A non-intrusive load monitoring method and system for steady-state and transient hybrid loads

Abstract

This invention discloses a non-intrusive load monitoring method and system that combines steady-state and transient states, belonging to the field of power load management and analysis technology. The monitoring method includes: collecting total electricity consumption data from users through smart meters to construct a typical load electricity consumption model; using an event detection algorithm to identify events, dividing them into steady-state and transient processes, and extracting steady-state and transient features respectively; using a method based on steady-state features to identify the load, and obtaining a candidate set of steady-state identification results based on the electricity consumption data in the electricity consumption model; using a method based on transient features to identify the load, and obtaining a candidate set of transient identification results based on the electricity consumption data in the electricity consumption model; constructing a time-series path characterizing the changes in load operating status, using the Viterbi algorithm to filter the candidate set of identification results, and determining the final identification result; eliminating erroneous identification results to a certain extent, further improving the accuracy of load identification.

Description

Technical Field

[0001] This invention belongs to the field of power load management and analysis technology, specifically relating to a non-intrusive load monitoring method and system that combines steady-state and transient states. 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. It offers advantages such as low cost, high reliability, and protection of user privacy, and has enormous application potential in the demand-side response field, thus receiving widespread attention and research in recent years.

[0003] Based on the different types of electricity consumption data characteristics used, load identification methods can be divided into two categories: load identification methods based on steady-state characteristics and load identification methods based on transient characteristics. Steady-state characteristics can be extracted from electricity consumption data during steady-state processes, in which the load's operating state remains unchanged. Transient characteristics can be extracted from electricity consumption data during transient processes, in which the load's operating state changes. Load identification methods based on steady-state characteristics are generally simpler and easier to implement, but they struggle to effectively identify loads or load combinations with similar steady-state characteristics. Load identification methods based on transient characteristics can achieve higher identification accuracy, but they rely on high-frequency data, require large amounts of data, and place high demands on smart meter hardware. Therefore, there is an urgent need for a load identification method that can effectively identify load combinations with sufficiently high accuracy. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide a non-intrusive load monitoring method and system that integrates steady-state and transient states.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] A non-intrusive load monitoring method for steady-state and transient hybrid loads includes the following steps:

[0007] S1 collects total electricity consumption data from users through smart meters and constructs electricity consumption models for typical loads;

[0008] S2, use an event detection algorithm to identify events, divide steady-state processes into steady-state processes and transient processes, and extract steady-state features and transient features respectively;

[0009] S3. Load identification is performed using a method based on steady-state characteristics, and a candidate set of steady-state identification results is obtained based on the electricity consumption data in the electricity consumption model.

[0010] S4. Load identification is performed using a method based on transient features, and a candidate set of transient identification results is obtained based on the electricity consumption data in the electricity consumption model.

[0011] S5. Construct a time-series path to characterize the changes in load operating status, and use the Viterbi algorithm to filter the candidate sets of steady-state and transient identification results to determine the final identification result.

[0012] Furthermore, the total electricity consumption data includes: active power, reactive power, voltage, current, and timestamp;

[0013] The power consumption model includes: the number of operating states of the load, the corresponding active and reactive power consumption values, and the power change data of the load when state switching occurs.

[0014] Furthermore, the optimized model for load identification based on steady-state characteristics is as follows:

[0015]

[0016] 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 power and reactive power values ​​corresponding to the j-th operating state of the i-th load, respectively, where N is the number of loads and n is the number of loads. i Let x be the number of operating states for the i-th load. i,j (t) is used to characterize the operating state of the i-th load 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.

[0017] Furthermore, the optimization constraints for load identification based on steady-state characteristics are as follows:

[0018]

[0019] Furthermore, the steps for solving the candidate set of load identification results based on steady-state characteristics are as follows:

[0020] 1) When identifying the load for each steady-state process, the operating state x of the load is... i,j (t) is used as the decision variable, and the objective function is solved iteratively using the discrete particle swarm optimization algorithm;

[0021] 2) Update and record several global optimal results and their corresponding objective function values ​​during each iteration; output them after the iteration is completed as a candidate set of load identification results based on steady-state characteristics.

