A non-intrusive load monitoring method based on multi-label classification
By using multi-label classification and integer programming algorithms, and utilizing electricity consumption data collected by smart meters, a multi-label classification model is constructed. This solves the problems of insufficient load identification efficiency and accuracy in existing technologies, and achieves efficient and accurate load status identification and power prediction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SOUTHEAST UNIV
- Filing Date
- 2023-01-05
- Publication Date
- 2026-04-24
AI Technical Summary
Existing non-intrusive load monitoring technologies are insufficient in terms of load identification efficiency and accuracy, making it difficult to effectively identify the operating status of multiple loads in a user's home.
A multi-label classification method is adopted. The total electricity consumption data is collected by smart meters to build a multi-label classification model. The RAKEL method of random forest is used for training and identification, and the power consumption of the load is predicted by combining integer programming algorithm.
It improves the efficiency and accuracy of load identification, can accurately identify the operating status of multiple loads, and predict their power consumption, which has practical value.
Smart Images
Figure CN115983480B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power load management and analysis technology, specifically a non-intrusive load monitoring method based on multi-label classification. Background Technology
[0002] Non-intrusive load monitoring technology aims to collect total electricity consumption data from users by installing smart meters only at the user's electricity inlet, thereby identifying the user's load operating status. The total electricity consumption data for a user is composed of the electricity consumption data of each operating load in the user's home, superimposed with background noise. Typically, multiple loads may be operating simultaneously; therefore, this technology provides a non-intrusive load monitoring method. Summary of the Invention
[0003] (a) Technical problems to be solved
[0004] To address the shortcomings of existing technologies, this invention provides a non-intrusive load monitoring method based on multi-label classification, which improves the efficiency of load identification and achieves higher identification accuracy, thus possessing strong practical value and significance.
[0005] (II) Technical Solution
[0006] To achieve the above objectives, the present invention provides the following technical solution:
[0007] Firstly, a non-intrusive load monitoring method based on multi-label classification is provided, including:
[0008] Collect users' total electricity consumption data and divide the data into training and testing sets to build electricity consumption models for different types of loads;
[0009] A multi-label classification model is constructed to characterize the non-intrusive load identification problem;
[0010] Construct a multi-label classification problem solving model, and train the multi-label classification problem solving model parameters using training and validation sets;
[0011] The trained multi-label classification problem-solving model is applied to load identification to obtain the operating status of the load;
[0012] Based on the load's operating status, predict the load's power consumption.
[0013] Preferably, the total electricity consumption data of the power users includes active power, reactive power, voltage, and current harmonics; and 85% of the total electricity consumption data of the users is randomly selected as the test set, and the remaining 15% is used as the verification set.
[0014] Preferably, the power consumption model for different types of loads includes the number of operating states of the load, and the corresponding active and reactive power consumption values.
[0015] Preferably, the construction of the multi-label classification model to characterize the non-invasive load identification problem includes:
[0016] Establish the relationship between the user's total electricity consumption data and the individual electricity consumption data of each load:
[0017]
[0018]
[0019] Where P(t) and Q(t) are the total active power and total reactive power of the user at time t, respectively, N is the number of loads, and M is the total reactive power of the user at time t. i p represents the number of operating states for the i-th load. i,j (t), q i,j (t) represent the rated active power and reactive power corresponding to the j-th operating state of the i-th load, respectively, x i,j (t)∈{0,1} represents the operating state of the i-th load at time t, x i,j (t) = 1 indicates that the load is in its j-th operating state; otherwise, x i,j (t) = 0 indicates that the state is not in that state; e p (t) and e q (t) represents the background noise of active power and reactive power at time t, respectively.
[0020] Preferably, the step of constructing a multi-label classification problem-solving model and training the multi-label classification problem-solving model parameters using a training set and a validation set includes:
[0021] The RAKEL multi-label classification method based on random forest is used to solve the non-intrusive load identification problem. The specific calculation steps of RAKEL are as follows:
[0022] For the total electricity consumption data samples collected from users at different times, when applying the multi-label classification problem, the working status corresponding to each load is regarded as a label and constitutes a label set;
[0023] Select several tag combinations from the tag set, where each tag combination contains k tags;
[0024] Build a classifier for each label combination and train it;
[0025] When identifying unknown samples, the identification results of each classifier are obtained, and a voting method is used to determine the final identification result.
