A DTW combined bi-lstm power load online identification method
By combining DTW with Bi-LSTM for online power load identification, and integrating cloud platform and sensing devices, we have achieved accurate online monitoring of the operating status of electrical equipment. This solves the problem of high-precision identification when multiple electrical appliances are running simultaneously, and improves the accuracy and efficiency of load identification.
Patent Information
- Application Number
- CN202311709040.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-12-13
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2043-12-13
AI Technical Summary
Existing load identification methods struggle to achieve high-precision identification when multiple electrical appliances are running simultaneously. Furthermore, deep learning models require a large amount of sample data, making it difficult to balance monitoring accuracy and efficiency. Moreover, existing methods are ill-suited to address issues such as overlapping load features and reduced discriminative power.
An online power load identification method combining DTW and Bi-LSTM is adopted. The method combines real-time monitoring of the operating status of electrical equipment through a cloud platform, uses the CUSUM algorithm to detect events, and uses short-time Fourier transform and stacked sparse autoencoder for feature extraction and dimensionality reduction. A DTW algorithm template library is constructed and combined with the Bi-LSTM model for load identification. The final identification result is output by combining the probabilistic models of DTW and Bi-LSTM.
It enables precise online monitoring of the operating status of electrical equipment, improves the accuracy and efficiency of load identification, effectively handles complex operating conditions, and reduces the need for sample data.
Smart Images

Figure CN117972384B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of non-intrusive load monitoring technology, and in particular to an online power load identification method using DTW combined with Bi-LSTM. Background Technology
[0002] In the face of global energy shortages, the efficient use of electricity is a key means to alleviate the current energy problem. Currently, households account for 40% of global energy consumption, and their carbon emissions are increasing daily. Optimizing household energy consumption and reducing energy waste are essential for achieving energy conservation and emission reduction. To fully tap the potential for energy conservation and emission reduction on the user side, the key lies in monitoring the energy consumption of various loads in households. This cannot be limited to the total household load data recorded by traditional meters. Power grid companies and users can adjust power distribution and consumption plans separately based on granular electricity monitoring.
[0003] The above goals can be achieved through the current construction of "smart grids". Non-intrusive load monitoring, as an important means and method for improving the construction of "smart grids", is essential for achieving "granular" electricity consumption monitoring. However, the existing electrical equipment is diverse and generates a large amount of data, making it difficult to achieve high-precision load identification. In particular, when multiple loads are running simultaneously, problems such as overlapping features and reduced distinguishability can easily occur, compromising the accuracy of load identification and further increasing the difficulty of load identification.
[0004] Current load identification methods struggle to extract highly discriminative features in the face of the increasing number of household appliances. Furthermore, existing load features suffer from overlap, poor effectiveness, and extraction difficulties. Therefore, most methods employ deep learning to train models by extracting features (such as active power, active-reactive power curves, VI trajectories, and current harmonics) to achieve load identification, significantly improving overall recognition performance. However, training deep learning network models requires a large amount of sample data, often resulting in a trade-off between accuracy and efficiency in load monitoring. Additionally, current methods rarely address the complex situation of multiple loads operating simultaneously, which may place higher demands on the performance of load identification models.
[0005] Therefore, proposing a DTW combined with Bi-LSTM online power load identification method to solve the difficulties of existing technologies is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention
[0006] In view of this, the present invention provides a method for online identification of power load using DTW combined with Bi-LSTM, which combines a cloud platform to monitor the operating status of electrical equipment in real time, thereby achieving accurate online identification of power load.
[0007] To achieve the above objectives, the present invention adopts the following technical solution: a method for online identification of power load using DTW combined with Bi-LSTM, comprising the following steps: S1. Data Acquisition Steps: Collect aggregated power load data from users; S2, Event Detection Steps: Perform event detection on the aggregated power load data collected in S1 using the CUSUM algorithm to obtain unknown electrical appliance event signals; S3, Feature Extraction Step: Use the short-time Fourier transform and the unknown event signal obtained in S2 of the stacked sparse autoencoder to perform feature extraction and feature dimensionality reduction, and obtain the load characteristics of the unknown electrical appliance after dimensionality reduction. S4. Model training steps: The feature extraction and feature dimensionality reduction methods in S3 are used to characterize the event signals of known electrical appliances, and the characteristic known electrical appliance event signals are used to train the Bi-LSTM model. S5. Load identification step: Use the DTW algorithm and the Bi-LSTM model trained in S4 to identify the load of unknown electrical appliances, and combine the identification results of the two through a probability model to obtain the final identification probability vector. S6. Application steps: The remote monitoring device realizes online monitoring of the operating status of electrical equipment based on the final identification probability vector obtained in S5.
