CN113033607B - Non-invasive load monitoring event extraction method and storage medium - Google Patents
Non-invasive load monitoring event extraction method and storage medium Download PDFInfo
- Publication number
- CN113033607B CN113033607B CN202110182121.2A CN202110182121A CN113033607B CN 113033607 B CN113033607 B CN 113033607B CN 202110182121 A CN202110182121 A CN 202110182121A CN 113033607 B CN113033607 B CN 113033607B
- Authority
- CN
- China
- Prior art keywords
- load data
- power load
- power
- segment
- monitoring event
- 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
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
Abstract
The invention provides a non-invasive load monitoring event extraction method and a storage medium, which are used for processing time-series power load data fragments based on a non-time-series method, avoiding depending on a time-series model with high calculation requirements, inputting visual statistical characteristics into a decision tree model instead, and having stronger expansibility.
Description
Technical Field
The invention relates to the technical field of electricity utilization, in particular to a non-invasive load monitoring event extraction method and a storage medium.
Background
Under the intelligent electricity utilization technology system, non-invasive load monitoring (NILM) is an important ring, and aims to deeply analyze the electricity utilization load types of users only by means of monitoring devices installed at buses and acquire electricity utilization information with different fineness. Event-based non-invasive load monitoring can extract trusted pieces of electricity usage events for sample data slicing, and for detected event pieces, a key technology in non-invasive load monitoring is how to accurately identify. At present, two main ideas are commonly adopted in the field: the database comparison method is used for comparing unknown event fragments to confirm the categories of the unknown event fragments by accumulating a large number of marked samples, and has the core difficulty of calculation efficiency and universality; pattern recognition attempts to learn the feature classes, and has the defects of complex training process, poor interpretability and low current accuracy.
Accordingly, a novel non-invasive load monitoring event extraction method is presented herein to preserve the positive effects of existing non-invasive load monitoring event extraction methods while eliminating the negative effects of existing non-invasive load monitoring event extraction methods.
Disclosure of Invention
The invention aims to avoid dependence on a time sequence model with high calculation requirement and have high calculation speed by processing time sequence power load data fragments based on a non-time sequence method.
In order to achieve the above objective, the present invention provides a non-invasive load monitoring event extraction method, comprising the following steps: establishing a power load database, wherein the power load database comprises a plurality of power load data fragments, and the power load data fragments are power load data in a historical time period; collecting the plurality of power load data fragments for statistical analysis to obtain input characteristics; constructing a sample set to be trained, wherein the sample set comprises the plurality of power load data fragments and the input features; building a decision tree model for training the sample set; inputting the sample set into the decision tree model for training to obtain a judgment model; and acquiring real-time power load data fragments and inputting the real-time power load data fragments into the judgment model to obtain the corresponding electric equipment type.
Further, in the step of acquiring the plurality of power load data segments and performing statistical analysis to obtain input characteristics, the method specifically includes the following steps: calculating a difference between a maximum value of the power load data and a minimum value of the power load data in the power load data segment; calculating a ratio of a maximum value of the electrical load data to a minimum value of the electrical load data; a segment length of the power load data segment is calculated, the segment length of the power load data segment being a number of sample data points in the power load data segment.
Further, in the step of acquiring the plurality of pieces of power load data and performing statistical analysis to obtain the input characteristics, after the step of calculating the number of pieces of power load data, the method further includes the steps of: calculating an absolute value of a difference between adjacent ones of the power load data segments; comparing the absolute value with a preset value, and dividing the power load data segment into a platform segment and a jump segment; wherein the absolute values in the platform segments are all smaller than the preset value, and the absolute values in the jump segments are all smaller than the preset value; and sequentially carrying out statistical analysis on the station fragments and the hopping fragments.
Further, before the step of creating the electrical load database, the method further comprises the steps of: acquiring the instantaneous voltage and the instantaneous current of a user bus; multiplying the instantaneous voltage and the instantaneous current to obtain active power in a historical time period; multiplying the instantaneous voltage by the offset 1/4 period instantaneous current to obtain reactive power in a historical time period; wherein the power load data segment includes active power and reactive power.
Further, in the step of acquiring the instantaneous voltage and the instantaneous current of the subscriber bus, the instantaneous voltage and the instantaneous current are acquired by using a current-voltage acquisition unit, and the acquisition frequency is 5-15 Hz.
