A nonlinear load state recognition method based on time-frequency analysis

Abstract

The application relates to a nonlinear load identification method based on time-frequency analysis, constructs a load identification architecture using frequency domain features, and aims to solve the problem that transient characteristics cannot be effectively applied in nonlinear load identification in non-intrusive load monitoring. The application mainly comprises: a current feature extraction technology based on fast Fourier transform and Hilbert-Huang transform, and a classification method based on an extreme learning machine as the main body, and the method for extracting transient load characteristics and identifying nonlinear loads is verified by training and testing on a public data set.

Description

Technical Field

[0001] This invention belongs to the field of non-intrusive power load monitoring technology and relates to a nonlinear load state identification method based on time-frequency analysis. Background Technology

[0002] Electricity has become the most important end-use energy source in modern society. To address the global energy shortage, achieve social decarbonization, and improve the reliability of power supply and the efficiency of electricity utilization on the user side, research and practice on smart grids are being conducted worldwide. On the demand side of the grid, the widespread application of smart electronic devices and high-energy-consuming equipment has significantly increased the proportion of nonlinear loads at the grid's demand end. How to obtain the load characteristics of nonlinear loads from total user energy consumption data and accurately identify them is a key focus for the power sector.

[0003] Non-intrusive Load Monitoring (NILM) technology requires only one sensor with communication capabilities installed at the power supply inlet of the electrical load. By analyzing the total load power consumption data, it can obtain power consumption information for each (major) appliance within the user's premises. NILM technology is of great significance for improving the observability of demand-side electricity. Its deployment can contribute to the construction of a clean, low-carbon, safe, and efficient energy system and the achievement of "dual carbon" goals, promote pollution and carbon reduction in key industries, and expand and deepen the application of electricity data in social services.

[0004] Since its inception in the late 20th century, the Nonlinear Load Imaging (NILM) technology has received widespread attention. Currently, various methods within this field are applied to nonlinear load identification, among which time-frequency domain analysis theory has become a crucial tool for NILM technology in achieving nonlinear load identification. Specifically, the Fast Fourier Transform (FFT) and Hilbert-Huang Transform (HFT) theories, matured in the field of speech signal processing, extract target load features from high-frequency load signals to support subsequent load identification. Due to the diverse types of user-side loads, the widespread occurrence of load aliasing in complex scenarios, and the significant interference of noise, accurately acquiring and identifying the characteristics of the target nonlinear load is quite challenging.

[0005] To address the aforementioned issues, this invention makes improvements based on time-frequency analysis theory, proposes a complete nonlinear load identification architecture based on the extreme learning machine model, and uses public datasets to train and test the classification model, verifying that the method of this invention can accurately identify complex nonlinear loads on the user side. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide a nonlinear load state identification method based on time-frequency analysis, which can accurately identify complex nonlinear loads on the user side.

[0007] The technical problem solved by this invention is achieved through the following technical solution:

[0008] A nonlinear load state identification method based on time-frequency analysis, characterized by the following steps:

[0009] Step 1: Extract the target time series using the time-frequency feature extraction module, and extract features from the extracted series. The extracted features include time-domain features and frequency-domain features.

[0010] Step 2: Based on the time-domain and frequency-domain features extracted in Step 1, establish a multi-feature fusion classification learning model: The learning model is based on the extreme learning machine model, which completes the identification of load by fusing multiple input features.

[0011] Step 3: Use the public dataset for training and adjustment: Adjust the sliding window length based on the extracted time-frequency feature performance and load recognition performance to obtain the optimal test results.

[0012] Furthermore, the method for extracting the target time series using the time-frequency feature extraction module and extracting features from the extracted series, including time-domain features and frequency-domain features, is as follows:

[0013] The initial sliding window length is defined as N. The total current signal of the user is truncated according to this window length to obtain the target current time series:

[0014] {i(n)|n=1,2,...,N-1,N} (1)

[0015] Where i(n) represents the instantaneous current value at each time point of the current time series sample;

[0016] For the current time series given by equation (1), the frequency domain features of the target current signal are extracted using a variant of the discrete Fourier method, the short-time Fourier, which is a frequency domain analysis method. The frequency domain features mentioned in this invention include the frequency domain amplitude FA and the frequency domain phase FP.