[0026] Preferably, applying the trained multi-label classification problem-solving model to load identification and obtain the load's operating status includes:
[0027] A random forest model was used for load identification, and grid search and k-fold cross-validation were used to determine the model parameters. The specific steps are as follows:
[0028] Set the initialization parameters of the random forest model, as well as the range and step size of each parameter, and let the parameter k = 10 for k-fold cross-validation;
[0029] Under a specific set of model parameters, the test set data is divided into ten groups on average and trained ten times. During each training process, one group of data is selected to test the training results of the model, and the remaining nine groups of data are used to train the model. The average of the ten training results is taken as the training result of the model parameters of that group.
[0030] After all parameter combinations have been trained, the set of model parameters with the best training results is selected as the final parameters of the load identification model.
[0031] Preferably, predicting the power consumption of the load based on its operating status includes:
[0032] Target:
[0033]
[0034] Where P(t) and Q(t) are the user's total active power and total reactive power data at time t, respectively. i,j (t), q i,j (t) represents the active and reactive power values corresponding to the j-th operating state of the i-th load identified as being in the on state, respectively, and N is the number of loads identified as being in the running state. i Let x be the number of operating states of the i-th load identified as being in the on state. i,j (t) is used to characterize the operating state of the i-th load identified as being in the on state at time t, x i,j (t) = 0 indicates that the load is not in its j-th operating state, x i,j (t) = 1 indicates that the load is in its j-th operating state;
[0035] constraint:
[0036] For a load with multiple operating states, it can be in at most one operating state at any given time, as follows:
[0037]
[0038] When estimating the power consumption of a load, the operating state x of the load that will be identified as active will be used. i,j (t) is used as the decision variable. An integer programming algorithm is used to iteratively solve the objective function to determine the operating state of each load and further predict the power consumption of each load, as follows:
[0039]
[0040] in, p is the estimated active power consumption of the i-th load at time t. i (t) represents the rated active power corresponding to the working state of the i-th load at time t.
[0041] Secondly, a device is provided, comprising:
[0042] One or more processors;
[0043] Memory, used to store one or more programs.
[0044] When the one or more programs are executed by the one or more processors, the one or more processors execute the non-intrusive load monitoring method based on multi-label classification.
[0045] Thirdly, a computer-readable storage medium is provided that stores a computer program, which, when executed by a processor, implements the aforementioned non-intrusive load monitoring method based on multi-label classification.
[0046] (III) Beneficial Effects
[0047] This invention discloses a non-intrusive load monitoring method based on multi-label classification. During the load identification process, a multi-label classification algorithm is used to identify the operating status of the load, and the power consumed by the load is further predicted based on the identification results. This improves the efficiency of load identification and achieves higher identification accuracy, which has strong practical value and significance. Attached Figure Description
[0048] Figure 1 This is a schematic diagram of the method flow of the present invention. Detailed Implementation
[0049] The technical solutions in 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.
[0050] Example
[0051] like Figure 1 As shown, this embodiment of the invention provides a non-intrusive load monitoring method based on multi-label classification, including:
[0052] Step 1: Collect the user's total electricity consumption data through smart meters installed at the power inlet of the power user, and divide the user's total electricity consumption data into training set and test set to build electricity consumption models for different types of loads.
[0053] Specifically, the total electricity consumption data for power users includes active power, reactive power, voltage, and current harmonics. The electricity consumption model for a typical load includes the number of load operating states and the corresponding active and reactive power consumption values. 85% of the total electricity consumption data from users is randomly selected as the test set, and the remaining 15% is used as the validation set.
[0054] Step 2: Construct a multi-label classification model to characterize the non-intrusive load identification problem.
[0055] In the load identification problem, the relationship between the total electricity consumption data of users and the individual electricity consumption data of each load is as follows:
[0056]
[0057]
[0058] Where P(t) and Q(t) are the total active power and total reactive power of the user at time t, respectively, N is the number of loads, and M is the total reactive power. i p represents the number of operating states of the i-th load (excluding the off state). i,j (t), q i,j (t) represent the rated active power and reactive power corresponding to the j-th operating state of the i-th load, respectively, x i,j (t)∈{0,1} represents the operating state of the i-th load at time t, x i,j (t) = 1 indicates that the load is in its j-th operating state; otherwise, x i,j (t) = 0 indicates that the state is not in that state. p (t) and e q (t) represents the background noise of active power and reactive power at time t, respectively.
[0059] The total electricity consumption data collected from power users is used as a data sample, and the various parameters contained within it are used as data features. The operating status of each load is used as the label of the sample. By analyzing and identifying each sample, its corresponding label is determined, thus identifying the operating status of each load based on the total electricity consumption data. Considering that multiple loads may be operating simultaneously, meaning a sample may have multiple different labels, multi-label classification can be used to characterize the non-intrusive load identification problem.
