A load state recognition method based on improved KNN multi-label classification

Abstract

The present application relates to a kind of load state identification method based on improved KNN multi-label classification, to solve the problem of poor load state identification effect caused by the load event overlapping of multiple appliances in the load event detected from total power information.The present application mainly includes: by the duration of load event, the power change amount of load event termination time and starting time and the switching rule of appliance, filter the event that meets the condition from known event library;By KNN algorithm, give the load event label information that meets the confidence requirement;For the load event that does not meet the requirement, screen possible known event library event combination, and generate a series of load event curves, find the K load event curves of minimum distance, and complete the multi-label classification of load event according to the label of event;According to the label information given, complete load state identification.

Description

Technical Field

[0001] This invention belongs to the field of non-intrusive power load monitoring technology, and relates to a load status identification method based on improved KNN multi-label classification. Background Technology

[0002] In recent years, with the increasing prominence of global warming, energy issues, and environmental pressures, the construction of smart grids, as one of the important ways to solve these problems, has attracted more and more attention. The grid-connected operation of distributed clean energy sources and load demand response are crucial links in achieving sustainable development within the construction of smart grids, and effective load monitoring is a vital aspect of building smart grids.

[0003] Power load monitoring is divided into intrusive load monitoring and non-intrusive load monitoring (NILM). Unlike intrusive load monitoring, which requires the installation of digital sensors on each user's internal load equipment, non-intrusive load monitoring can collect user power consumption information by installing digital sensors at the user's power supply inlet and monitor the user's internal load power consumption through load status recognition algorithms.

[0004] NILM was first proposed by Har, an engineer at MIT, in the 1980s. After nearly 40 years of development, NILM has yielded many theoretical methods. One classic framework is: data acquisition – data preprocessing – load event detection and load feature extraction – load component and state identification. In this process, load event detection, load feature extraction, and load component and state identification are the core components of the load identification algorithm. Each appliance has unique load characteristics; after detecting load events from the overall user electricity consumption information, they need to be identified.

[0005] Currently, NILM based on this framework generally assumes that the extracted load events are formed by one type of appliance. However, in real life, there is a certain probability that the load characteristics of two or more appliances will overlap, which will lead to a decrease in the accuracy of the NILM algorithm. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing technologies and provide a load status identification method based on improved KNN multi-label classification, which can accurately identify overlapping load events and improve the identification effect of the NILM method.

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

[0008] A load state identification method based on improved KNN multi-label classification, characterized by the following steps:

[0009] Step 1: Use the data of the load status identification as the detection data, use change point monitoring to extract load events that can reflect the start or stop of the load and the occurrence time of the load events, and initialize the parameter settings based on expert experience, including the parameter settings of the number of nearest neighbors K, the nearest neighbor support k, the range change parameter β, and the confidence level α0.

[0010] Step 2: For the detected load events, calculate the duration T of the load event and the range of power change ΔP between the end time and the start time of the load event according to formula (1). In the known event library, find all known events whose duration and power change between the end time and the start time are within the range of variation to form a load event library to be selected.

[0011]

[0012] Where x is the duration T of the load event or the power change ΔP between the end and start times of the load event; β is the range variation parameter;

[0013] Step 3: Based on the occurrence time of the detected load events, remove events that do not meet the switching rules from the candidate load event library;

[0014] Step 4: Calculate the DTW distance between the detected load event and all events in the candidate load event library, and calculate the confidence level α according to formula (2). Select the K load events with the smallest DTW distance. For the load event with the most identical labeling information, if its maximum confidence level α < α0, assign the corresponding labeling information to the detected load event and proceed to step 10; otherwise, proceed to step 5.

[0015]

[0016] Among them, DTW(X) i ,X j () is event X i With Event X j The DTW distance, max(X) is the maximum value in event X, and length(X) is the duration of event X;

[0017] Step 5: Determine whether there is a step load segment with negative power change among the detected events. If so, proceed to step 6; otherwise, proceed to step 7.

[0018] Step 6: Assign labeling information to the step-type load segments detected in Step 5, the off events of appliances in the on state when the load event is detected, and the transition events between different states of multi-state appliances, in accordance with the method in Step 4. Then, remove the step-type load segments with negative power changes in the detected load events according to Equation (3).