[0022] Furthermore, the optimized model for load identification based on transient features is as follows:

[0023]

[0024] The function f2() compares the similarity between transient features extracted from electricity consumption data and feature templates. A smaller function value indicates a greater similarity between the transient feature and the template feature. feature(k) represents the transient feature extracted from the electricity consumption data of the k-th event. i,j This represents the feature template corresponding to the j-th state switching process of the i-th load.

[0025] Furthermore, the steps for solving the candidate set of load identification results based on transient features are as follows:

[0026] 1) When identifying the load for each transient process, a dynamic warping algorithm is used to calculate the similarity between transient features and feature templates;

[0027] 2) Extract the load state switching processes corresponding to the minimum values ​​of several objective functions, and use them as a candidate set of load identification results based on steady-state characteristics.

[0028] Furthermore, the load state switching process can be described using the variable x. i,j This means that x takes three values: -1, 0, and 1. i,j =0 indicates that the j-th operating state of the i-th load has not changed, x i,j =1 indicates that the j-th operating state of the i-th load begins during the transient process, x i,j =-1 indicates that the j-th working state of the i-th load has ended during the transient process;

[0029] The load state switching situation of the kth transient process is used The corresponding candidate set of transient recognition results is represented by an array. express.

[0030] Furthermore, the steps for obtaining the final load identification result are as follows:

[0031] S51, the input parameters are: model λ=(A,B,π) and observation sequence O=(o1,o2,...,o K The output is the recognition result X for each hidden sequence. s =(X s,1 ,X s,2 ,...,X s,K The observation sequence can be viewed as the electrical features extracted from the total user electricity consumption data for each steady-state process. These features are used for load identification based on steady-state characteristics. The hidden state corresponds to each steady-state process. Will By normalizing the objective function value corresponding to each result, the emission probability B can be obtained; The transition probability A can be obtained by normalizing the objective function value corresponding to each result.

[0032] S52, Initialization: The forward probabilities δ1(i) = π(i) * b1(i) of each hidden state in S51, and their corresponding optimal paths ψ1(i) = 0, where i = 1, 2, ..., N s ;

[0033] S53, recursively calculate, the forward probability of the k-th step is δ k (i)=max[δ k-1 (j1)*a k-1 (j2)]b k (i); where j1 = 1, 2, ..., N s and j2 = 1, 2, ..., N e And satisfy the formula X s,k-1 +X e,k-1 =X s,k The optimal path is ψ k (i)=argmax[δ k-1 (j1)*a k-1 (j2)];

[0034] S54, Termination: When recursively reaching the end of the observation sequence, calculate the maximum forward probability P of the last step. * =maxδ K (i) and the corresponding path

[0035] S55, Optimal path backtracking: For k = K-1, K-2, ..., 2, 1, calculate the path. and the corresponding load identification results This result is the final load identification result obtained after verification and analysis.

[0036] Furthermore, it includes a data acquisition module, an event recognition module, a steady-state recognition module, a transient state recognition module, and a result determination module;

[0037] Data acquisition module: Collects total electricity consumption data from users through smart meters and constructs electricity consumption models for typical loads;

[0038] Event recognition module: Uses event detection algorithms to identify events, classify them into steady-state and transient processes, and extract steady-state and transient features respectively;

[0039] Steady-state identification module: uses a method based on steady-state characteristics to identify loads and obtains a candidate set of steady-state identification results based on the electricity consumption data in the electricity consumption model;

[0040] Transient identification module: Uses a method based on transient features to identify loads and obtains a candidate set of transient identification results based on the electricity consumption data in the electricity consumption model;

[0041] Result determination module: Construct a time-series path representing the changes in load operating status, use the Viterbi algorithm to filter the candidate sets of steady-state and transient identification results, and determine the final identification result.