[0017] To obtain the frequency domain characteristics, the original current signal is first processed by short-time Fourier transform, as shown in equation (2):

[0018]

[0019] In the formula, F(m) is the frequency domain signal of the original current signal i(n) of the user's total port collected;

[0020] Based on equation (2), the frequency domain amplitude FA and frequency domain phase FP can be calculated, as shown in equation (3):

[0021]

[0022] In the formula, ReF(m) and ImF(m) are the real and imaginary parts of F(m), respectively;

[0023] For the extraction of time-domain features, the Hilbert-Huang transform is used to process the acquired raw current signal i(n) of the user's main port in order to extract time-domain features;

[0024] The time-domain features described in this invention include instantaneous amplitude IA, instantaneous phase IP, and instantaneous frequency IF.

[0025] Furthermore, the method for extracting the temporal features is as follows:

[0026] The Hilbert-Huang transform result HT(i(n)) of the original current signal i(n) is shown in equation (4):

[0027]

[0028] In the formula, u is the integral operator.

[0029] After processing by equation (3), the original current signal i(t) is analyzed and characterized, and the signal form is calculated according to equation (5):

[0030]

[0031] Where j represents the imaginary unit;

[0032] Then, the instantaneous amplitude IA and instantaneous phase IP are extracted as two time-domain features, and the calculation formula is given by equation (6):

[0033]

[0034] After obtaining the instantaneous amplitude IA and instantaneous phase IP, calculate according to equation (7):

[0035]

[0036] In the formula, This means taking the first derivative of the obtained instantaneous phase IP(t) in time. Since IP(t) is a discrete time series in form, the derivative is the difference between the two data points.

[0037] Furthermore, the classification learning model based on the Extreme Learning Machine is an artificial neural network, which consists of an output layer, an input layer, and multiple hidden layers. In this network structure, the weight values ​​and bias values ​​between the input layer and the hidden layers are randomly generated and not iteratively modified. After establishing the learning model, feature fusion is performed based on the time-frequency domain features output in step 1, and subsequent model training begins.

[0038] Output value t of output layer node k With input layer node value x i The mathematical relationship is given by equation (8):

[0039]

[0040] In the formula, ω j and b j These are the weight matrix and bias vector from the input layer to the j-th hidden layer node in the network structure, respectively, g(ω) j ·x i +b j ) refers to the activation function, β j This refers to the weight value from the j-th hidden layer node to the output layer node, and L refers to the number of hidden layer nodes.

[0041] When ω is determined j and b j Next, the output matrix H of the hidden layer can be calculated, and the calculation equation is given in equation (9):

[0042] Hβ=T (9)

[0043] Where T refers to the output matrix of the output layer; the solution to equation (9) is given by equation (10):

[0044]

[0045] In the formula, Let H be the pseudo-inverse matrix;

[0046] The five time-frequency domain features obtained in step 1, namely frequency domain amplitude FA, frequency domain phase FP, instantaneous amplitude IA, instantaneous phase IP and instantaneous frequency IF, are fused to complete the subsequent model training. The specific steps are as follows: adopt a fully connected network and use the results obtained as shown in equation (10) to train the established extreme learning machine model.

[0047] Furthermore, the method for adjusting the sliding window length based on the extracted time-frequency feature effect and load identification effect to obtain the optimal test effect is as follows: a target accuracy τ is preset. If the accuracy of the current model is insufficient for the target accuracy, then the length of τ is increased by Δt before and after the center of the current window length, and the accuracy of the model is tested again. If the target accuracy τ is still not achieved, the window length is increased repeatedly until the accuracy reaches the target.

[0048] The advantages and beneficial effects of this invention are as follows:

[0049] This nonlinear load state identification method based on time-frequency analysis extracts time-frequency domain features from the current signal collected from the user's main inlet using short-time Fourier transform and Hilbert-Huang transform. It then constructs an extreme learning machine model and performs supervised training using a public dataset, ultimately achieving the goal of identifying nonlinear loads on a non-intrusive platform. Attached Figure Description

[0050] Figure 1 This is the Extreme Learning Machine network structure described in this invention;

[0051] Figure 2 This is a flowchart illustrating the process of increasing window length as described in this invention. Detailed Implementation

[0052] The present invention will be further described in detail below through specific embodiments. The following embodiments are merely descriptive and not limiting, and should not be used to limit the scope of protection of the present invention.

[0053] An innovative method for identifying nonlinear load states based on time-frequency analysis includes the following steps:

[0054] Step 1: Based on the initially defined sliding window length of N, extract the time series of the target current from the total current collected at the user's main port. The resulting target current time series is defined as (11):

[0055] {i(n)|n=1,2,...,N-1,N} (11)

[0056] For the current time series given by equation (11), the short-time Fourier transform, a variant of the discrete Fourier transform commonly used in the field, is used to extract frequency domain features from the target current signal. The frequency domain features mentioned in this invention refer to frequency-domain amplitude (FA) and frequency-domain phase (FP).