Further, before the step of creating the electrical load database, after the step of multiplying the instantaneous voltage by the offset 1/4 period instantaneous current to obtain the reactive power in the historical time, the method further includes the following steps: filling in missing values in the active power during the history period and the reactive power during the history period.
Further, in the step of filling the missing value in the active power in the history period and the reactive power in the history period, the filling method is to fill the data of the next time corresponding to the missing value to the time corresponding to the missing value.
Further, the decision tree model comprises a gradient-lifting decision tree model; wherein the number of trees is 300, and the maximum depth of each tree is 3 layers.
Further, in the step of inputting the sample set into the decision tree model to train to obtain a judgment model, the method specifically includes the following steps: randomly dividing the sample set into training samples and test samples; inputting the training sample into the decision tree model for training to obtain a first judgment model; inputting the test sample into the first judgment model to perform verification operation, and optimizing the first judgment model according to a verification result to obtain the judgment model.
The invention also provides a storage medium, on which a computer program is stored, which when executed by a processor can implement the non-invasive load monitoring event extraction method.
The beneficial effects of the invention are as follows: the invention provides a non-invasive load monitoring event extraction method and a storage medium, which are used for processing time-series power load data fragments based on a non-time-series method, avoiding dependence on a time-series model with high calculation requirements, and inputting visual statistical characteristics into a decision tree model instead, and have strong expansibility.
Drawings
The invention is further described below with reference to the drawings and examples.
FIG. 1 is a flow chart of a non-invasive load monitoring event extraction method according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a load data segment of an electric device according to an embodiment of the present invention;
FIG. 3 is a flowchart illustrating a non-invasive load monitoring event extraction method according to an embodiment of the present invention in step S6);
fig. 4 is a flowchart of step S9) of a non-invasive load monitoring event extraction method according to an embodiment of the present invention.
Detailed Description
The following description of the embodiments refers to the accompanying drawings, which illustrate specific embodiments in which the invention may be practiced. The names of the elements, such as first, second, etc., mentioned in the present invention are merely distinguishing different elements, and may be better expressed. In the drawings, like elements are referred to by like reference numerals and adjacent or similar elements are referred to by similar reference numerals.
Embodiments of the present invention will be described in detail herein with reference to the accompanying drawings. This invention may take many different forms and should not be construed as limited to the particular embodiments set forth herein. These embodiments are provided so that this disclosure will be thorough and complete, and will fully enable others skilled in the art to understand the various embodiments of the invention and with various modifications as are suited to the particular use contemplated.
As shown in FIG. 1, one embodiment of the present invention provides a non-invasive load monitoring event extraction method, which is a non-invasive load monitoring method. The invention provides a non-sequential method-based power load data segment processing sequential, which avoids dependence on a sequential model with high calculation requirement, and replaces intuitive statistical features to be input into a decision tree model, so that the model has stronger expansibility, and the current model can accurately classify events of different electric appliances and has high calculation speed. The non-invasive load monitoring event extraction method includes the following steps S1-S10.
S1) acquiring the instantaneous voltage and the instantaneous current of a user bus; in this embodiment, the current-voltage collector is used to collect the instantaneous voltage and the instantaneous current, and the collection frequency is 5-15 Hz, which may be 10Hz, 8Hz, 9Hz, 11Hz, and 12Hz in this embodiment. The unit of acquisition is a segment, i.e., a preset historical period of time, each segment comprising 3-50 sample data points.
S2) multiplying the instantaneous voltage and the instantaneous current to obtain the active power in the history period. The active power sequence refers to the alternating current energy actually sent or consumed by the electric equipment in unit time.
S3) multiplying the instantaneous voltage by the offset 1/4 period instantaneous current to obtain reactive power in a historical time period; wherein the power load data segment includes active power and reactive power. Reactive power is the reversible energy exchange between the reactive element in the circuit and the circuit, and is actually the exchange power (not actually doing work) of electric energy and a magnetic field.
S4) filling the missing value of the active power in the history time period and the reactive power in the history time period; in this embodiment, the filling method is to fill the data of the next time corresponding to a missing value to the time corresponding to the missing value.