[0008] The above methods are optional. The specific content of S1 is as follows: The overall energy consumption information of the user side is obtained through sensing devices and uploaded to the cloud platform. The overall energy consumption information includes: active power signal and current signal.
[0009] The above methods are optional. The specific content of S2 is as follows: build a CUSUM algorithm model on the cloud platform, perform event detection on the active power sequence, determine the operating status of electrical appliances, extract the current signal based on the operating status of electrical appliances, and obtain the current signal of unknown electrical appliances, that is, the event signal of unknown electrical appliances.
[0010] The above methods are optional. The calculation formula for the CUSUM algorithm is: (1) in, , These represent the cumulative sum of load input and load cut-out events, respectively. , Let be the forward and reverse offsets of the current sequence point after removing the noise, respectively, denoted as . and , It is the average value of a sampling sequence of a certain length before the change point. This represents the noise value present in the sampled sequence.
[0011] The above methods are optional. The formula for calculating a periodic current signal is: (2) in, The first current sequence of load events One value, The first steady-state current before the event occurs One sampling point, The first steady-state current after the event occurs One sampling point, is the length of the current sequence.
[0012] The above method, optionally, has the following specific content in S3: S301. Obtain the time spectrum diagram corresponding to the unknown electrical event signal in S2 through STFT, and construct the numerical matrix of the time spectrum diagram as load characteristics to distinguish different electrical appliances; S302. Use SSAE to perform feature dimensionality reduction on the load characteristics to obtain the dimensionality-reduced unknown electrical load characteristics.
[0013] The above method, optionally, has the following specific content in S4: S401. The event signals of all known electrical appliances are characterized using the feature extraction and feature dimensionality reduction methods in S3 to obtain the load feature library and construct the DTW algorithm template library. S402. Divide the load feature data in the DTW algorithm template library to obtain training set and test set, which are used to train Bi-LSTM model.
[0014] The above method is optional; the specific content of S5 is as follows: S501. Calculate the similarity between the load characteristics of the unknown electrical appliance and the feature vector in the DTW algorithm template library. Calculate the shortest local cost metric weighted sum with the feature vector. The shortest local cost metric weighted sum is negatively correlated with the probability of identifying the electrical appliance category. Output the first identification probability vector. First identification probability vector The calculation formula is: (5) in, Representing the One test sample, Indicates the first Number of appliance types; Indicates the first The test sample is the same as the first DTW distance between similar appliances Indicates the first The test sample is the same as the first The smaller the DTW distance between similar electrical appliances, the higher the similarity. S502. Input the load characteristics of the unknown electrical appliance into the trained Bi-LSTM model to output the second identification probability vector; the second identification probability vector is... , Second identification probability vector The calculation formula is: (6) in, Softmax For activation functions; and They represent the first The output values of the forward and reverse layers of each test sample; S503. By combining the probability vectors of the two algorithms using a probability model, the final identified probability vector is: Identify probability vectors The calculation formula is: = (7) The above method, optionally, has the following specific content in S6: The remote monitoring device obtains the final identification probability vector in S5 as the load identification result by connecting to the cloud platform, and monitors the operating status of electrical equipment in real time.