[0060] Step 3: Construct a multi-label classification problem solving model and train the multi-label classification problem solving model parameters using the training set and validation set.
[0061] It employs the RAKEL multi-label classification method based on random forest to solve the load identification problem. The specific calculation steps of RAKEL are as follows:
[0062] Step 3.1) For the total electricity consumption data samples collected at different times, when applying the multi-label classification problem, the working state corresponding to each load can be regarded as a label and constitute a label set;
[0063] Step 3.2) Select several tag combinations from the tag set, where each tag combination contains k tags;
[0064] Step 3.3) Build a classifier for each label combination and train it;
[0065] Step 3.4) When identifying unknown samples, obtain the identification results of each classifier and use a voting method to determine the final identification result.
[0066] It employs a random forest model for load identification and uses grid search and k-fold cross-validation to determine model parameters. The specific steps are as follows:
[0067] Step 4.1) Set the initialization parameters of the random forest model, as well as the range of values and step size of each parameter. Let the parameter k for k-fold cross-validation be 10.
[0068] Step 4.2) Under a specific set of model parameters, the test set data is divided into ten groups on average and trained ten times. During each training process, one group of data is selected to test the training results of the model, and the remaining nine groups of data are used to train the model. The average of the ten training results is taken as the training result of the model parameters of that group.
[0069] Step 4.3) After all parameter combinations have been trained, select the set of model parameters with the best training results as the final parameters of the load identification model.
[0070] Step 4: Apply the trained model to load identification to obtain the load's operating status.
[0071] Step 5: Based on the load's operating status, predict the load's power consumption.
[0072] It involves the following optimization model:
[0073] The optimization objective is:
[0074]
[0075] Where P(t) and Q(t) are the user's total active power and total reactive power data at time t, respectively. i,j (t), q i,j (t) represents the active and reactive power values corresponding to the j-th operating state of the i-th load identified as being in the on state, respectively, and N is the number of loads identified as being in the running state. i Let x be the number of operating states of the i-th load identified as being in the on state. i,j (t) is used to characterize the operating state of the i-th load identified as being in the on state at time t, x i,j (t) = 0 indicates that the load is not in its j-th operating state, x i,j (t) = 1 indicates that the load is in its j-th operating state. The optimization constraint is:
[0076] For a load with multiple operating states, it can be in at most one operating state at any given time, that is:
[0077]
[0078] When estimating the power consumption of a load, the operating state x of the load that will be identified as active will be used. i,j (t) is used as the decision variable. An integer programming algorithm is used to iteratively solve the objective function to determine the operating state of each load and further estimate the power consumption of each load, i.e.:
[0079]
[0080] in, p is the estimated active power consumption of the i-th load at time t. i (t) represents the rated active power corresponding to the working state of the i-th load at time t.
[0081] In the selected embodiment of this invention, load data of residential users is set in the LVNS simulation platform, and their electricity consumption behavior over five days is simulated to obtain electricity consumption data as an application scenario. The model parameters of each load are shown in Table 1.
[0082] Table 1. Model parameters of the load
[0083]
[0084] The F1 score, including F1-micro and F1-macro, and the root mean square error (NMSE) from the machine learning field are used as metrics to measure the accuracy of load identification. The F1 score is used to evaluate the accuracy of load identification, and the NMSE is used to evaluate the accuracy of load power estimation. The proposed method is compared with load identification methods based on decision trees and support vector machines, as shown in Table 2.
[0085] Table 2 Load Identification Results
[0086] Load identification method F1-micro F1-macro Method of the present invention 0.994 0.992 Decision Tree 0.967 0.906 Support Vector Machine 0.9693 0.9021
[0087] As can be seen from Table 2, the load identification algorithms based on decision trees and support vector machines have relatively high F1-micro and F1-macro values, with the F1-micro value exceeding 0.96 and the F1-macro value around 0.9. The method proposed in this invention further improves the F1-micro and F1-macro values, both exceeding 0.99, indicating that the method of this invention can accurately and reliably complete the load identification task.
[0088] The power consumption estimates for each load are shown in Table 3.
[0089] Table 3 Power consumption estimation results for each load
[0090]
[0091] As can be seen from Table 3, the NMSE values of each load are at a low level, with the highest not exceeding 3% and the lowest only 0.1%, indicating that the estimated power consumption of the load is close to its actual power consumption, demonstrating the accuracy and effectiveness of the method proposed in this invention in estimating load power consumption.