[0019] P'(t)=|ΔP'|+P(t) t≥t' (3)

[0020] Where |ΔP'| is the absolute value of the power change of the step load segment, and t' is the termination time of the step load segment;

[0021] Step 7: First, remove events whose power change at the end and start times and whose duration exceeds the upper limit of the detected load event power change and duration. Combine the remaining events. Then, remove combinations that do not exceed the upper limit of the detected load event power change and duration from the new combinations. Continue combining until the combination contains 4 known load events or there are no new load event combinations. Filter all the known load event combinations according to Step 2.

[0022] Step 8: For each combination of load events, select the event with the shortest duration in the combination, and extract a series of load segments from the detected load events according to the duration of the event with a step size ΔT. Calculate the Euclidean distance between these extracted segments and the selected event, and select the position with the smallest Euclidean distance as the position where the event appears in the combination. Repeat the above operation until the positions of all events in the combination are determined. Use the samples existing in the sample library to generate a series of load event curves for the combination.

[0023] Step 9: Calculate the DTW distance between the detected load events and all combinations of load events, select the K load events with the smallest DTW distance, and count the number of times the label of each event appears. Assign the label of the load event that appears more than k times to the detected load event.

[0024] Step 10: Use the label information assigned to the detected events to complete the identification of the load status.

[0025] Furthermore, step 3 removes events that do not meet the switching rules from the candidate load event library based on the location of the detected load events. The specific steps are as follows:

[0026] Step 3-1: For appliances with only a single load state, if the appliance was in the on state before the detected load event, the selectable load event library should only contain the appliance's off event, and the appliance's on event should be removed from the selectable load event library. Otherwise, if the appliance was in the off state before the detected load event, the selectable load event library should only contain the appliance's on event, and the appliance's off event should be removed.

[0027] Step 3-2: For appliances with multiple load states, if the appliance was in the on state before the detected load event, the load event library to be selected can only contain the appliance's off event and the transition event from the current state to other states. Other events for that appliance in the load event library need to be removed. Otherwise, if the appliance was in the off state before the detected load event, the load event library to be selected can only contain the appliance's on event, and the appliance's off event and the transition event between different states need to be removed.

[0028] The advantages and beneficial effects of this invention are as follows:

[0029] This invention addresses the problem of the NILM model's difficulty in effectively identifying overlapping load events by designing a load status identification method based on improved KNN multi-label classification. This method significantly improves the accuracy of identifying overlapping load events and provides effective technical support for non-intrusive load monitoring based on event detection. Attached Figure Description

[0030] Figure 1 This is a flowchart of the load state recognition method based on the improved KNN multi-label classification of the present invention;

[0031] Figure 2 This is a schematic diagram illustrating the combination of load events according to the present invention. Detailed Implementation

[0032] The present invention will be further described in detail below through specific embodiments. The following embodiments are merely descriptive and not limiting, and should not be used to limit the scope of protection of the present invention.

[0033] This invention proposes a load state identification method based on improved KNN multi-label classification, and a detailed flowchart is shown below. Figure 1As shown, its innovation lies in the following aspects: First, by using the duration T of the load event, the power change ΔP between the end and start times of the load event, and the switching rules of the electrical appliances, events that meet the conditions are initially screened from the known event library; second, the KNN algorithm is used to assign label information to load events that meet the confidence requirements; for load events that do not meet the requirements, possible combinations of events from the known event library are screened, and a series of load event curves are generated. The K load event curves with the smallest distance are found, and multi-label classification of load events is completed based on the event labels; finally, load status identification is completed based on the assigned label information.

[0034] The above flowchart can be used to implement a load state identification method based on improved KNN multi-label classification, which mainly includes the following steps:

[0035] Step 1: Use the data of the load status identification as the detection data, and use variable point monitoring to extract load events that can reflect the start or stop of the load and the occurrence time of the load events. Based on expert experience, initialize the parameter settings including the number of nearest neighbors K, the nearest neighbor support k, the range change parameter β, and the confidence level α0.

[0036] Step 2: For the detected load events, calculate the duration T of the load event and the range of power change ΔP between the end time and the start time of the load event according to formula (1). In the known event library, find all known events whose duration and power change between the end time and the start time are within the range of variation to form a load event library to be selected.