S5) establishing a power load database, wherein the power load database comprises a plurality of power load data fragments, and the power load data fragments are power load data in a historical time period. As shown in fig. 2, the electric load data segment of the microwave oven of the electric equipment is shown, the solid line represents the active power, and the dotted line represents the reactive power.
S6) collecting the plurality of power load data segments, and carrying out statistical analysis to obtain input features. Specifically, as shown in fig. 3, step S6 specifically includes the following steps:s601) calculating a difference between a maximum value of the power load data and a minimum value of the power load data in the power load data segment. S602) calculating a ratio of a maximum value of the power load data to a minimum value of the power load data. S603) calculating a segment length of the power load data segment, the segment length of the power load data segment being a number of sample data points in the power load data segment. S604) calculates an absolute value of a difference of adjacent power load data in the power load data segment. S605) comparing the absolute value with a preset value, and dividing the power load data segment into a platform segment and a jump segment; the absolute values in the platform segments are smaller than the preset value, and the absolute values in the jump segments are smaller than the preset value. S606) sequentially carrying out statistical analysis on the platform fragments and the hopping fragments, and specifically, respectively calculating the average value/variance, the average value/variance of the differential sequence, the number of outliers and the like of the platform fragments and the hopping fragments. Because certain fluctuation is presented on a power curve when the electric appliance is started or shut down, the core of the patent is to capture the relation between the two types of power fluctuation characteristics (namely the platform segment and the jump segment) and different types of electric appliances, namely the relation between the active and reactive waveforms and the starting of a certain device is obtained. Generating a number of non-timed numerical features (in the form of a sequence of active power segments) for a sequence of events (active power and reactive power), respectively
For example, the method of the invention is equally applicable to reactive power segment sequences, the generation of the judgment model depends on the characteristics generated from both sequences) the following sequence numbers do not represent the calculation order, but are calculation methods of several parallel numerical characteristics), 1) calculate the amplitude range of the power load data segment ∈ ->
2) Maximum/minimum value in power load data +.>
3) Power load data segment Length m i The method comprises the steps of carrying out a first treatment on the surface of the 4) Dividing the power load data segment into a platform segment and a jump segment, and respectively calculating the average value/variance of the platform segment and the jump segment, the average value/variance of the differential sequence, the number of outliers and the like, wherein the sub-features are respectively calculated for the platform segment and the jump segment. These features are the necessary input features for the decision tree algorithm of this patent and have a significant positive effect on improving classification accuracy. The code of the dividing method of the platform segment and the jump segment is as follows:
s7) constructing a sample set to be trained, the sample set comprising the plurality of electrical load data segments and the input features.
S8) building a decision tree model for training the sample set. The decision tree model comprises a gradient lifting decision tree model; wherein the number of trees is 300, and the maximum depth of each tree is 3 layers. The gradient-lifting decision tree model (GBDT) is a variant of the generic decision tree model. The gradient lifting decision tree model continuously reduces the classification loss by constructing a plurality of iterative decision trees, the training of the model is serial, namely each decision tree is sequentially established, the loss (error) of the predicted value of the tree is calculated by comparing with the correct classification label, and the prediction target of the next tree is the error of the last tree under each sample. The input of the decision tree model is a numeric identification feature set (namely a power load data segment and an input feature) of each power utilization event time sequence sample, and the output is a corresponding power utilization class label and is coded in a one-hot mode. Specifically, the gradient-lifting decision tree model pseudocode is as follows:
where M represents building M decision trees, K represents all possible classification labels of the sample, and J represents that each decision tree is composed of J leaf nodes.
S9) inputting the sample set into the decision tree model for training to obtain a judgment model. As shown in fig. 4, step S9 specifically includes the following steps: s901) randomly dividing the sample set into a training sample and a test sample, wherein only the training sample is visible to the model in the training process, and the test sample is used for verifying the generalization performance of the model, and the ratio of the training sample to the test sample is 4:1. S902) inputting the training sample into the decision tree model for training to obtain a first judgment model; s903) inputting the test sample into the first judgment model for verification operation. S904) optimizing the first judgment model according to the verification result to obtain the judgment model.
S10) acquiring real-time power load data fragments and inputting the power load data fragments into the judgment model to obtain the corresponding electric equipment type.
