CN113687164B - Non-intrusive load event two-stage self-adaptive detection method - Google Patents
Non-intrusive load event two-stage self-adaptive detection method Download PDFInfo
- Publication number
- CN113687164B CN113687164B CN202110984291.2A CN202110984291A CN113687164B CN 113687164 B CN113687164 B CN 113687164B CN 202110984291 A CN202110984291 A CN 202110984291A CN 113687164 B CN113687164 B CN 113687164B
- Authority
- CN
- China
- Prior art keywords
- load
- load event
- event
- power
- events
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Theoretical Computer Science (AREA)
- General Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Marketing (AREA)
- Tourism & Hospitality (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Remote Monitoring And Control Of Power-Distribution Networks (AREA)
- Complex Calculations (AREA)
Abstract
The invention discloses a two-stage self-adaptive detection method for a non-invasive load event, which comprises the following steps: calculating to obtain an adaptive threshold value of each moment by utilizing the original load total power data based on outlier analysis so as to dynamically adapt to the fluctuation degree of the steady-state section of the electrical signal, and judging whether a load event occurs or not; step type load event detection of a first stage is carried out by improving an edge detection method and an adjacent load event segmentation method; taking signals of all sections except the step-type load event in the total load power data as detection objects, and performing long transient state type load event detection in the second stage by using a method of combining the moving average with the sliding t test; and collecting all events detected in the two stages, screening one by one, and identifying the load events with the starting and stopping point power difference smaller than the self-adaptive threshold as pseudo load events and eliminating the pseudo load events, thereby completing the self-adaptive detection of the load events.
Description
Technical Field
The invention belongs to the field of load event detection of non-invasive load monitoring, and particularly relates to a two-stage self-adaptive detection method of a non-invasive load event.
Background
Under the unified deployment of strategic goals of carbon peaking and carbon neutralization, the construction of the intelligent power grid in China is continuously and deeply promoted. Non-invasive load monitoring as an important technology in the field of advanced measurement of smart grids [1] The method comprises the following steps that (NILM) power utilization information of each (main) electrical appliance in a user can be obtained only through analysis of Load power utilization total data, implementation basis can be provided for taking measures of series energy efficiency upgrading, energy saving and emission reduction, and therefore the aim of carbon neutralization is achieved. The existing NILM method can be generally divided into an event method and a non-event method according to different implementation principles [2] 。
Load event detection is the fundamental link of the event-based NILM method. The load event is a conversion process of the working state of the electric appliance, and comprises the switching of the on and off of the electric appliance and the switching between the non-zero working states of power. Transient section and steady section of load electricity consumption total data can be divided through load event detection, and transient and steady characteristics of the electric appliance can be extracted according to the transient section and the steady section [1] . Therefore, the accuracy of event detection is directly related to the follow-up load characteristicThe accuracy of the acquisition is improved, so that the final effects of the state identification and the load decomposition of the electric appliance are influenced. However, most of the studies reported for NILM do not perfectly solve the load event detection problem, and the major difficulties and challenges are:
(1) The detection effect is sensitive to the amplitude threshold. Most documents [1][3] The load event detection is executed by adopting a fixed threshold, and when the threshold is set to be too small, the electric signal fluctuation caused by noise or the running characteristic of an electric appliance is easily detected as an event by mistake; when the threshold is set too large, the load event of the appliance with less power is easily missed.
(2) It is difficult to detect the complete transient process of different load events. Some methods detect a single change point in the time series of the total signal of the load [4] Only the approximate time of occurrence of the load event can be determined [5] . However, event detection can further achieve accurate extraction of the steady-state features and complete expression of the corresponding transient features only if complete capture of the transient segment is achieved, i.e., the start and end times thereof are determined.
(3) It is difficult to separate different events (Simultaneous events) in the vicinity from each other. The adjacent load events refer to different load events with very short time intervals, and the design method of the prior document is mostly based on the principle of switch continuity [1] The theoretical assumption of (Switch Continuity Principle) assumes that only one event occurs in a short time interval and therefore the proximity load event is directly detected as an event. Event detection method, in particular for use with sliding windows [3][6][7] In theory, it is not possible to resolve different events with a time interval smaller than the window length.
Therefore, in order to solve the above problems, a non-intrusive load event detection method that can adapt to the background noise or electrical signal fluctuation and detect different types of load event transient is needed.
[ reference documents ]
[1]G.W.Hart.Nonintrusive appliance load monitoring[J].Proceedings of the IEEE,1992,80(12):1870-1891.
[2]K D Anderson,M E Bergés,A Ocneanu,et al.Event detection for Non Intrusive load monitoring[C].2012-38th Annual Conference on IEEE Industrial Electronics Society,Montreal,2012:3312-3317.
[3]M Lu,Z Li.A Hybrid Event Detection Approach for Non-Intrusive Load Monitoring[J].IEEE Transactions on Smart Grid,2020,11(1):528-540.
[4]B Wild,K S Barsim,B Yang.A new unsupervised event detector for non-intrusive load monitoring[C].2015IEEE Global Conference on Signal and Information Processing(GlobalSIP),Orlando,FL,2015,73-77.
