described as a discrete time series ,in Indicates time The state of the electrical appliance is an unobservable hidden state. is the running time. The output current, voltage and power of the electrical appliances can be output as observation values, and the observation sequence is .

[0059] use Indicates that device 1 is in the slave state To status The state transition probability and , .

[0060] The specific state transition probability of each device is obtained through statistical calculation based on the collected transient steady-state voltage, current and power.

[0061] The state transition probability matrix of device 1 can be obtained from the state transition probability of device 1: (12), in, Represents the number of operating states of electrical appliances.

[0062] Combined with the actual operation, the power consumption mode of the load equipment in the actual environment is analyzed, and the state transition probability matrix of each device is optimized. Consider the printer load device in the office environment, which contains three states: open, print, and close. The operation process of the printer is as follows: Figure 7 As shown in the figure, in actual situations, the user's behavior determines the operation process and does not exist, that is, the new state transition diagram and state transition probability matrix are updated as follows Figure 8 By analyzing the power consumption mode of the load device in accordance with the above combined with the actual environment, the optimized state transition probability matrix of other load devices can be obtained.

[0063] Step 402: Construction of different device state transition probability matrices Under the natural working state of each device, the state transition probability between different devices is collected, and it is calculated by counting the transition frequency between different devices; the calculation formula for the state transition probability from device A to device B is as follows: (13), In the formula The frequency of device A switching on to device B switching on. The frequency of device A's transition from the on state to other states .

[0064] By counting the previous state transition probabilities between different states of each device, the state transition probability matrix from each state of each device to each state of other devices is obtained. For example, there are three devices A, B, and C. Assume that device A has two states (on state, off state), represented by A1 and A2 respectively. Device B has three states (on state, standby state, off state), represented by B1, B2, and B3 respectively. Device C has two states (on state, off state), represented by C1 and C2 respectively.

[0065] Then the state transition probability matrix between the three devices can be expressed as: (14), In the process of observing and counting the transition frequencies between the states of each device, it is known that the previous state belongs to a state of device A, and the current state belongs to a state of device B. However, in actual operation, it is not completely certain which device state the previous state belongs to. However, the specific probability of which device state it belongs to can be determined by the similarity information matrix calculated in step 5. Assume that the probability of judging that it is the state of device A is 0.94, the probability of judging that it is the state of device B is 0.04, and the probability of judging that it is the state of device C is 0.02. In actual situations, the probability matrix of the transition of the three device states can be expressed as: (15).

[0066] Step 5: Construct a similarity information matrix based on the device status information database; In order to more accurately judge the status of the equipment, the concept of similarity information matrix is ​​introduced. The similarity information matrix is ​​constructed by calculating the similarity of each transient and steady-state electrical information.

[0067] In this embodiment, the similarity of each state feature is calculated using the Euclidean distance formula, and the state feature data is first standardized before calculating the Euclidean distance. The standardized formula is as follows: (16), The Euclidean distance formula is as follows: (17), The similarities between the state to be judged and each state feature of each device are calculated in sequence to form a similarity information matrix.

[0068] Taking the current transient steady-state information as an example, assuming there are four devices, the similarity information matrix is ​​constructed as shown in Table 3: Table 3 Similarity information matrix construction table

[0069] Step 6: Construct a neural network training set based on the composite feature vector composed of the device state information library, the state transition probability matrix and the similarity information matrix; including: Step 601: Obtain the transient current characteristics and power characteristics of the device from the high-frequency voltage and current information of the device and normalize the collected timestamp, season, and weather; Step 602: Count the state transition frequencies of the same device or between different devices, then calculate the state transition probability and construct a state transition probability matrix; Step 603: Based on the peak-to-peak value, absolute average value, effective value of transient current and voltage and the effective value of steady-state current and voltage, fundamental wave, third harmonic, fifth harmonic and power characteristics, a similarity information matrix is ​​constructed by calculating the similarity between the new event and the state information library data.