An embodiment of the invention provides a non-invasive load monitoring event extraction method, which is used for processing time-series power load data fragments based on a non-time-series method, avoiding dependence on a time-series model with high calculation requirements, inputting visual statistical characteristics into a decision tree model instead, and having stronger expansibility.
An embodiment of the present invention provides a storage medium having a computer program stored thereon, which when executed by a processor, implements the non-invasive load monitoring event extraction method. The non-invasive load monitoring event extraction method processes the time-series power load data fragments based on a non-time-series method, avoids depending on a time-series model with high calculation requirements, and inputs the time-series model into a decision tree model instead of visual statistical characteristics, so that the non-invasive load monitoring event extraction method has strong expansibility, and the current model can accurately monitor events of different electric appliances and has high calculation speed.
It should be noted that numerous variations and modifications are possible in light of the fully described invention, and are not limited to the specific examples of implementation described above. The above-described embodiments are merely illustrative of the present invention and are not intended to be limiting. In general, the scope of the present invention should include those variations or alternatives and modifications apparent to those skilled in the art.
Claims (8)
1. A method of non-intrusive load monitoring event extraction, comprising the steps of:
establishing a power load database, wherein the power load database comprises a plurality of power load data fragments, and the power load data fragments are power load data in a historical time period;
collecting the plurality of power load data fragments for statistical analysis to obtain input characteristics;
constructing a sample set to be trained, wherein the sample set comprises the plurality of power load data fragments and the input features;
building a decision tree model for training the sample set;
inputting the sample set into the decision tree model for training to obtain a judgment model;
acquiring real-time power load data fragments and inputting the real-time power load data fragments into the judgment model to obtain corresponding electric equipment types;
the step of acquiring the plurality of power load data fragments for statistical analysis to obtain input characteristics specifically comprises the following steps:
calculating a difference between a maximum value of the power load data and a minimum value of the power load data in the power load data segment;
calculating a ratio of a maximum value of the electrical load data to a minimum value of the electrical load data;
calculating a segment length of a power load data segment, the segment length of the power load data segment being a number of sample data points in the power load data segment;
calculating an absolute value of a difference between adjacent ones of the power load data segments;
comparing the absolute value with a preset value, and dividing the power load data segment into a platform segment and a jump segment; wherein the absolute values in the platform segments are all smaller than the preset value, and the absolute values in the jump segments are all smaller than the preset value;
and sequentially carrying out statistical analysis on the station fragments and the hopping fragments.
2. The method for non-intrusive load monitoring event extraction of claim 1, wherein,
before the step of establishing the electrical load database, the method further comprises the following steps:
acquiring the instantaneous voltage and the instantaneous current of a user bus;
multiplying the instantaneous voltage and the instantaneous current to obtain active power in a historical time period;
multiplying the instantaneous voltage by the offset 1/4 period instantaneous current to obtain reactive power in a historical time period;
wherein the power load data segment includes active power and reactive power.
3. The method for non-intrusive load monitoring event extraction of claim 2, wherein,
in the step of acquiring the instantaneous voltage and the instantaneous current of the subscriber bus,
the instantaneous voltage and the instantaneous current are acquired by using a current-voltage collector, and the acquisition frequency is 5-15 Hz.
4. The method for non-intrusive load monitoring event extraction of claim 2, wherein,
before the step of building the power load database, after the step of multiplying the instantaneous voltage by the offset 1/4 period instantaneous current to obtain the reactive power in the historical time, the method further comprises the following steps:
filling in missing values in the active power during the history period and the reactive power during the history period.
5. The method for non-intrusive load monitoring event extraction of claim 4, wherein,
in the step of filling the missing value in the active power in the history period and the reactive power in the history period, the filling method is to fill the data of the next moment corresponding to the missing value to the moment corresponding to the missing value.
6. The method for non-intrusive load monitoring event extraction of claim 1, wherein,
the decision tree model comprises a gradient lifting decision tree model; wherein the number of trees is 300, and the maximum depth of each tree is 3 layers.
7. The method for non-intrusive load monitoring event extraction of claim 1, wherein,
the step of inputting the sample set into the decision tree model to train to obtain a judgment model specifically comprises the following steps:
randomly dividing the sample set into training samples and test samples;
inputting the training sample into the decision tree model for training to obtain a first judgment model;
inputting the test sample into the first judgment model for verification operation;
and optimizing the first judgment model according to the verification result to obtain the judgment model.