[5]H.A.D.Azzini,R.Torquato and L.C.P.da Silva,"Event detection methods for nonintrusive load monitoring[C].2014IEEE PES General Meeting Conference&Exposition,National Harbor,MD,2014:1-5
[6]A U Rehman,T T Lie,B Vallès,et al.Event-Detection Algorithms for Low Sampling Nonintrusive Load Monitoring Systems Based on Low Complexity Statistical Features[J].IEEE Transactions on Instrumentation and Measurement,2020,69(3):751-759.
[7]Yi S,Yin X,Diao Y,et al.A New Event-detection Method Based on Composite Windows in NILM for Industrial Settings[C].IEEE Sustainable Power and Energy Conference(iSPEC).IEEE,2019:2768-2771.
Disclosure of Invention
Aiming at the prior art, the invention provides a two-stage self-adaptive detection method for a non-invasive load event, and simultaneously, a two-stage self-adaptive detection framework for the non-invasive load event is designed based on the geometrical characteristics of the load event on a load total power curve. As shown in fig. 1 below, the frame is composed of four functional modules. The system specifically comprises an adaptive threshold setting module, a step type load event detection module, a long transient state type load event detection module and a load event post-processing module. And after the four modules start, the four modules are sequentially executed to jointly complete the self-adaptive detection function of the load event.
In order to solve the above technical problem, the present invention provides a two-stage adaptive detection method for non-intrusive load event, which comprises the following steps:
step one, self-adaptive threshold setting based on outlier analysis: calculating to obtain a self-adaptive power change threshold value at each moment based on the original load total power data, and judging whether a load event occurs or not; setting a threshold value through outlier analysis, and enabling the threshold value to be dynamically adaptive to the fluctuation degree of the steady-state section of the electrical signal so as to improve the detection capability of load events with different amplitudes;
step two, step type load event detection: step type load event detection is carried out in a first stage by improving an edge detection method and an adjacent load event segmentation step;
step three, detecting a long transient load event: taking signals of all sections except the step-type load event in the total load power data as detection objects, and performing long transient state type load event detection in the second stage by using a method of combining the moving average with the sliding t test;
step four, post-processing of load events: and collecting the step type load events and the long transient type load events measured in the first stage and the second stage, screening each detected load event, and regarding the load events with the power difference between the start point and the stop point smaller than the self-adaptive threshold as pseudo load events and eliminating the pseudo load events.
Further, the invention provides a two-stage adaptive detection method for non-intrusive load events, wherein the method comprises the following steps:
the specific steps of the first step are as follows:
firstly, a load total power time sequence { P (T) | T =1,2, 3.. The T } with the length of T is calculated according to the formula (1) to obtain a difference sequence { delta P (T) | T =1,2, 3.. The T-1}
ΔP(t)＝P(t+1)-P(t) (1)
Then, detecting and eliminating outliers larger than the sequence average value for a load total power first-order difference absolute value sequence { delta P (T) | T =1,2, 3.. The., T-1} by using a sliding window; and (3) detecting outliers by applying a Laeya criterion, and performing iterative computation according to a formula (2):
|ΔP(t outlier )|＞μ(|ΔP(t i )|)+3·σ(|ΔP(t i )|)(2)
in the formula (2), μ (-) and σ (-) are the average and standard deviation calculation, respectively, t i Is the ith length of W 1 At a point in time, t, within the sliding window outlier Time point of outlier, t outlier ∈{t i ,W 1 ·(i-1)+1≤t i ≤W 1 I }; detecting that outliers are removed immediately until no outliers meeting the conditions exist in the current window; for the detected outlier, replacing the outlier by the arithmetic mean of the adjacent data still reserved before and after the outlier to obtain an updated sequence { | Δ P' (t) | };
finally, with each time as the center, selecting the length W 2 And satisfies W 2 ＜W 1 (ii) a Then, the self-adaptive threshold value of the moment is preliminarily calculated according to the power fluctuation condition in the window
As shown in formula (3):
in the formula (3), t-W 2 /2≤t i ≤t+W 2 K is a proportionality coefficient, k is 3; the adaptive threshold setting results are shown in equation (4):
in the formula (4), D th_min And D th_max Upper and lower adaptive threshold limits, respectively.