[0042] The beneficial effects of this invention are:

[0043] 1. This invention can, during the load identification process, establish multiple steady-state and transient processes in conjunction based on the logical relationship between the load operating states of adjacent steady-state and transient processes, and perform verification analysis on the identification results. This can eliminate erroneous identification results to a certain extent, further improving the accuracy of load identification. It has strong practical value and significance.

[0044] 2. Furthermore, by constructing the method of this invention, in the load identification problem based on event detection, the dependence on high-frequency data of transient processes can be greatly reduced by introducing steady-state timing logic to assist identification, thereby reducing data sampling requirements and achieving lightweighting of smart meter hardware requirements. Attached Figure Description

[0045] 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, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a flowchart illustrating the non-intrusive load monitoring method for steady-state and transient hybridization of the present invention.

[0047] Figure 2 This is a schematic diagram of the calculation process of the present invention, which uses the Viterbi algorithm to verify and analyze the candidate set of recognition results and determine the final recognition result. Detailed Implementation

[0048] 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.

[0049] like Figure 1 As shown, a non-intrusive load monitoring method for steady-state and transient hybrid loads includes the following steps:

[0050] S1 collects total electricity consumption data of users through smart meters installed at the power supply inlet of electricity users, and constructs a typical load electricity consumption model;

[0051] The total electricity consumption data for power users includes parameters such as active power, reactive power, voltage, current, and timestamps. The electricity consumption model for typical loads includes the number of load operating states, the corresponding active and reactive power consumption values, and the power change data when the load switches states.

[0052] S2 uses an event detection algorithm to identify events, extract the event occurrence time and transient data, and classify steady-state and transient processes. Taking the active power time-series data obtained in S1 as the object, the event detection algorithm is used to identify events in the time-series data, and then the total electricity consumption data is classified. The specific method is as follows:

[0053] S21 As time progresses, monitor the continuous change of active power P = Pt - Pt-1. If |P| is less than the threshold, continue monitoring as time progresses. If |P| is greater than the threshold, proceed to S22.

[0054] S22 records the start time of the event as ts = t-1, classifies the electricity consumption data between the end time of the previous event and ts into a steady-state process, and proceeds to S23;

[0055] S23 Continue monitoring over time. If |P| is greater than the threshold, continue monitoring over time. If |P| is less than the threshold, proceed to S24.

[0056] S24 records the event end time as te = t-1, classifies the electricity consumption data between the event start time ts and the event end time te into a transient process, and proceeds to S21;

[0057] After the S25 event detection algorithm is completed, the steady-state process dataset and the transient process dataset are arranged in timestamp order to form a hybrid steady-state and transient time series dataset.

[0058] S3, based on the typical power consumption model in S1, uses a method based on steady-state features to identify the load of the steady-state process data object in S22, and obtains a candidate set of steady-state identification results, which includes several optimal solutions and corresponding objective function values; and the steady-state features are extracted from the power consumption data of the steady-state process.

[0059] The optimization model involved is as follows:

[0060] The optimization objective is:

[0061]

[0062] 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 power and reactive power values ​​corresponding to the j-th operating state of the i-th load, respectively, where N is the number of loads and n is the number of loads. i Let x be the number of operating states for the i-th load. i,j (t) is used to characterize the operating state of the i-th load 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;

[0063] The optimization constraints are:

[0064] For a load with multiple operating states, it can be in at most one operating state at any given time, that is:

[0065]

[0066] When identifying the load for each steady-state process, the operating state x of the load is... i,j (t) is used as the decision variable. The objective function is solved iteratively using the discrete particle swarm optimization algorithm. In each iteration, several global optimal results and their corresponding objective function values ​​are updated and recorded. After the iteration is completed, the results are output as a candidate set of load identification results based on steady-state characteristics.

[0067] The load operating state of the k-th steady-state process is available. This indicates that the corresponding steady-state identification result candidate set can be represented by an array. express.