[0015] As can be seen from the above technical solution, compared with the prior art, the present invention provides a method for online identification of power load using DTW combined with Bi-LSTM, which has the following beneficial effects: This invention integrates multiple technologies, including data acquisition, cloud platform, data processing, and remote monitoring equipment, to achieve online monitoring of the operating status of electrical equipment. Specifically, it can be divided into three steps: event detection, feature extraction, and load identification. The various tasks of this method revolve around the collection, detection, processing, and monitoring of power load status information. It has a reasonable structure, strong operability, and can effectively solve the problems faced by current load identification methods. While ensuring the operational efficiency of the identification model, it further improves the model's accuracy in identifying electrical equipment. Furthermore, when constructing the load feature library, it is only necessary to combine various electrical appliances and add the load characteristics of each combination to the feature library to achieve monitoring of various complex operating conditions. Attached Figure Description
[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0017] Figure 1 This is a flowchart of an online power load identification method using DTW combined with Bi-LSTM provided by the present invention; Figure 2 This is a diagram showing the identification results of the load features extracted by the STFT-SSAE algorithm compared with other features in an example test. Figure 3 This is a diagram illustrating the overall recognition performance of the DTW-Bi-LSTM load identification architecture tested in an embodiment. Figure 3 a is the overall recognition result of the PLAID dataset. Figure 3 b shows the overall recognition performance of the UK-DALE dataset; Figure 4 This is a schematic diagram of the structure of a remote power load monitoring device provided by the present invention; Among them, 401 is the processor, 402 is the memory, 403 is the communication bus, 404 is the communication interface, 404-1 is the user interface, and 404-2 is the network interface. Detailed Implementation
[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0019] In this application, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0020] Reference Figure 1 As shown, this invention discloses an online power load identification method using DTW combined with Bi-LSTM, comprising the following steps: S1. Data Acquisition Steps: Collect aggregated power load data from users; S2, Event Detection Steps: Perform event detection on the aggregated power load data collected in S1 using the CUSUM algorithm to obtain unknown electrical appliance event signals; S3, Feature Extraction Step: Use the short-time Fourier transform and the unknown event signal obtained in S2 of the stacked sparse autoencoder to perform feature extraction and feature dimensionality reduction, and obtain the load characteristics of the unknown electrical appliance after dimensionality reduction. S4. Model training steps: The feature extraction and feature dimensionality reduction methods in S3 are used to characterize the event signals of known electrical appliances, and the characteristic known electrical appliance event signals are used to train the Bi-LSTM model. S5. Load identification step: Use the DTW algorithm and the Bi-LSTM model trained in S4 to identify the load of unknown electrical appliances, and combine the identification results of the two through a probability model to obtain the final identification probability vector. S6. Application steps: The remote monitoring device realizes online monitoring of the operating status of electrical equipment based on the final identification probability vector obtained in S5.
[0021] Furthermore, the specific content of S1 is as follows: The overall energy consumption information of the user side is obtained through sensing devices and uploaded to the cloud platform. The overall energy consumption information includes: active power signal and current signal.
[0022] Furthermore, S2 specifically involves: building a CUSUM algorithm model on the cloud platform, performing event detection on the active power sequence, determining the operating status of electrical appliances, extracting current signals based on the operating status of electrical appliances, and obtaining the current signals of unknown electrical appliances, i.e., unknown electrical appliance event signals.
[0023] Furthermore, the calculation formula for the CUSUM algorithm is as follows: (1) in, , These represent the cumulative sum of load input and load cut-out events, respectively. , Let be the forward and reverse offsets of the current sequence point after removing the noise, respectively, denoted as . and , It is the average value of a sampling sequence of a certain length before the change point. This represents the noise value present in the sampled sequence.
[0024] Specifically, sequence points The offset is less than This is often overlooked.
[0025] Furthermore, the formula for calculating the periodic current signal is: (2) in, The first current sequence of load events One value, The first steady-state current before the event occurs One sampling point, The first steady-state current after the event occurs One sampling point, is the length of the current sequence.
[0026] Specifically, after building the CUSUM algorithm model on the cloud platform, event detection is performed on the active power sequence, and small offsets of the sequence data are accumulated. and That is, the difference between the current sequence point and the average value of the sample data under normal operating conditions, when the cumulative sum... or When a change occurs, it can be assumed that an event has occurred. The offset is then accumulated based on the sequence data. When the accumulated sum exceeds a set threshold, it can be considered that an event has occurred. After determining the event occurrence time, the current sequence before each load event can be calculated using the parallel circuit principle. and the current sequence after occurrence By taking the difference, the current sequence of the load event can be obtained. As a preliminary load characteristic.