[0037]

[0038] Where x is the duration T of the load event or the power change ΔP between the end and start times of the load event;

[0039] Step 3: Based on the occurrence time of the detected load events, remove events that do not meet the switching rules from the candidate load event library. The specific steps are as follows:

[0040] Step 3-1) For appliances with only a single load state, if the appliance was in the on state before the detected load event, then the selectable load event library can only contain the appliance's off event, and the appliance's on event needs to be removed from the selectable load event library; otherwise, if the appliance was in the off state before the detected load event, then the selectable load event library can only contain the appliance's on event, and the appliance's off event needs to be removed.

[0041] Step 3-2) For appliances with multiple load states, if the appliance was in the on state before the detected load event, the load event library to be selected can only contain the appliance's off event and the transition event from the current state to other states. Other events for the appliance in the load event library to be selected need to be removed. Otherwise, if the appliance was in the off state before the detected load event, the load event library to be selected can only contain the appliance's on event, and the appliance's off event and the transition event between different states need to be removed.

[0042] Step 4: Calculate the DTW distance between the detected load event and all events in the candidate load event library, and calculate the confidence level α according to formula (2). Select the K load events with the smallest DTW distance. For the load event with the most identical labeling information, if its maximum confidence level α < α0, assign the corresponding labeling information to the detected load event and proceed to step 10; otherwise, proceed to step 5.

[0043]

[0044] Among them, DTW(X) i ,X j () is event X i With Event X j The DTW distance, max(X) is the maximum value in event X, and length(X) is the duration of event X;

[0045] Step 5: Determine whether there is a step load segment with negative power change among the detected events. If so, proceed to step 6; otherwise, proceed to step 7.

[0046] Step 6: Assign labeling information to the step-type load segments detected in Step 5, the off events of appliances in the on state when the load event is detected, and the transition events between different states of multi-state appliances, in accordance with the method in Step 4. Then, remove the step-type load segments with negative power changes in the detected load events according to Equation (3).

[0047] P'(t)=|ΔP'|+P(t) t≥t' (3)

[0048] Where |ΔP'| is the absolute value of the power change of the step load segment, and t' is the termination time of the step load segment;

[0049] Step 7: First, remove events whose power change at the known event termination time and start time, and whose duration exceeds the upper limit of the detected load event power change and duration. Combine the remaining events. Then, remove combinations from the new combinations that do not exceed the upper limit of the detected load event power change and duration. Continue combining until the combination contains four known load events or no new load event combinations are found. An example diagram of load event combinations is shown below. Figure 2 As shown, all known load event combinations obtained are filtered according to step 2;

[0050] Step 8: For each combination of load events, select the event with the shortest duration in the combination, and extract a series of load segments from the detected load events according to the duration of the event with a step size ΔT. Calculate the Euclidean distance between these extracted segments and the selected event, and select the position with the smallest Euclidean distance as the position where the event appears in the combination. Repeat the above operation until the positions of all events in the combination are determined. Use the samples existing in the sample library to generate a series of load event curves for the combination.

[0051] Step 9: Calculate the DTW distance between the detected load events and all combinations of load events, select the K load events with the smallest DTW distance, and count the number of times the label of each event appears. Assign the label of the load event that appears more than k times to the detected load event.

[0052] Step 10: Use the label information assigned to the detected events to complete the identification of the load status.

[0053] Examples of research materials:

[0054] Precision (P), recall (R), and F1 score were used to evaluate the load state identification method based on improved KNN multi-label classification. The calculation methods for precision (P), recall (R), and F1 score are as follows:

[0055]

[0056]

[0057]

[0058] Wherein, TP represents the number of actual load events for appliance a and the number of predicted load events for appliance a; FP represents the number of actual load events for appliance a and the number of predicted load events for appliance a; and FN represents the number of actual load events for appliance a and the number of predicted load events for appliance a.