[0068] S4, based on the typical power consumption model in S1, uses a transient feature-based method to identify the load of the transient process data object in S24, and obtains a candidate set of transient identification results, which includes several optimal solutions and corresponding objective function values; and the transient features are extracted from the power consumption data of the transient process.

[0069] The optimization model involved is as follows:

[0070] The optimization objective is:

[0071]

[0072] The function f2() compares the similarity between transient features extracted from electricity consumption data and feature templates. A smaller function value indicates a greater similarity between the transient feature and the template feature. feature(k) represents the transient feature extracted from the electricity consumption data of the k-th event. i,jThis represents the feature template corresponding to the j-th state switching process of the i-th load; in addition, it can be implemented based on pattern recognition methods to learn the electrical characteristics of the load during state switching, thereby achieving the goal of load identification.

[0073] When identifying the load for each transient process, a dynamic warping algorithm is used to calculate the similarity between transient features and feature templates, and to extract the load state switching processes corresponding to the minimum values ​​of several objective functions as a candidate set of load identification results based on transient features.

[0074] The load state switching process can be achieved using variable x. i,j This indicates that it can take three values: -1, 0, and 1, x i,j =0 indicates that the j-th operating state of the i-th load has not changed, x i,j =1 indicates that the j-th operating state of the i-th load begins during the transient process, x i,j =-1 indicates that the j-th working state of the i-th load has ended during the transient process;

[0075] The load state switching situation of the kth transient process can be used This indicates that the corresponding candidate set of transient recognition results can be represented by an array. express.

[0076] S5, based on the transient steady-state time-series hybrid dataset of S25, constructs a time-series path characterizing the changes in load operating status, and uses the Viterbi algorithm to filter the candidate set of identification results to determine the final identification result;

[0077] The specific steps involved, such as Figure 2 As shown:

[0078] S51, the input parameters are: model λ=(A,B,π) and observation sequence O=(o1,o2,...,o K The output is the recognition result X for each hidden sequence. s =(X s,1 ,X s,2 ,...,X s,K The observation sequence can be viewed as the electrical features extracted from the total user electricity consumption data for each steady-state process. These features are used for load identification based on steady-state characteristics. The hidden state corresponds to each steady-state process. Will By normalizing the objective function value corresponding to each result, the emission probability B can be obtained; The transition probability A can be obtained by normalizing the objective function value corresponding to each result.

[0079] S52, Initialization: The forward probabilities δ1(i) = π(i) * b1(i) of each hidden state in S51, and their corresponding optimal paths ψ1(i) = 0, where i = 1, 2, ..., N s ;

[0080] S53, recursively calculate, the forward probability of the k-th step is δ k (i)=max[δ k-1 (j1)*a k-1 (j2)]b k (i); where j1 = 1, 2, ..., N s and j2 = 1, 2, ..., N e And satisfy the formula X s,k-1 +X e,k-1 =X s,k The optimal path is ψ k (i)=argmax[δ k-1 (j1)*a k-1 (j2)];

[0081] S54, Termination: When recursively reaching the end of the observation sequence, calculate the maximum forward probability P of the last step. * =maxδ K (i) and the corresponding path

[0082] S55, Optimal path backtracking: For k = K-1, K-2, ..., 2, 1, calculate the path. and the corresponding load identification results This result is the final load identification result obtained after verification and analysis.

[0083] In addition, a non-intrusive load monitoring system that combines steady-state and transient states may include a data acquisition module, an event recognition module, a steady-state recognition module, a transient recognition module, and a result determination module.

[0084] The data acquisition module collects total electricity consumption data from users through smart meters and constructs electricity consumption models for typical loads.