In the second step, the load event detection in the first stage is completed through the detection and judgment of the step-type load event, and the method comprises the following steps:
step 2-1) extracting load events from the total load power time sequence by using an improved edge detection method:
firstly, a power change trend satisfying the formula (5) is searched in a time sequence of total power of the loadPotential change point t tp ，
t tp ＝{t | sgn(ΔP(t))≠sgn(ΔP(t+1))} (5)
In formula (5), sgn (-) is a sign function with a value range of { -1,0,1 };
during scanning the total power sequence of the load, once the ith power change trend change point
If the requirement of formula (6) is satisfied, the product is determined
Is the start time of a load event;
in the following scanning process, if the power variation trend changes at the ith' power variation trend changing point
To
If there is no point satisfying the formula (6) in the time range, then
Will be identified as the end time of this load event; and from this moment a search for the start moment of the next new load event is deployed; wherein, t th A length threshold value of a steady-state section outside a section where a load event is located in a time sequence of the total load power;
step 2-2) adjacent load event segmentation:
the adjacent load events refer to different load events with very short time intervals, and are divided into two conditions of same direction and reverse direction according to the power change direction relation of the two adjacent load events;
for segmentation of reverse adjacent load events, the steps are as follows:
step 1: for duration greater than preset threshold value delta t th Performs piece-wise Linear Representation (PLR); for the ith load event E (i) The sequence of load event powers is
Thereby obtaining a corresponding sequence of segmentation points
Step 2: taking the power extreme point in the segment point sequence as a boundary point to the power sequence of the load event
And performing segmentation, wherein the judgment condition of the power extreme point is as follows:
and 3, step 3: if the boundary point exists, determining the starting and stopping time of a new load event in each segmented power sequence, otherwise, jumping to the step 5; let the power variation between the start and stop moments of the current power segment be Δ P seg In this section, the absolute value of the power variation is found to be greater than α · | Δ P seg Taking two time points with the shortest time interval as the starting and stopping time of the candidate newly-added load event; wherein, alpha is a given proportionality coefficient;
and 4, step 4: merging the candidate load events which are connected end to end into one load event for subsequent processing;
and 5, step 5: repeating the steps 1 to 4 until all load events needing to be segmented are segmented;
for the segmentation of the equidirectional adjacent load events, the steps are as follows:
for load event E (i) Power sequence
The corresponding difference sequence is obtained and calculated,
calculating the power change between adjacent target elements, and if equation (9) is satisfied, determining the load event power sequence
Dividing the two target elements, and respectively using the divided power sequences as the power sequences of two different newly-added load events;
in the formula (9), β is a defined scaling factor threshold value, and a smaller value thereof indicates a more severe segmentation condition,
respectively one before and one after the adjacent target elements,
step 2-3) load event judgment, namely judging step type load events in all the load events obtained in the step 2-1) and the step 2-2) and taking the step type load events as the detection results of the first-stage load events, wherein the step type load events comprise the following steps:
if a load event E (i) If equation (10) is satisfied, the load event can be determined to be a step-type load event;
in the formula (10), N th Setting a sample point number threshold; Δ P (t) is the sequence
Of (1), t' outlier Is a sequence of
Where the time scale of outliers greater than the mean of the sequence, gamma is a defined scaling factor threshold,
for power changes between the start and stop times of the load event.
The third step comprises the following specific steps:
and performing second-stage load event detection on signals of all sections except the step-type load event in the load power data: scanning the residual load total power sequence point by point, if the current time point meets the requirement of the formula (11) and the previous time point is in a steady-state section, judging that a load event transient section is entered from the current time point and continuing to scan,
in the formula (11), for each time point t,
and
respectively is the average value of the power in the front scanning window and the rear scanning window of the point;
respectively the standard deviation of the power sequence under the corresponding window; the window length of the front and rear scanning windows satisfies W before ＝W after = w; at a given level of significance
In the case of the above-described situation,
for the threshold value of t-test quantity obtained by table lookup,
has a degree of freedom of 2w-2, wherein
Taking the value as 0.05;
if equation (12) holds at time t and then t th If the expression (11) is not established any more in the time length range, the moment t is judged to be the ending moment of the transient section, namely, a new steady-state section is entered from the moment;
the fourth step comprises the following specific steps:
aggregating the step-type load events detected in step two and the long transient-type load events detected in step three, and for each detected load event E (i) Screening according to the formula (13), and determining the load events with the start and stop point power difference smaller than the self-adaptive threshold as pseudo load events and removing the pseudo load events;
in equation (13), the steady-state segment length threshold t th Set to 5, scaling factor thresholds α, β, γ are set to 0.75,0.2,0.75, N, respectively th Set to 5.
Meanwhile, the invention also provides a system for realizing the non-invasive load event two-stage self-adaptive detection method, which comprises a self-adaptive threshold setting module, a step type load event detection module, a long transient type load event detection module and a load event post-processing module;
the self-adaptive threshold setting module is used for calculating a self-adaptive power change threshold at each moment based on the original load total power data and judging whether a load event occurs or not; power change caused by a load event is eliminated from fluctuation of a total load signal through outlier analysis, so that a threshold value is dynamically adaptive to the fluctuation degree of a steady-state section of an electrical signal, and the detection capability of the load event with different amplitudes is improved;
the step type load event detection module is used for detecting the load event in the first stage by improving an edge detection method and adjacent load event segmentation, judging the step type load event in the step type load event, extracting and marking a corresponding transient process section from a load total power time sequence, and directly sending the transient process section into a subsequent load event post-processing step;
the long transient load event detection module detects the load event at the second stage by using signals of all sections except the step load event in the total load power data as detection objects and by using a method combining the sliding average and the sliding t test to detect the non-step load event; the non-step load event is called a long transient load event;
the load event post-processing module screens all load events including step type and long transient state type load events and eliminates pseudo load events of which the power change is smaller than a set adaptive threshold.