[0027] Furthermore, the specific details of S3 are as follows: S301. Obtain the time spectrum diagram corresponding to the unknown electrical event signal in S2 through STFT (short-time Fourier transform), and construct the numerical matrix of the time spectrum diagram as load characteristics to distinguish different electrical appliances; S302. Use SSAE (Stacked Sparse Autoencoder) to perform feature dimensionality reduction on the load features to obtain the dimensionality-reduced unknown electrical load features.
[0028] Specifically, the formula for calculating the numerical matrix of the time-frequency spectrum in S301 is as follows: (3) Among them, the current signal is The sequence length is Window width is The center position of the window is The sliding step size is The time window function is The windowing function is , It is the STFT (short-time Fourier transform) function.
[0029] The two-dimensional time spectrum obtained by STFT (short-time Fourier transform) has high resolution. If image-based load type identification is carried out directly, it will place high demands on the hardware. Therefore, SSAE (stacked sparse autoencoder) is used to reduce the dimensionality of the two-dimensional time spectrum to reduce the computational burden on the model. The calculation formula for the SSAE (Stacked Sparse Autoencoder) dimensionality reduction algorithm is as follows: (4) in, The input vector; This is the output vector; Output intermediate vectors for the hidden layer; Represents the activation functions of the encoder and decoder; and These represent the preset weights for the encoding and decoding layers, respectively. and These represent the preset biases for the encoding and decoding layers, respectively.
[0030] The specific process of S301 is as follows: The time-frequency spectrum of the unknown electrical appliance's current signal is obtained through STFT (short-time Fourier transform). For any set of current signals satisfying the Dirichlet constraints... The sequence length is Window width is The sliding step size is The time window function is Place the time window at the beginning of the current sequence and determine the center position of the window. With the set step size Sliding time window, if it exists Then, windowing processing is applied to the signals within the current window, i.e. Perform STFT (short-time Fourier transform) to obtain ; The specific process of S302 is as follows: SSAE (Stacked Sparse Autoencoder) is used to perform feature dimensionality reduction on the two-dimensional time spectrum as the load feature. When using SSAE (Stacked Sparse Autoencoder) for feature learning training, L1 regularization is introduced to make the activation units of the hidden layer of SSAE (Stacked Sparse Autoencoder) as close to zero as possible, which can improve the generalization ability of the feature learning model and extract the features from the two-dimensional time spectrum.
[0031] Furthermore, the specific details of S4 are as follows: S401. The event signals of all known electrical appliances are characterized using the feature extraction and feature dimensionality reduction methods in S3 to obtain the load feature library and construct the DTW algorithm template library. S402. Divide the load feature data in the DTW algorithm template library to obtain training set and test set, which are used to train Bi-LSTM model.
[0032] Specifically, the Bi-LSTM load identification model is trained using the training set data, and the unknown electrical appliances are identified using the test set data; the structural parameter settings of the Bi-LSTM load identification model during the training phase are shown in Table 1.
[0033] Table 1. Bi-LSTM Network Structure Parameter Settings
[0034] Furthermore, the specific details of S5 are as follows: S501. Calculate the similarity between the load characteristics of the unknown electrical appliance and the feature vector in the DTW algorithm template library. Calculate the shortest local cost metric weighted sum with the feature vector. The shortest local cost metric weighted sum is negatively correlated with the probability of identifying the electrical appliance category. Output the first identification probability vector. First identification probability vector The calculation formula is: (5) in, Representing the One test sample, Indicates the first Number of appliance types; Indicates the first The test sample is the same as the first DTW distance between similar appliances Indicates the first The test sample is the same as the first The smaller the DTW distance between similar electrical appliances, the higher the similarity. S502. Input the load characteristics of the unknown electrical appliance into the trained Bi-LSTM model to output the second identification probability vector; the second identification probability vector is... , Second identification probability vector The calculation formula is: (6) in, Softmax For activation functions; and They represent the first The output values of the forward and reverse layers of each test sample; S503. By combining the probability vectors of the two algorithms using a probability model, the final identified probability vector is: Identify probability vectors The calculation formula is: = (7).
[0035] Furthermore, the specific details of S6 are as follows: The remote monitoring device obtains the final identification probability vector in S5 as the load identification result by connecting to the cloud platform, and monitors the operating status of electrical equipment in real time.
[0036] Specifically, the final identification probability vector is obtained by synthesizing the probability model. Based on the probability of the unknown appliance being estimated as each type of appliance in the vector, the appliance with the highest probability is determined as the type of unknown appliance.