[0057] To obtain the frequency domain characteristics, the original current signal is first processed by short-time Fourier transform, as shown in equation (12):

[0058]

[0059] In the formula, F(m) is the frequency domain signal of the original current signal i(n) of the user's total port. Based on equation (12), the frequency domain amplitude FA and frequency domain phase FP can be calculated, as shown in equation (13):

[0060]

[0061] The Hilbert-Huang transform is used to extract the instantaneous amplitude (IA), instantaneous phase (IP), and instantaneous frequency (IF) in the time domain. The Hilbert-Huang transform result HT(i(n)) of the original current signal i(n) is shown in Equation (14):

[0062]

[0063] After processing by equation (13), the original current signal i(t) is analyzed and characterized, and its form can be expressed as in equation (15):

[0064]

[0065] Then, the instantaneous amplitude IA and instantaneous phase IP are extracted as two time-domain features, and the calculation formula is as shown in equation (16):

[0066]

[0067] After obtaining the instantaneous amplitude IA and instantaneous phase IP, calculate according to equation (17):

[0068]

[0069] Step 2: Construct an Extreme Learning Machine network, consisting of one output layer, one input layer, and one hidden layer. The network structure diagram is shown below. Figure 1 The network structure is given. The weights and biases between the input and hidden layers are randomly generated and not iteratively modified. After establishing the learning model, feature fusion is performed based on the multi-time-frequency domain features output in step 1, and subsequent model training begins.

[0070] Output value t of output layer node k With input layer node value x i The mathematical relationship is given by equation (18):

[0071]

[0072] In the formula, ω jand b j These are the weight matrix and bias vector from the input layer to the j-th hidden layer node in the network structure, respectively, g(ω) j ·x i +b j ) refers to the activation function, β j This refers to the weight value from the j-th hidden layer node to the output layer node, and L refers to the number of hidden layer nodes.

[0073] When ω is determined j and b j Next, the output matrix H of the hidden layer is calculated, and the calculation equation is given in equation (19):

[0074] Hβ=T (19)

[0075] Here, T refers to the output matrix of the output layer. The solution to equation (19) is given by equation (20):

[0076]

[0077] The five time-frequency domain features obtained in step 1, namely frequency domain amplitude FA, frequency domain phase FP, instantaneous amplitude IA, instantaneous phase IP, and instantaneous frequency IF, are fused to complete the subsequent model training. The Extreme Learning Machine (ELM) structure adopts a fully connected network, and the established ELM model is trained using the results obtained as shown in Equation (20).

[0078] Step 3: To improve the performance of the learning model established in Step 2, the sliding window length N, a key parameter, should be increased bidirectionally. A target accuracy τ is preset. If the current model's accuracy is insufficient to meet the target, the target window length is increased. The model accuracy is then repeatedly tested by increasing the window length by Δt forward and backward from the current window center until the target accuracy is achieved. The flowchart for increasing the window length is shown below. Figure 2 Provided.

[0079] Examples of research materials:

[0080] The test used total power data from 56 households selected from the publicly available PLAID dataset containing high-frequency data. The training and testing sets for the model were randomly allocated in a 3:1 ratio. Evaluation metrics employed included state identification accuracy metrics commonly used in the field of non-intrusive load monitoring, such as accuracy, precision, recall, and F1 score.

[0081] The key parameters set in the embodiment are as follows: the initial sliding window length N is 100, the number of hidden layer units L in the extreme learning machine model is 10, the window length is increased by Δt by 20 on one side, and the target accuracy τ is 93%.

[0082] Training and testing were conducted based on the above experimental configuration, and the test results are as follows:

[0083] Electrical appliances accuracy Accuracy Recall rate F1 score air conditioner 0.99 0.95 0.95 0.95 Fluorescent lamps 0.99 0.91 1.00 0.95 laptop 0.96 0.77 0.85 0.81 refrigerator 0.99 1.00 0.85 0.92 Hair dryer 1.00 1.00 1.00 1.00 heater 1.00 1.00 1.00 1.00 fan 0.97 0.81 0.85 0.83 Incandescent lamp 1.00 1.00 1.00 1.00 Micro-wave oven 0.99 0.95 1.00 0.98 Vacuum cleaner 1.00 1.00 1.00 1.00 washing machine 0.98 1.00 0.85 0.92 average 0.99 0.95 0.94 0.94

[0084] The results show that the identification method described in this invention significantly outperforms the non-invasive load monitoring field in identifying various electrical appliances, and performs exceptionally well in some common electrical appliances, such as hair dryers, heaters, incandescent lamps, and vacuum cleaners.