Compared with the prior art, the invention has the beneficial effects that:
the invention provides a two-stage self-adaptive detection framework and a method for a non-intrusive load event. Firstly, the self-adaptive threshold value is set to have high detection capability on a low-power amplitude load event, and meanwhile, the false detection condition caused by electric signal fluctuation is also obviously reduced; secondly, the two-stage processing mode can well deal with different geometrical shape characteristics of a step type load event and a long transient state type load event, fully considers the power change trend in the load event and realizes the complete capture of the transient state process of the electric appliance; meanwhile, the detection capability of the adjacent load event which is easy to occur in the low-frequency data is improved. The method has higher load event detection accuracy and keeps stronger robustness in different user scenes.
Drawings
FIG. 1 is a framework for two-phase adaptive detection of non-intrusive load events in accordance with the present invention;
FIG. 2 is a graph of the total power of a user's load in one day according to an embodiment of the present invention;
FIG. 3 is the result of adaptive threshold calculation in the present invention;
FIG. 4 is a graph showing the results of piecewise linearization in the present invention;
FIG. 5 is a power waveform of a refrigerator and a water heater operating simultaneously according to an embodiment of the present invention;
fig. 6 is the load event detection result of the present invention.
Detailed Description
The invention will be further described with reference to the following drawings and specific examples, which are not intended to limit the invention in any way.
The invention provides a non-intrusive load event two-stage self-adaptive detection system, wherein a frame structure of the system mainly comprises four functional modules: the device comprises an adaptive threshold setting module, a step type load event detection module, a long transient type load event detection module and a load event post-processing module. The four modules are started in advance and then sequentially executed to jointly complete the self-adaptive detection function of the load event, as shown in the following figure 1.
And the self-adaptive threshold setting module is used for calculating a self-adaptive power change threshold at each moment based on the original load total power data and judging whether a load event occurs or not. The power change caused by the load event is eliminated from the fluctuation of the total load signal as much as possible through outlier analysis, so that the threshold value can be accurately and dynamically adaptive to the fluctuation degree of the steady-state section of the electrical signal, and the detection capability of the load event with different amplitudes is improved.
The step type load event detection module detects the load event in the first stage by improving an edge detection method and an adjacent load event segmentation step, judges the step type load event, extracts and marks a corresponding transient process section from a total signal, and directly sends the transient process section to a subsequent load event post-processing step.
The long transient load event detection module takes signals of all sections except the step type load event in the total load power data as detection objects, and performs load event detection of a second stage by using a method of combining a sliding average with a sliding t test so as to detect a non-step type load event. Wherein the non-step load event exhibits a shape characteristic of a long transient, also referred to herein as a long transient load event.
And the load event post-processing module screens all load events including step type and long transient state load events and eliminates pseudo load events with small power change.
Based on the framework, the invention provides a non-intrusive load event two-stage self-adaptive detection method, namely a specific implementation method is provided for main modules in the framework, and the method is as follows:
step one, self-adaptive threshold setting based on outlier analysis: calculating to obtain a self-adaptive power change threshold value at each moment based on the original load total power data, and judging whether a load event occurs or not; setting a threshold value through outlier analysis, and enabling the threshold value to be dynamically adaptive to the fluctuation degree of the steady-state section of the electrical signal so as to improve the detection capability of load events with different amplitudes;
step two, step type load event detection: step type load event detection is carried out in a first stage by improving an edge detection method and an adjacent load event segmentation step;
step three, detecting a long transient load event: taking signals of all sections except the step-type load event in the total load power data as detection objects, and performing long transient state type load event detection in the second stage by using a method of combining the moving average with the sliding t test;
step four, post-processing of load events: and collecting the step type load events and the long transient type load events measured in the first stage and the second stage, screening each detected load event, and identifying the load events with the starting and stopping point power difference smaller than the adaptive threshold as pseudo load events and eliminating the pseudo load events.
The load event detection threshold in the present invention refers to a power change threshold, i.e. it is possible to indicate the occurrence of a load event only if the degree of change of the total power signal of the load exceeds the threshold.
The details of each step are described in detail below.
The self-adaptive threshold setting method based on outlier analysis comprises the following steps:
firstly, a load total power time sequence { P (T) | T =1,2, 3.. The T } with the length of T is calculated according to the formula (1) to obtain a difference sequence { delta P (T) | T =1,2, 3.. The T-1}
ΔP(t)＝P(t+1)-P(t) (1)
Next, a sequence of first order difference absolute values of the total power of the load { Δ P (T) | T =1,2, 3.. The, T-1} detects and rejects single-sided outliers (large values) using a sliding window. Theoretically, a plurality of different data outlier detection methods can be selected, and the outlier is detected by applying the Laplace criterion, namely the iterative computation formula (2):
|ΔP(t outlier )|＞μ(|ΔP(t i )|)+3·σ(|ΔP(t i )|)(2)
wherein, mu (-) and sigma (-) are respectively the calculation of average value and standard deviation, t i Is the ith length of W 1 At a point in time, t, within the sliding window outlier Time point of outlier, t outlier ∈{t i ,W 1 ·(i-1)+1≤t i ≤W 1 I }. And detecting outliers and then eliminating the outliers until no outliers meeting the conditions exist in the current window. For the detected outlier, the arithmetic mean of the adjacent data still kept before and after the outlier is used for replacing the outlier, and the updated sequence { | Δ P' (t) | } is obtained.