[0085] Event recognition module: Uses event detection algorithms to identify events, classify them into steady-state and transient processes, and extract steady-state and transient features respectively;

[0086] Steady-state identification module: uses a method based on steady-state characteristics to identify loads and obtains a candidate set of steady-state identification results based on the electricity consumption data in the electricity consumption model;

[0087] Transient identification module: Uses a method based on transient features to identify loads and obtains a candidate set of transient identification results based on the electricity consumption data in the electricity consumption model;

[0088] Result determination module: Construct a time-series path representing the changes in load operating status, use the Viterbi algorithm to filter the candidate set of identification results, and determine the final identification result.

[0089] The technical solution of the present invention will be illustrated below with specific embodiments:

[0090] Load data of residential users were set in the LVNS simulation platform, and their electricity consumption behavior for ten days was simulated. The electricity consumption data was used as an application scenario. The method of this invention was compared with the result of load identification using only discrete particle swarm optimization algorithm. The model parameters of each load are shown in Table 1.

[0091] Table 1. Model parameters of the load

[0092] Television (TV) 270 1 Single-state load Incandescent Lamp 1 (INC1) 120 0.38 Single-state load Furnace (FUR) 500 0.84 Multi-state load Oven (TOA) 900 1 Single-state load Food processing machine (FOO) 800 1 Single-state load Refrigerator (RFR) 150 0.92 Single-state load Incandescent lamp 2 (INC2) 40 1 Single-state load STO (Stove Toll Collection) 1100 0.9 Single-state load

[0093] Using the evaluation metrics F1 score, root mean square error (RMSE), mean square error (MSE), and standard root mean square error (NMSE) from the field of machine learning as indicators to measure the accuracy of load identification, the load identification results of the method of the present invention are shown in Table 2.

[0094] Table 2 Load identification results of the method of the present invention

[0095]

[0096]

[0097] The results of load identification using only the discrete particle swarm optimization algorithm are shown in Table 3.

[0098] Table 3 shows the load identification results using only the Discrete Particle Swarm Optimization (DPSO) algorithm.

[0099] TV set 0.8262 48.1954 9.2740 0.5873 Incandescent lamp 1 0.9635 21.3602 3.8021 0.2748 oven 0.9610 78.9334 23.2097 0.2764 toaster 0.9009 78.0576 6.7700 0.4432 Food processing machine 0.7363 91.9857 10.5767 0.7693 refrigerator 0.9015 39.3395 11.2864 0.4449 Incandescent lamp 2 0.8354 12.5957 3.9663 0.5576 electric furnace 0.9971 41.3123 7.4525 0.0772

[0100] As shown in Tables 2 and 3, compared to using only the discrete particle swarm optimization algorithm, the load identification results of the method of this invention show improved F1 values ​​for most loads, while RMSE, MSE, and NMSE values ​​decreased significantly. Only one load showed a decrease in identification accuracy, but this load has a relatively small rated power and will not significantly affect the overall identification accuracy. In the application scenario, seven out of eight loads had F1 values ​​higher than 0.9, with five loads having F1 values ​​exceeding 0.97, and the electric furnace reaching the highest F1 value of 0.9981. This demonstrates that the method of this invention has good reliability and can effectively improve the accuracy of load identification.