[0037] This embodiment uses the publicly available datasets PLAID and UK-DALE for verification. The PLAID dataset has a sampling frequency of 30kHz and a power frequency of 60Hz, and includes 16 types of electrical appliances, 329 electrical devices, and 1876 sets of measurement data. The UK-DALE dataset has a sampling frequency of 16kHz and a power frequency of 50Hz, and includes electrical device data from 5 households.
[0038] This embodiment selects five types of appliances with significant differences—air conditioners, incandescent lamps, refrigerators, microwave ovens, and washing machines—from the PLAID dataset as experimental subjects. Data from 5000 cycles of each of these five appliances is used as experimental data, resulting in a total of 25,000 load data sets, including 24,500 training data sets and 500 test data sets. To verify the effectiveness of the load features extracted by the STFT-SSAE algorithm provided in this invention, a comparative experiment on load identification performance under an LSTM model is conducted. Various load features are selected, including current waveform, third current harmonic, STFT (short-time Fourier transform) time-spectrum matrix, and STFT (short-time Fourier transform) time-spectrum matrix after KPCA dimensionality reduction (STFT-KPCA), and compared with the load features used in this invention. Each load feature is tested 100 times under the same model. The identification results of the load features extracted by the STFT-SSAE algorithm compared with other features are shown below. Figure 2 As shown in Table 2, the identification results of each load feature are presented. Experimental results reveal that when the load features extracted using the STFT-SSAE algorithm are presented, the model's average recognition accuracy reaches 91.15%, which is significantly improved compared to other load features, demonstrating the effectiveness and superiority of the feature extraction algorithm provided in this invention.
[0039] Table 2 Identification results of different load characteristics
[0040] This embodiment uses 25,000 load data points from the PLAID dataset and load samples from five categories of appliances (table lamps, refrigerators, televisions, laptops, and hair dryers) from the UK-DALE dataset as test samples. The data processing method is consistent with that of the PLAID dataset, resulting in a total of 50,000 experimental data points. To verify the effectiveness of the DTW-Bi-LSTM load identification architecture provided in this invention, load features extracted by the STFT-SSAE algorithm were used, and 100 repeated experiments were conducted on both the PLAID and UK-DALE datasets. To more comprehensively evaluate the load identification performance, four metrics—accuracy, recall, precision, and F1 score—were used as evaluation standards. This embodiment tests the overall identification performance of the DTW-Bi-LSTM load identification architecture as follows: Figure 3 As shown in the figure, the results of accuracy, recall, precision, and F1 score are plotted as box plots, where the solid line in the middle represents the median of the experimental results. It can be seen that the recognition performance of the DTW-Bi-LSTM load recognition architecture differs by about 1% on different datasets.
[0041] The present invention provides a remote power load monitoring device, the structural schematic of which is shown in the attached figure. Figure 4 As shown, it can be a computer, tablet computer, mobile phone, or other device. This device includes at least a processor 401, a memory 402, a communication bus 403, and a communication interface 404. Specifically: In this embodiment, the processor 401 may be a CPU, the memory 402 may be a high-speed RAM, the communication bus 403 is used to realize the connection and communication between various components, and the communication interface 404 may be a wired interface or a wireless interface. Depending on the target audience, it can be divided into a user interface 404-1 and a network interface 404-2. The user interface 404-1 includes a display screen, keyboard, etc., and the network interface 404-2 is responsible for receiving data through the network. The memory 402 stores the application program, and the processor 401 can call the program code stored in the memory 402 to execute it. Figure 1 Any of the method steps described.
[0042] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for system or system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and relevant parts can be referred to the descriptions in the method embodiments. The systems and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.