Finally, with each time as the center, selecting the length W 2 And satisfies W 2 ＜W 1 . Then, the power fluctuation situation in the window is used for preliminarily calculating the timeAdaptive threshold
As shown in formula (3).
Wherein, t-W 2 /2≤t i ≤t+W 2 The/2-1, k is the proportionality coefficient, and the present invention is set to 3. Meanwhile, in order to avoid load event detection omission caused by an overlarge threshold value or detect power change with an undersized amplitude value as a load event, the invention limits the minimum value and the maximum value of the self-adaptive threshold value, and the final result is shown as a formula (4).
Wherein D is th_min And D th_max Upper and lower adaptive threshold limits, respectively.
In the present invention, the window length W 1 ，W 2 300s and 30s, respectively, threshold boundary [ D ] th_min ,D th_max ]Is [10,60 ]]W. Therefore, for a certain user total power load data shown in fig. 2, a load event detection threshold that can be adapted to the degree of total power load fluctuation as shown in fig. 3 can be calculated.
Step two, step type load event detection
The invention completes the first stage load event detection by detecting and judging the step type load event, and mainly comprises the following three steps.
1) Load events are extracted using an improved edge detection method. The invention changes the detection logic of 'difference making point by point' of the traditional edge detection algorithm and evaluates the power difference between the power change trend change points. Firstly, searching a power change trend change point t satisfying the formula (5) in a load total power sequence tp 。
t tp ＝{t | sgn(ΔP(t))≠sgn(ΔP(t+1))} (5)
Wherein sgn (. Cndot.) is a sign function having a value range of { -1,0,1 }.
During scanning the total power sequence of the load, once the ith power change trend change point
If the requirement of equation (6) is satisfied, it is identified as the starting time of a load event.
In the following scanning process, if the power variation trend changes at the ith' power variation trend changing point
To
If there is no point satisfying the formula (6) in the time range, then
Will be identified as the end time of this load event. And from this point on a search for the start time of the next new load event is deployed. Wherein, t th Is a steady state segment length threshold.
2) And (4) dividing the adjacent load events. The adjacent load events refer to different load events with very short time intervals, which can be falsely detected as one load event by an edge detection method, and can be divided into two conditions of same direction and reverse direction according to the power change direction relation of the two adjacent load events. The segmentation for reverse proximity loading events essentially comprises the steps of:
step 1: for duration longer than preset threshold value delta t th The load event power sequence (set to 3 samples) performs a piece-wise Linear Representation (PLR). For the ith load event E (i) The power sequence thereof is
Corresponding segmented point sequences can be obtained
Taking fig. 4 as an example, the load event power sequence covered by the thick line in the figure can obtain the segmented linearization result shown by the dotted line.
Step 2: taking the power extreme point in the segment point sequence as a boundary point to the power sequence of the load event
And (6) carrying out segmentation. The judgment condition of the power extreme point is as follows:
and 3, step 3: and if the boundary point exists, determining the starting and stopping time of the new load event in each segmented power sequence, otherwise, jumping to the step 5. Setting the power change between the start and stop moments of the current power section as delta P seg In this section, the absolute value of the power change is sought to be greater than α · | Δ P seg And l, taking two time points with the shortest time interval as the starting and stopping time of the candidate newly-added load event. Where α is a given scaling factor.
And 4, step 4: in order to avoid mistakenly dividing the spike pulse in the total load power and the starting spike power transient of part of the electric appliances into two load events, the candidate load events which are connected end to end and are stopped together are combined into one load event for subsequent processing.
And 5, step 5: and repeating the steps 1-4 until the judgment of all the load events is completed.
Segmentation for equidirectional adjacent load events: first, for a load event E (i) Power sequence
Finding corresponding difference sequences
The target element with larger absolute value is selected according to the formula (8):
then, the power variation between adjacent target elements is calculated, such as the first target element
With a second target element
Change in power therebetween
Fig. 5 shows power waveforms when a refrigerator and a water heater are operated simultaneously.
Finally, if equation (9) holds, the load event power sequence is applied
Two different newly added load events are divided and respectively drawn between the two target elements.
Where β is a defined scaling factor threshold, and smaller values indicate more severe segmentation conditions.
3) And judging a step type load event. And judging step type load events in all the load events obtained in the two steps and taking the step type load events as a first-stage load event detection result. Specifically, depending on the shape characteristics of the step-type load event-a significant change in power change occurs over a short time interval if load event E (i) If equation (10) is satisfied, it can be determined as a step-type load event.
Wherein, N th Setting a sample point number threshold; Δ P (t) is the sequence
Element of (1), t o ′ utlier Is a sequence of
The time scale of the medium unilateral (greater) outliers, gamma is a defined scaling factor threshold,
for power changes between the start and stop times of the load event.