8. A storage medium having stored thereon a computer program which, when executed by a processor, performs the method of any of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110182121.2A CN113033607B (en) | 2021-02-10 | 2021-02-10 | Non-invasive load monitoring event extraction method and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110182121.2A CN113033607B (en) | 2021-02-10 | 2021-02-10 | Non-invasive load monitoring event extraction method and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113033607A CN113033607A (en) | 2021-06-25 |
| CN113033607B true CN113033607B (en) | 2023-07-14 |
Family
ID=76461381
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110182121.2A Active CN113033607B (en) | 2021-02-10 | 2021-02-10 | Non-invasive load monitoring event extraction method and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113033607B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109492667A (en) * | 2018-10-08 | 2019-03-19 | 国网天津市电力公司电力科学研究院 | A kind of feature selecting discrimination method for non-intrusive electrical load monitoring |
| CN110504679A (en) * | 2019-07-25 | 2019-11-26 | 深圳供电局有限公司 | A kind of non-intrusion type load discrimination method based on KM matching algorithm |
| CN111126780A (en) * | 2019-10-31 | 2020-05-08 | 内蒙古电力(集团)有限责任公司包头供电局 | Non-invasive load monitoring method and storage medium |
-
2021
- 2021-02-10 CN CN202110182121.2A patent/CN113033607B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109492667A (en) * | 2018-10-08 | 2019-03-19 | 国网天津市电力公司电力科学研究院 | A kind of feature selecting discrimination method for non-intrusive electrical load monitoring |
| CN110504679A (en) * | 2019-07-25 | 2019-11-26 | 深圳供电局有限公司 | A kind of non-intrusion type load discrimination method based on KM matching algorithm |
| CN111126780A (en) * | 2019-10-31 | 2020-05-08 | 内蒙古电力(集团)有限责任公司包头供电局 | Non-invasive load monitoring method and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113033607A (en) | 2021-06-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111368904B (en) | Electrical equipment identification method based on electric power fingerprint | |
| CN106411257A (en) | Photovoltaic power station state diagnosis method and device | |
| CN109840691B (en) | Non-invasive polynomial electric quantity estimation method based on deep neural network | |
| CN103869192B (en) | Intelligent grid line loss detection method and system | |
| CN106443244A (en) | Method and system for identifying type of electric devices | |
| CN112396098A (en) | Non-embedded apartment electrical appliance load identification and analysis method, system and application | |
| CN117559443A (en) | Ordered power utilization control method for large industrial user cluster under peak load | |
| CN113033607B (en) | Non-invasive load monitoring event extraction method and storage medium | |
| CN116861316B (en) | Electrical appliance monitoring method and device | |
| CN110968703B (en) | Method and system for constructing abnormal metering point knowledge base based on LSTM end-to-end extraction algorithm | |
| CN113659564A (en) | Low-voltage distribution network topology identification method and system based on voltage fluctuation feature clustering | |
| CN112595918A (en) | Low-voltage meter reading fault detection method and device | |
| Paul et al. | An intelligent system for domestic appliance identification using deep dense 1-D convolutional neural network | |
| CN115168437B (en) | Method and system for realizing portrait of electricity user based on data analysis | |
| CN111651448A (en) | Low-voltage topology identification method based on noise reduction differential evolution | |
| CN111080089A (en) | Method and device for determining critical factors of line loss rate based on random matrix theory | |
| CN115712871A (en) | Power electronic system fault diagnosis method combining resampling and integrated learning | |
| CN115828091A (en) | Non-invasive load identification method and system based on end-cloud cooperation | |
| CN115392663A (en) | Data acquisition and processing method based on big data | |
| CN110245617B (en) | Artificial intelligence analysis method based on transient recording waveform | |
| CN111124816B (en) | HDP algorithm-based server log analysis method and system | |
| CN114358160A (en) | Data anomaly detection method in power system | |
| CN109615558B (en) | User electricity habit analysis method based on electricity parameter big data statistics | |
| Ma et al. | State duration based event detection for domestic power disaggregation | |
| Martin et al. | Non-Intrusive Load Monitoring for Feeder-Level EV Charging Detection: Sliding Window-based Approaches to Offline and Online Detection |
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 |