[0092] Embodiments of this application may be provided as methods or computer program products. Therefore, this application may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application may take the form of a computer program product implemented 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. The solutions in the embodiments of this application may be implemented in various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.
[0093] 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.
[0094] 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.
[0095] 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.
[0096] It should be noted that, in this document, relational terms such as "first" and "second" are used only 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. Furthermore, 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 limitations, 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.
Claims
1. A non-intrusive load monitoring method based on multi-label classification, characterized in that, include: Collect users' total electricity consumption data and divide the data into training and testing sets to build electricity consumption models for different types of loads; A multi-label classification model is constructed to characterize the non-intrusive load identification problem; Construct a multi-label classification problem solving model, and train the multi-label classification problem solving model parameters using training and validation sets; The trained multi-label classification problem-solving model is applied to load identification to obtain the operating status of the load; Based on the load's operating status, predict the load's power consumption; The construction of a multi-label classification model to characterize the non-intrusive load identification problem includes: Establish the relationship between the user's total electricity consumption data and the individual electricity consumption data of each load: Where P(t) and Q(t) are the total active power and total reactive power of the user at time t, respectively, and N is the number of loads. Let i be the number of working states for the i-th load. , These are the rated active power and reactive power corresponding to the j-th operating state of the i-th load, respectively. This indicates the operating state of the i-th load at time i. This indicates that the load is in its j-th operating state, and vice versa. This indicates that the state is not currently in that state. and These represent the background noise at time t, representing active power and reactive power, respectively. The construction of a multi-label classification problem-solving model, and the training of the model parameters using a training set and a validation set, includes: The RAKEL multi-label classification method based on random forest is used to solve the non-intrusive load identification problem. The specific calculation steps of RAKEL are as follows: For the total electricity consumption data samples collected from users at different times, when applying the multi-label classification problem, the working status corresponding to each load is regarded as a label and constitutes a label set; Select several tag combinations from the tag set, where each tag combination contains k tags; Build a classifier for each label combination and train it; When identifying unknown samples, the identification results of each classifier are obtained, and a voting method is used to determine the final identification result; The method of predicting the power consumption of the load based on its operating status includes: Target: Where P(t) and Q(t) are the user's total active power and total reactive power data at time t, respectively. , These represent the active and reactive power values corresponding to the j-th operating state of the i-th load identified as being in the on state, respectively, and N is the number of loads identified as being in the running state. Let be the number of operating states of the i-th load identified as being in the on state. Used to characterize the operating state of the i-th load identified as being in the on state at time t. This indicates that the load is not in its j-th operating state. This indicates that the load is in its j-th operating state; constraint: For a load with multiple operating states, it can be in at most one operating state at any given time, as follows: When estimating the power consumption of a load, the operating state of the load will be identified as active. As decision variables, an integer programming algorithm is used to iteratively solve the objective function to determine the operating state of each load and further predict the power consumption of each load, as follows: in, Let be the estimated active power consumption of the i-th load at time t. Let be the rated active power corresponding to the operating state of the i-th load at time t.
2. The non-intrusive load monitoring method based on multi-label classification according to claim 1, characterized in that: The user's total electricity consumption data includes active power, reactive power, voltage, and current harmonics; We randomly selected 85% of the total electricity consumption data of users as the test set and the remaining 15% as the validation set.
3. The non-intrusive load monitoring method based on multi-label classification according to claim 2, characterized in that: The power consumption models for different types of loads include the number of operating states of the load, as well as the corresponding active and reactive power consumption values.
4. The non-intrusive load monitoring method based on multi-label classification according to claim 3, characterized in that: The step of applying the trained multi-label classification problem-solving model to load identification and obtaining the load's operating status includes: A random forest model was used for load identification, and grid search and k-fold cross-validation were used to determine the model parameters. The specific steps are as follows: Set the initialization parameters of the random forest model, as well as the range and step size of each parameter, and let the parameter k = 10 for k-fold cross-validation; Under a specific set of model parameters, the test set data is divided into ten groups on average and trained ten times. During each training process, one group of data is selected to test the training results of the model, and the remaining nine groups of data are used to train the model. The average of the ten training results is taken as the training result of the model parameters of that group. After all parameter combinations have been trained, the set of model parameters with the best training results is selected as the final parameters of the load identification model.
5. A device, characterized in that, include: One or more processors; Memory, used to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors perform a non-intrusive load monitoring method based on multi-label classification as described in any one of claims 1-4.
6. A computer-readable storage medium storing a computer program, characterized in that, When executed by the processor, the program implements a non-intrusive load monitoring method based on multi-label classification as described in any one of claims 1-4.