Step three, detecting long transient load event
And performing second-stage load event detection on signals of all sections except the step-type load event in the load power data: scanning the residual load total power sequence point by point, if the current time point satisfies the requirement of formula (11) and the previous time point is in a steady-state section, then judging to enter a load event transient section from the current time point and continuing to scan,
wherein, for each point in time t,
and
respectively the average power values in the front scanning window and the rear scanning window of the point;
respectively the standard deviation of the power sequence under the corresponding window; the window length of the front and rear scanning windows arranged in the paper satisfies W before ＝W after = w; at a given level of significance
In the case of a situation in which,
the degree of freedom is 2w-2, wherein the t-test quantity critical value can be obtained by table look-up
Take 0.05.
If equation (12) holds at time t and then t th If the expression (11) is no longer true in the time length range, the time t is determined as the ending time of the transient section, that is, a new steady-state section is entered from the time t.
Step four, load event post-processing
And collecting all load events detected in the two stages, and performing unified post-processing. For each load event detected E (i) And (4) screening according to the formula (13), and identifying the load events with the start-stop point power difference smaller than the adaptive threshold as pseudo load events and eliminating the pseudo load events.
The above-mentioned parameters are set as follows: steady state zone length threshold t th Is set to 5, Δ t th Set to 3, scaling factor thresholds α, β, γ are set to 0.75,0.2,0.75, respectively, N th Set to 5. By using the above method, load event detection is performed on the total power data of the load shown in fig. 2, wherein the partial load event detection result is shown in fig. 6, and the load event section is covered by a circle. The method disclosed by the invention can be found to have a good detection effect on both step type and long transient state type load events.
Although the present invention has been described in connection with the accompanying drawings, the present invention is not limited to the above-described embodiments, which are intended to be illustrative rather than restrictive, and many modifications may be made by those skilled in the art without departing from the spirit of the present invention as disclosed in the appended claims.
Claims (5)
1. A two-stage adaptive detection method for non-intrusive load events, the method comprising:
step one, self-adaptive threshold setting based on outlier analysis: calculating to obtain a self-adaptive power change threshold value at each moment based on the original load total power data, and judging whether a load event occurs or not; setting a threshold value through outlier analysis, and enabling the threshold value to be dynamically adaptive to the fluctuation degree of the steady-state section of the electrical signal so as to improve the detection capability of load events with different amplitudes; the method comprises the following specific steps:
firstly, a load total power time sequence { P (T) | T =1,2, 3.. Times, T } with a length of T is calculated according to the formula (1) to obtain a difference sequence { Δ P (T) | T =1,2, 3.. Times, T-1}
ΔP(t)＝P(t+1)-P(t) (1)
Then, detecting and eliminating outliers larger than the average value of the sequences by utilizing a sliding window for a load total power first-order difference absolute value sequence { | | delta P (T) | | T =1,2,3,. The.. And T-1 }; and (3) detecting outliers by applying a Laeya criterion, and performing iterative computation according to a formula (2):
|ΔP(t outlier )|＞μ(|ΔP(t i )|)+3·σ(|ΔP(t i )|) (2)
in the formula (2), μ (-) and σ (-) are the average value and standard deviation calculation, respectively, t i Is the ith length of W 1 At a point in time, t, within the sliding window outlier Time point of outlier, t outlier ∈{t i ,W 1 ·(i-1)+1≤t i ≤W 1 I }; detecting that outliers are removed immediately until no outliers meeting the conditions exist in the current window; for detected outliers, use their predecessorsReplacing the arithmetic mean value of the adjacent data still remained later to obtain an updated sequence { | delta P' (t) | };
finally, with each time as the center, selecting the length as W 2 And satisfies W 2 ＜W 1 (ii) a Then, the self-adaptive threshold value of the moment is preliminarily calculated according to the power fluctuation condition in the window
As shown in formula (3):
in the formula (3), t-W 2 /2≤t i ≤t+W 2 K is a proportionality coefficient, k is 3; the adaptive threshold setting result is shown in equation (4):
in the formula (4), D th_min And D th_max Upper and lower adaptive threshold limits, respectively;
step two, step type load event detection: step type load event detection of a first stage is carried out by improving an edge detection method and an adjacent load event segmentation step;
step three, detecting a long transient load event: taking signals of all sections except the step-type load event in the total load power data as detection objects, and performing long transient state type load event detection in the second stage by using a method of combining the moving average with the sliding t test;
step four, post-processing of load events: and collecting the step type load events and the long transient type load events measured in the first stage and the second stage, screening each detected load event, and identifying the load events with the starting and stopping point power difference smaller than the adaptive threshold as pseudo load events and eliminating the pseudo load events.