[0101] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0102] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A non-intrusive load monitoring method that combines steady-state and transient states, characterized in that, The monitoring method includes the following steps: S1. Collect total electricity consumption data of users through smart meters and construct a typical load electricity consumption model. The electricity consumption model includes: the number of working states of the load, the corresponding active power and reactive power consumption values, and the power change data of the load when the state changes. S2, use an event detection algorithm to identify events, divide steady-state processes into steady-state processes and transient processes, and extract steady-state features and transient features respectively; S3. Load identification is performed using a method based on steady-state characteristics, and a candidate set of steady-state identification results is obtained based on the electricity consumption data in the electricity consumption model. S4. Load identification is performed using a method based on transient features, and a candidate set of transient identification results is obtained based on the electricity consumption data in the electricity consumption model. S5. Construct a time-series path to characterize the changes in load operating status, and use the Viterbi algorithm to filter the candidate sets of steady-state and transient identification results to determine the final identification result. The optimization model for load identification based on steady-state characteristics is as follows: in , For users in The total active power and total reactive power data at any given time. , The first The first load The active and reactive power values ​​corresponding to each operating state For the number of loads, For the first Number of operating states of each load Used to characterize Time of the first The operating status of each load This indicates that the load is not in its first position. A work status, This indicates that the load is in its first... One working status; The optimization constraints for load identification based on steady-state characteristics are: The steps for solving the candidate set of load identification results based on steady-state characteristics are as follows: 1) When identifying the load for each steady-state process, the operating state of the load is... As decision variables, the objective function is solved iteratively using the discrete particle swarm optimization algorithm; 2) Update and record several global optimal results and their corresponding objective function values ​​during each iteration; output them after the iteration is completed as a candidate set of load identification results based on steady-state characteristics; The optimized model for load identification based on transient features is as follows: Where the function This function is used to compare the similarity between transient features extracted from electricity consumption data and feature templates. The smaller the function value, the greater the similarity between the transient feature and the template feature. Indicates from the first Transient features extracted from the electricity consumption data of each event. Indicates the first The first load Feature templates corresponding to each state transition process; The steps for solving the candidate set of load identification results based on transient features are as follows: 1) When identifying the load for each transient process, a dynamic warping algorithm is used to calculate the similarity between transient features and feature templates; 2) Extract the load state switching processes corresponding to the minimum values ​​of several objective functions, and use them as a candidate set of load identification results based on steady-state characteristics; The load state switching process can use variables This means taking the values ​​-1, 0, and 1. Indicates the first The first load The working status has not changed. Indicates the first step in the transient process The first load A work status begins. Indicates the first step in the transient process The first load The work status has ended; No. The load state switching situation of each transient process is used The corresponding candidate set of transient recognition results is represented by an array. express; The steps to obtain the final load identification result are as follows: S51, input parameters are: model and observation sequence The output is the recognition result of each hidden sequence. The observation sequence can be viewed as the electrical features extracted from the total user electricity consumption data for each steady-state process. These features are used for load identification based on steady-state characteristics. The hidden state corresponds to each steady-state process. ;Will The emission probability can be obtained by normalizing the objective function value corresponding to each result. ;Will The transition probability can be obtained by normalizing the objective function value corresponding to each result. ; S52, Initialization, Forward probabilities of each hidden state in S51 and its corresponding optimal path ,and ; S53, recursive calculation, the first The forward probability of step is ;in as well as And satisfy the formula The optimal path is ; S54, Termination: When recursively reaching the end of the observation sequence, calculate the maximum forward probability of the last step. and the corresponding path ; S55, optimal path backtracking, for Calculation path and the corresponding load identification results This result is the final load identification result obtained after verification and analysis.

2. The non-intrusive load monitoring method for steady-state and transient hybrid loads according to claim 1, characterized in that, The total electricity consumption data includes: active power, reactive power, voltage, current, and timestamp.

3. A system for performing the non-intrusive load monitoring method for steady-state and transient hybridization as described in any one of claims 1-2, characterized in that, It includes a data acquisition module, an event recognition module, a steady-state recognition module, a transient recognition module, and a result determination module; Data acquisition module: Collects total electricity consumption data from users through smart meters and constructs electricity consumption models for typical loads; Event recognition module: Uses event detection algorithms to identify events, classify them into steady-state and transient processes, and extract steady-state and transient features respectively; Steady-state identification module: uses a method based on steady-state characteristics to identify loads and obtains a candidate set of steady-state identification results based on the electricity consumption data in the electricity consumption model; Transient identification module: Uses a method based on transient features to identify loads and obtains a candidate set of transient identification results based on the electricity consumption data in the electricity consumption model; Result determination module: Construct a time-series path representing the changes in load operating status, use the Viterbi algorithm to filter the candidate sets of steady-state and transient identification results, and determine the final identification result.

Images