[0043] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for online identification of power load using DTW combined with Bi-LSTM, characterized in that, Includes the following steps: S1. Data Acquisition Steps: Collect aggregated power load data from users; S2, Event Detection Steps: Perform event detection on the aggregated power load data collected in S1 using the CUSUM algorithm to obtain unknown electrical appliance event signals; S3, Feature Extraction Step: Short-time Fourier Transform and stacked sparse autoencoder are used to extract features and reduce the dimension of the unknown event signals obtained in S2 to obtain the load features of the unknown electrical appliances after dimension reduction. S4. Model training steps: The feature extraction and feature dimensionality reduction methods in S3 are used to characterize the event signals of known electrical appliances, and the characteristic known electrical appliance event signals are used to train the Bi-LSTM model. S5. Load identification step: Use the DTW algorithm and the Bi-LSTM model trained in S4 to identify the load of unknown electrical appliances, and combine the identification results of the two through a probability model to obtain the final identification probability vector. S6. Application steps: The remote monitoring device realizes online monitoring of the operating status of electrical equipment based on the final identification probability vector obtained in S5.
2. The online power load identification method based on DTW combined with Bi-LSTM according to claim 1, characterized in that, The specific content of S1 is as follows: The overall energy consumption information of the user side is obtained through sensing devices and uploaded to the cloud platform. The overall energy consumption information includes: active power signal and current signal.
3. The online power load identification method based on DTW combined with Bi-LSTM according to claim 2, characterized in that, The specific content of S2 is as follows: build a CUSUM algorithm model on the cloud platform, perform event detection on the active power sequence, determine the operating status of electrical appliances, extract the current signal based on the operating status of electrical appliances, and obtain the current signal of unknown electrical appliances, that is, the event signal of unknown electrical appliances.
4. The online power load identification method based on DTW combined with Bi-LSTM according to claim 3, characterized in that, The calculation formula for the CUSUM algorithm is: in, , These represent the cumulative sum of load input and load cut-out events, respectively. , Let be the forward and reverse offsets of the current sequence point after removing the noise, respectively, denoted as . and , It is the average value of a sampling sequence of a certain length before the change point. This represents the noise value present in the sampled sequence.
5. The online power load identification method using DTW combined with Bi-LSTM according to claim 3, characterized in that, The formula for calculating a periodic current signal is: in, The first current sequence of load events One value, The first steady-state current before the event occurs One sampling point, The first steady-state current after the event occurs One sampling point, is the length of the current sequence.
6. The online power load identification method based on DTW combined with Bi-LSTM according to claim 1, characterized in that, The specific details of S3 are as follows: S301. Obtain the time spectrum diagram corresponding to the unknown electrical event signal in S2 through STFT, and construct the numerical matrix of the time spectrum diagram as load characteristics to distinguish different electrical appliances; S302. Use SSAE to perform feature dimensionality reduction on the load characteristics to obtain the dimensionality-reduced unknown electrical load characteristics.
7. The online power load identification method based on DTW combined with Bi-LSTM according to claim 1, characterized in that, The specific details of S4 are as follows: S401. The event signals of all known electrical appliances are characterized using the feature extraction and feature dimensionality reduction methods in S3 to obtain the load feature library and construct the DTW algorithm template library. S402. Divide the load feature data in the DTW algorithm template library to obtain training and test sets, which are used to train the Bi-LSTM model.
8. The online power load identification method based on DTW combined with Bi-LSTM according to claim 1, characterized in that, The specific details of S5 are as follows: S501. Calculate the similarity between the load characteristics of the unknown electrical appliance and the feature vector in the DTW algorithm template library. Calculate the shortest local cost metric weighted sum with the feature vector. The shortest local cost metric weighted sum is negatively correlated with the probability of identifying the electrical appliance category. Output the first identification probability vector. First identification probability vector The calculation formula is: in, Representing the One test sample, Indicates the first Each type of electrical appliance; Indicates the first The test sample is the same as the first DTW distance between similar appliances Indicates the first The test sample is the same as the first The smaller the DTW distance between similar electrical appliances, the higher the similarity. S502. Input the load characteristics of the unknown electrical appliance into the trained Bi-LSTM model to output the second identification probability vector; the second identification probability vector is... , Second identification probability vector The calculation formula is: in, Softmax For activation functions; and They represent the first The output values of the forward and reverse layers of each test sample; S503. By combining the probability vectors of the two algorithms using a probability model, the final identified probability vector is: Identify probability vectors The calculation formula is: = 。 9. The online power load identification method based on DTW combined with Bi-LSTM according to claim 1, characterized in that, The specific details of S6 are as follows: The remote monitoring device obtains the final identification probability vector in S5 as the load identification result by connecting to the cloud platform, and monitors the operating status of electrical equipment in real time.