2. The method of claim 1, wherein the step two step detection of the first stage load event is performed by detecting and determining a step-type load event, and comprises:
step 2-1) extracting load events from the total load power time sequence by using an improved edge detection method:
firstly, a power change trend change point t meeting the formula (5) is searched in a load total power time series tp ，
t tp ＝{t|sgn(ΔP(t))≠sgn(ΔP(t+1))} (5)
In formula (5), sgn (-) is a sign function with a value range of { -1,0,1 };
during the scanning process of the total power sequence of the load, once the ith power change trend change point
If the requirement of formula (6) is satisfied, the product is determined
Is the start time of a load event;
in the following scanning process, if the power variation trend is at the ith' power variation trend changing point
To is that
If there is no point satisfying the formula (6) in the time range, then
Will be identified as the end time of this load event; and fromAt this moment, the search for the starting moment of the next new load event is developed; wherein, t th A length threshold value of a steady-state section outside a section where a load event is located in a time sequence of the total load power;
step 2-2) adjacent load event segmentation:
the adjacent load events refer to different load events with very short time intervals, and are divided into two conditions of same direction and reverse direction according to the power change direction relation of the two adjacent load events;
for segmentation of reverse adjacent load events, the steps are as follows:
step 1: for duration greater than preset threshold value delta t th Performs piece-wise Linear Representation (PLR); for the ith load event E (i) The sequence of load event powers is
Thereby obtaining a corresponding sequence of segmented points
Step 2: taking the power extreme point in the segment point sequence as a boundary point to the load event power sequence
And performing segmentation, wherein the judgment conditions of the power extreme point are as follows:
and 3, step 3: if the boundary point exists, determining the starting and stopping time of a new load event in each segmented power sequence, otherwise jumping to the step 5; setting the power change between the start and stop moments of the current power section as delta P seg In this section, the absolute value of the power variation is found to be greater than α · | Δ P seg Taking two time points with the shortest time interval as candidate new negativesThe starting and stopping time of the load event; wherein, alpha is a given proportionality coefficient;
and 4, step 4: merging the end-to-end candidate load events into a load event for subsequent processing;
and 5, step 5: repeating the steps 1 to 4 until all load events needing to be segmented are segmented;
for the segmentation of the equidirectional adjacent load events, the steps are as follows:
for load event E (i) Power sequence
The corresponding difference sequence is obtained by calculating the difference sequence,
calculating the power change between adjacent target elements, and if the formula (9) is satisfied, then carrying out the power sequence of the load event
Dividing the two target elements, and respectively using the divided power sequences as the power sequences of two different newly-added load events;
in the formula (9), β is a defined scaling factor threshold value, and a smaller value thereof indicates a more severe division condition,
respectively in adjacent target elementsThe former one and the latter one,
step 2-3) load event judgment, namely judging step type load events in all the load events obtained in the step 2-1) and the step 2-2) and taking the step type load events as a first-stage load event detection result, wherein the method comprises the following steps:
if a load event E (i) If the formula (10) is satisfied, the load event can be determined to be a step-type load event;
in the formula (10), N th Setting a sample point number threshold; Δ P (t) is the sequence
Of (1), t' outlier Is a sequence of
Where the time scale of outliers greater than the mean of the sequence, gamma is a defined scaling factor threshold,
for power changes between the start and stop times of the load event.
3. The two-stage adaptive non-intrusive loading event detection method as defined in claim 1, wherein the specific steps of step three are as follows:
and performing second-stage load event detection on signals of all sections except the step-type load event in the load power data: scanning the residual load total power sequence point by point, if the current time point satisfies the requirement of formula (11) and the previous time point is in a steady-state section, then judging to enter a load event transient section from the current time point and continuing to scan,
in the formula (11), at each time point t,
and
respectively is the average value of the power in the front scanning window and the rear scanning window of the point;
respectively the standard deviation of the power sequence under the corresponding window; the window length of the front and rear scanning windows satisfies W before ＝W after = w; at a given level of significance
In the case of a situation in which,
for the threshold value of t-test quantity obtained by table lookup,
has a degree of freedom of 2w-2, wherein
Taking the value as 0.05;
if equation (12) holds at time t and then t th If the expression (11) is not established any more in the duration range, the time t is judged to be the ending time of the transient section, namely, a new steady-state section is entered from the time;
4. the two-stage adaptive non-intrusive loading event detection method as defined in claim 1, wherein the detailed steps of step four are as follows:
aggregating the step-type load events detected in step two and the long transient-type load events detected in step three, and for each detected load event E (i) Screening according to the formula (13), and identifying the load event with the start-stop point power difference smaller than the adaptive threshold as a pseudo load event and removing the pseudo load event;
in equation (13), the steady-state segment length threshold t th Set to 5, scaling factor thresholds α, β, γ are set to 0.75,0.2,0.75, N, respectively th Set to 5.