[0070] Step 604: Combine the one-dimensional array consisting of the transient and steady-state electrical information, the normalized timestamp, the weather, and the season and the one-dimensional array consisting of the flattened and reduced-dimensional state transition probability matrix into a one-dimensional array, which is a composite feature.

[0071] Among them, when the composite feature vectors are combined, the state transition probability matrix and the similarity information matrix are flattened by rows to convert the two-dimensional data into one-dimensional data.

[0072] The collected data set contains the device voltage and current waveforms at a sampling frequency of 30kHz. There are 11 different types of household appliances, with a total of 235 independent devices and 1074 sets of data in the data set. Each set of data contains the transient and steady-state process data when the equipment is switched on and off as required by this embodiment. 20 of the real sampling data of each type of equipment are randomly selected to form a test set, and the test set has a total of 220 device sampling data. The remaining sampling data constitute the training set, and the training set has a total of 1705 device sampling data, which includes real device sampling and synthetic data. The input of the training set is the status information of the device, the status information of the previous device, the probability matrix of the transfer of the previous device state to the current device after flattening and dimensionality reduction, and the normalized timestamp, weather, and season.

[0073] Step 7: training the load identification network based on the neural network training set to obtain a trained load identification network; Since load identification is a multi-classification problem, the activation function of the output layer of the network is Softmax, and the output is a Dimensional vector , the sum of all elements in the vector is 1, where the value of each element The representative predicted value is The cross entropy loss function is used, and the training goal is to minimize the loss function, that is, when the input is given to the network When the characteristics of the class device are The value of is the largest.

[0074] The optimization algorithm uses an adaptive moment estimation optimization algorithm, takes composite features as the input of the classification network, and uses equipment categories as labels to train the classification network for load identification.

[0075] Step 8: Based on the trained load identification network, the equipment is identified to obtain the type of equipment and the corresponding status.

[0076] Embodiment 2 This embodiment provides a non-intrusive load identification system for electrical equipment, including: A data acquisition module, used to obtain high-frequency voltage and current data of electrical equipment that has undergone equipment state changes; A state information library construction module is used to extract the transient and steady-state voltage and current characteristics of each electrical device based on the electrical high-frequency voltage and current data, calculate the power characteristics of each device according to the transient and steady-state voltage and current characteristics, and construct a device state information library by combining the transient and steady-state voltage and current characteristics and power characteristics of each device; A state transition probability matrix construction module is used to construct a state transition probability matrix of the same device and a state transition probability matrix between different devices based on a device state information library; A similarity information matrix construction module is used to calculate the similarity between the state to be judged and each state feature of each state of each device in turn based on the device state information library to form a similarity information matrix; A load identification network training data set construction module is used to construct a load identification network training data set by combining a composite feature vector composed of a device state information base, a state transition probability matrix between the same device and different devices, and a similarity information matrix; A load identification network training module, used for training the load identification network based on the load identification network training data set to obtain a trained load identification network; The load identification module is used to identify the electrical equipment to be identified based on the trained load identification network to obtain the type of equipment.

[0077] Embodiment 3 This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps in the above-mentioned method for non-intrusive load identification of electrical equipment are implemented.

[0078] Embodiment 4 This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the above-mentioned method for non-intrusive load identification of electrical equipment are implemented.

[0079] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A non-intrusive load identification method for electrical equipment, characterized in that: The steps include: Acquire high-frequency voltage and current data of electrical equipment that has undergone equipment state changes; Based on the electrical high-frequency voltage and current data, extract the transient and steady-state voltage and current characteristics of each electrical device, calculate the power characteristics of each device based on the transient and steady-state voltage and current characteristics, and build a device status information library by combining the transient and steady-state voltage and current characteristics and power characteristics of each device; Based on the device state information database, construct the state transition probability matrix of the same device and the state transition probability matrix between different devices; Based on the device state information database, the similarity between the state to be judged and each state feature of each device is calculated in turn to form a similarity information matrix; The load identification network training data set is constructed by combining the composite feature vector composed of the equipment state information database, the state transition probability matrix and the similarity information matrix; Training the load identification network based on the load identification network training data set to obtain a trained load identification network; The class of the equipment is obtained by identifying the electrical equipment to be identified based on the trained load identification network.

2. A non-intrusive load identification method for electrical equipment according to claim 1, characterized in that: The state change of electrical equipment is judged by an event detection method that combines heuristic threshold triggering and CUSUM-based threshold triggering.

3. A non-intrusive load identification method for electrical equipment according to claim 1, characterized in that: The transient voltage and current characteristics include the peak-to-peak value, absolute average value and effective value of the transient waveform; the steady-state voltage and current characteristics include the steady-state current effective value, fundamental wave, third harmonic and fifth harmonic of the steady-state waveform.

4. A non-intrusive load identification method for electrical equipment according to claim 1, characterized in that: The process of constructing the state transition probability matrix of the same device includes: According to the different steady-state characteristics of the equipment, the equipment is classified, and the state transition probability of each equipment is calculated according to the transient and steady-state voltage and current characteristics and power characteristics of each electrical equipment. The state transition probability matrix of each equipment is calculated according to the state transition probability of each equipment; The power consumption mode of the load equipment in the actual environment is analyzed, and the state transition probability matrix of each device is updated to obtain the optimized state transition probability matrix.

5. The non-intrusive load identification method for electrical equipment according to claim 1, characterized in that: The process of constructing the state transition probability matrix between different devices includes: The state transition probability between different devices is calculated by counting the transfer frequencies between different devices; By counting the previous state transition probabilities between different states of each device, the state transition probability matrix from each state of each device to each state of other devices is obtained; The state transition probability matrix between all devices is constructed by combining each state of each device to each state of other devices; The specific probability of each device state is determined based on the similarity matrix, and the state transition probability matrix between all devices is updated to obtain the state transition probability matrix between all devices in the actual situation.

6. A non-intrusive load identification method for electrical equipment according to claim 1, characterized in that: When constructing the similarity information matrix, the similarity of each state feature is calculated using the Euclidean distance formula.

7. A non-intrusive load identification method for electrical equipment according to claim 1, characterized in that: The method also includes recording the time, weather and season information while obtaining the transient and steady-state current, voltage and power characteristics of the equipment, and normalizing the collected time, weather and season data and adding them to the load identification network training data set.

8. A non-intrusive load identification system for electrical equipment, characterized in that: include: A data acquisition module, used to obtain high-frequency voltage and current data of electrical equipment that has undergone equipment state changes; A state information library construction module is used to extract the transient and steady-state voltage and current characteristics of each electrical device based on the electrical high-frequency voltage and current data, calculate the power characteristics of each device according to the transient and steady-state voltage and current characteristics, and construct a device state information library by combining the transient and steady-state voltage and current characteristics and power characteristics of each device; A state transition probability matrix construction module is used to construct a state transition probability matrix of the same device and a state transition probability matrix between different devices based on a device state information library; A similarity information matrix construction module is used to calculate the similarity between the state to be judged and each state feature of each state of each device in turn based on the device state information library to form a similarity information matrix; A load identification network training data set construction module is used to construct a load identification network training data set by combining a composite feature vector composed of a device state information base, a state transition probability matrix between the same device and different devices, and a similarity information matrix; A load identification network training module, used for training the load identification network based on the load identification network training data set to obtain a trained load identification network; The load identification module is used to identify the electrical equipment to be identified based on the trained load identification network to obtain the type of the equipment.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in a method for non-intrusive load identification of electrical equipment as described in any one of claims 1 to 7 are implemented.

10. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of a method for non-intrusive load identification of electrical equipment as described in any one of claims 1-7 are implemented.

Images