[0059] Using a household's actual electricity consumption data for one week as test data, this method was used to identify the load status, and the following test results were obtained:

[0060] refrigerator 0.94 0.92 0.93 washing machine 0.93 0.85 0.89 electric water heater 0.96 0.95 0.95 TV set 0.90 0.88 0.89 rice cooker 0.94 0.94 0.94 induction cooker 0.91 0.90 0.90 oven 0.96 0.94 0.95 Range hood 0.84 0.80 0.82 Micro-wave oven 0.90 0.91 0.91 Incandescent lamp 0.94 0.88 0.91 air conditioner 0.86 0.88 0.87

[0061] As can be seen from the above research materials, the method of the present invention has a good load identification effect.

[0062] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0063] 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.

[0064] 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.

[0065] 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.

[0066] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A load state identification method based on improved KNN multi-label classification, characterized in that: Includes the following steps: Step 1: Use the data from the load status identification as detection data. Utilize variable point monitoring to extract load events that reflect load on / off states and their occurrence times. Initialize parameter settings, including the number of nearest neighbors. Nearest Neighbor Support Range variation parameters and confidence level Including parameter settings; Step 2: For the detected load events, calculate the duration of the load events according to formula (1). and the power change between the end and start times of the load event. The range of variation is determined by searching the known event library for events whose duration and the power change between their start and end times fall within the specified range. This forms the candidate load event library. (1) in, It is the duration of the load event. Or the power change between the end and start times of a load event. ; It is a range-varying parameter; Step 3: Based on the occurrence time of the detected load events, remove events that do not meet the switching rules from the candidate load event library; Step 4: Calculate the DTW distance between the detected load event and all events in the candidate load event library, and calculate the confidence level according to equation (2). Select the one with the smallest DTW distance. For each load event, the load event with the most identical labeling information, if its highest confidence level is... If the event is detected, the corresponding labeling information will be assigned to the detected load event, and the process will proceed to step 10; otherwise, the process will proceed to step 5. (2) in, For the event With the event DTW distance, For the event The maximum value in, For the event Duration; Step 5: Determine whether there is a step load segment with negative power change among the detected events. If so, proceed to step 6; otherwise, proceed to step 7. Step 6: Assign labeling information to the step-type load segments detected in Step 5, the off events of appliances in the on state when the load event is detected, and the transition events between different states of multi-state appliances, in accordance with the method in Step 4. Then, remove the step-type load segments with negative power changes in the detected load events according to Equation (3). (3) in, This represents the absolute value of the power change in a step-type load segment. This is the termination time of the step-type load segment; Step 7: First, remove events whose power change at the end and start times and whose duration exceeds the upper limit of the detected load event power change and duration. Combine the remaining events. Then, remove combinations that do not exceed the upper limit of the detected load event power change and duration from the new combinations. Continue combining until the combination contains 4 known load events or there are no new load event combinations. Filter all the known load event combinations according to Step 2. Step 8: For each combination of load events, select the event with the shortest duration in the combination, and then apply the remaining duration in increments. Extract a series of load segments from the detected load events, calculate the Euclidean distance between these extracted segments and the selected events, select the position with the smallest Euclidean distance as the position where the event appears in the combination, repeat the above operation in step 8 until the position of all events in the combination is determined, and generate a series of load event curves for the combination using samples existing in the known sample library. Step 9: Calculate the DTW distance between the detected load event and all combined load events, and select the one with the smallest DTW distance. Each load event is counted, and the number of times each event's label appears is recorded. Events with more than one occurrence are then counted. The load event label is assigned to the detected load event; Step 10: Use the label information assigned to the detected events to complete the identification of the load status. Furthermore, step 3, based on the time of occurrence of the detected load event, removes events from the candidate load event library that do not meet the switching rules. The specific steps are as follows: Step 3-1: For appliances with only a single load state, if the appliance was in the on state before the detected load event, the selectable load event library should only contain the appliance's off event, and the appliance's on event should be removed from the selectable load event library. Otherwise, if the appliance was in the off state before the detected load event, the selectable load event library should only contain the appliance's on event, and the appliance's off event should be removed. Step 3-2: For appliances with multiple load states, if the appliance was in the on state before the detected load event, the load event library to be selected can only contain the appliance's off event and the transition event from the current state to other states. Other events for that appliance in the load event library need to be removed. Otherwise, if the appliance was in the off state before the detected load event, the load event library to be selected can only contain the appliance's on event, and the appliance's off event and the transition event between different states need to be removed.

Images