[0085] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0086] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0087] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0088] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0089] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A nonlinear load state identification method based on time-frequency analysis, characterized in that: Includes the following steps: Step 1: Extract the target time series using the time-frequency feature extraction module, and extract features from the extracted series. The extracted features include time-domain features and frequency-domain features. Step 2: Based on the time-domain and frequency-domain features extracted in Step 1, establish a multi-feature fusion classification learning model. This learning model is based on the extreme learning machine model and completes the identification of load by fusing multiple input features. Step 3, use public datasets for training and adjustment: adjust the sliding window length based on the extracted time-frequency feature performance and load recognition performance to obtain the optimal test results; The method for extracting features from the target time series using a time-frequency feature extraction module, and then extracting features from the extracted sequence, including time-domain features and frequency-domain features, is as follows: The initial sliding window length is defined as N. The total current signal of the user is truncated according to this window length to obtain the target current time series: {i(n)|n=1,2,…,N-1,N} (1) Where i(n) represents the instantaneous current value at each time point of the current time series sample; For the current time series given by equation (1), the frequency domain features of the target current signal are extracted using a variant of the discrete Fourier method—the short-time Fourier—based on frequency domain analysis. These frequency domain features include the frequency domain amplitude FA and the frequency domain phase FP. To obtain the frequency domain characteristics, the original current signal is first processed by short-time Fourier transform, as shown in equation (2): In the formula, F(m) is the frequency domain signal of the original current signal i(n) of the user's total port collected; Based on equation (2), the frequency domain amplitude FA and frequency domain phase FP can be calculated, as shown in equation (3): In the formula, ReF(m) and ImF(m) are the real and imaginary parts of F(m), respectively; For the extraction of time-domain features, the Hilbert-Huang transform is used to process the acquired user port raw current signal i(n) to extract time-domain features; the time-domain features include instantaneous amplitude IA, instantaneous phase IP, and instantaneous frequency IF.

2. The nonlinear load state identification method based on time-frequency analysis according to claim 1, characterized in that: The method for extracting the temporal features is as follows: The Hilbert-Huang transform result HT(i(n)) of the original current signal i(n) is shown in Equation (4): In the formula, u is the integral operator; After processing by equation (3), the original current signal i(t) is analyzed and characterized, and the signal form is calculated according to equation (5): Where j represents the imaginary unit; Then, the instantaneous amplitude IA and instantaneous phase IP are extracted as two time-domain features, and the calculation formula is given by equation (6): After obtaining the instantaneous amplitude IA and instantaneous phase IP, calculate according to equation (7): In the formula, This means taking the first derivative of the obtained instantaneous phase IP(t) in time. Since IP(t) is a discrete time series in form, the derivative is the difference between the two data points.

3. The nonlinear load state identification method based on time-frequency analysis according to claim 1, characterized in that: The classification learning model based on Extreme Learning Machine (ELM) is an artificial neural network, consisting of an output layer, an input layer, and multiple hidden layers. In this network structure, the weights and biases between the input and hidden layers are randomly generated and not iteratively modified. After establishing the learning model, feature fusion is performed based on the time-frequency domain features output in step 1, and subsequent model training begins. Output value t of the output layer node k With input layer node value x i The mathematical relationship is given by equation (8): In the formula, ω j and b j These are the weight matrix and bias vector from the input layer to the j-th hidden layer node in the network structure, respectively, g(ω) j ·x i +b j ) refers to the activation function, β j This refers to the weight value from the j-th hidden layer node to the output layer node, and L refers to the number of hidden layer nodes; When ω is determined j and b j Next, the output matrix H of the hidden layer can be calculated, and the calculation equation is given in equation (9): Hβ=T (9) Where T refers to the output matrix of the output layer; the solution to equation (9) is given by equation (10): In the formula, Let H be the pseudo-inverse matrix; The five time-frequency domain features obtained in step 1, namely frequency domain amplitude FA, frequency domain phase FP, instantaneous amplitude IA, instantaneous phase IP and instantaneous frequency IF, are fused to complete the subsequent model training. The specific steps are as follows: adopt a fully connected network and use the results obtained as shown in equation (10) to train the established extreme learning machine model.

4. The nonlinear load state identification method based on time-frequency analysis according to claim 1, characterized in that: The method for adjusting the sliding window length based on the extracted time-frequency feature effect and load identification effect to obtain the optimal test effect is as follows: a target accuracy τ is preset. If the accuracy of the current model is insufficient for the target accuracy, then the length of τ is increased by Δt before and after the center of the current window length, and the accuracy of the model is tested again. If the target accuracy τ is still not achieved, the window length is increased again until the accuracy reaches the target.

Images