5. A system for implementing a method for two-stage adaptive detection of a non-intrusive load event according to any of claims 1 to 4, comprising an adaptive threshold setting module, a step-type load event detection module, a long-transient-type load event detection module and a load event post-processing module;
the self-adaptive threshold setting module is used for calculating a self-adaptive power change threshold at each moment based on the original load total power data and judging whether a load event occurs or not; power change caused by a load event is eliminated from fluctuation of a total load signal through outlier analysis, so that a threshold value is dynamically adaptive to the fluctuation degree of a steady-state section of an electrical signal, and the detection capability of the load event with different amplitudes is improved;
the step type load event detection module is used for detecting the load event in the first stage by improving an edge detection method and adjacent load event segmentation, judging the step type load event in the step type load event, extracting and marking a corresponding transient process section from a load total power time sequence, and directly sending the transient process section into a subsequent load event post-processing step;
the long transient load event detection module detects the load event at the second stage by using signals of all sections except the step load event in the total load power data as detection objects and by using a method combining the sliding average and the sliding t test to detect the non-step load event; the non-step load event is called a long transient load event;
and the load event post-processing module screens all load events including step type and long transient state load events and eliminates pseudo load events of which the power change is smaller than a set self-adaptive threshold value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110984291.2A CN113687164B (en) | 2021-08-25 | 2021-08-25 | Non-intrusive load event two-stage self-adaptive detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110984291.2A CN113687164B (en) | 2021-08-25 | 2021-08-25 | Non-intrusive load event two-stage self-adaptive detection method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113687164A CN113687164A (en) | 2021-11-23 |
| CN113687164B true CN113687164B (en) | 2022-11-08 |
Family
ID=78582816
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110984291.2A Active CN113687164B (en) | 2021-08-25 | 2021-08-25 | Non-intrusive load event two-stage self-adaptive detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113687164B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115333102B (en) * | 2022-10-13 | 2022-12-23 | 天津大学 | Power load power mode multi-time scale self-adaptive detection method |
| CN115907567B (en) * | 2023-02-21 | 2023-05-09 | 浙江大学 | Load event detection method and system based on robust random segmentation forest algorithm |
| CN116861318B (en) * | 2023-09-05 | 2023-11-21 | 国网浙江省电力有限公司余姚市供电公司 | User electricity load classification method, device, equipment and storage medium |
Family Cites Families (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105678734B (en) * | 2014-11-21 | 2018-09-25 | 中国科学院沈阳自动化研究所 | A kind of heterologous test image scaling method of image matching system |
| CN104483575B (en) * | 2014-12-22 | 2017-05-03 | 天津求实智源科技有限公司 | Self-adaptive load event detection method for noninvasive power monitoring |
| CN106786534B (en) * | 2016-12-28 | 2019-07-26 | 天津求实智源科技有限公司 | A kind of non-intrusive electrical load transient process discrimination method and system |
| CN109902356A (en) * | 2019-01-30 | 2019-06-18 | 北京大学 | Electric power signal non-intrusion type loads decomposition method |
| US20220181877A1 (en) * | 2019-02-14 | 2022-06-09 | Massachusetts Institute Of Technology | Voltage waveform transient identification and autonomous load coordination |
| CN110866840A (en) * | 2019-11-12 | 2020-03-06 | 南昌工程学院 | Database modeling method for power load characteristic quantity training based on knowledge graph |
| CN112039059B (en) * | 2020-08-20 | 2022-02-18 | 国网湖南省电力有限公司 | Long transient load event detection method based on power step continuity judgment |
| CN112396007B (en) * | 2020-11-24 | 2021-08-24 | 国网江苏省电力有限公司营销服务中心 | Non-invasive three-threshold detection method and system for load sudden change event of residents |
-
2021
- 2021-08-25 CN CN202110984291.2A patent/CN113687164B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN113687164A (en) | 2021-11-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113687164B (en) | Non-intrusive load event two-stage self-adaptive detection method | |
| US9817045B2 (en) | Methods and system for nonintrusive load monitoring | |
| CN110504679B (en) | KM matching algorithm-based non-invasive load identification method | |
| CN113159077B (en) | Three-phase inverter fault identification method based on hybrid convolutional neural network | |
| CN112396007B (en) | Non-invasive three-threshold detection method and system for load sudden change event of residents | |
| CN111027408A (en) | Load identification method based on support vector machine and V-I curve characteristics | |
| Luan et al. | An adaptive two-stage load event detection method for nonintrusive load monitoring | |
| CN106802379A (en) | A kind of load switch detection of adaptive threshold and recognition methods and system | |
| CN109324285A (en) | A kind of load switch event detecting method and system | |
| CN117131110B (en) | Method and system for monitoring dielectric loss of capacitive equipment based on correlation analysis | |
| CN109085423A (en) | A kind of load switch event detecting method and system | |
| CN111413647B (en) | CLLLC resonant converter open-circuit fault real-time detection method and system | |
| CN111191908B (en) | Load event detection method, system and storage medium | |
| CN115333102B (en) | Power load power mode multi-time scale self-adaptive detection method | |
| CN115112989B (en) | Non-invasive load monitoring method based on low-frequency data | |
| CN108594156B (en) | Improved current transformer saturation characteristic identification method | |
| CN108594035B (en) | Load detection method and system | |
| CN112881793B (en) | Non-invasive load event detection method combined with time threshold | |
| CN112701683A (en) | Non-invasive load detection and decomposition method, system and storage medium | |
| Li et al. | A fusion framework using integrated neural network model for non-intrusive load monitoring | |
| CN111398810A (en) | Slip ring electric spark detection and diagnosis system and detection and diagnosis method | |
| CN117491730A (en) | Non-invasive near-simultaneous continuous event detection and electrical appliance classification method | |
| Sahrane et al. | A Clustering Event Detection Approach for Non-Intrusive Load Monitoring | |
| TWI769762B (en) | System and method for analyzing waveform applied to servo motor system | |
| CN111461081A (en) | Current signal segmentation algorithm |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |