CN112952827A

CN112952827A - Non-invasive full-load identification technology for accurately identifying charging of electric bicycle

Info

Publication number: CN112952827A
Application number: CN202110376510.9A
Authority: CN
Inventors: 任亚星
Original assignee: Individual
Current assignee: Individual
Priority date: 2021-04-08
Filing date: 2021-04-08
Publication date: 2021-06-11

Abstract

A non-invasive full load identification technique for accurately identifying the charging of an electric bicycle. By adopting a unique characteristic strengthening algorithm and a sliding window method, whether the electric bicycle is charged or not is accurately identified when multiple loads are simultaneously used, and all load types in use are identified. Adopt non-invasive installation, the equipment is not registered one's residence, is not inserted electric bicycle charging wire, is not inserted the load power supply line. The invention is mainly used for automatically monitoring the charging and instant alarming of the electric bicycles in families, enterprises and factories, and the type recognition, power detection and state diagnosis of electric appliances in use, and is widely applied to various fields of intelligent homes, intelligent buildings, intelligent control, energy-saving application and the like.

Description

Non-invasive full-load identification technology for accurately identifying charging of electric bicycle

Technical Field

The invention relates to a full-load type identification technology capable of identifying the charging of an electric vehicle battery, which is mainly used for automatically monitoring the charging of electric bicycles in families, enterprises and factories and giving an alarm in real time, and identifying the types, detecting the power and diagnosing the states of electric appliances in use, and is widely applied to various fields of intelligent home, intelligent buildings, intelligent control, energy-saving application and the like.

Background

The electric bicycle is charged indoors, in corridors, corridors and flying lines, and the fire and explosion caused by the charging cause great harm to the lives and properties of people. The problem of illegal charging of the electric vehicle has attracted high attention of supervision departments, becomes a topic of high social attention, and is urgently needed to be avoided.

Because electric bicycle self is small and exquisite, battery detachable characteristics for there is very big degree of difficulty in supervising electric bicycle's charging.

At present, no technology and equipment for automatically identifying the charging of the electric bicycle exist in China.

Meanwhile, the technology and products capable of identifying the load type, especially the identification under the condition of simultaneous working of multiple loads, are very few in market and low in accuracy.

The current load identification technology generally selects the transient characteristic and the steady-state characteristic of the load separately for matching identification. The adopted analysis technology comprises machine learning based on neural network algorithm, cluster analysis technology and the like. From the aspect of object orientation, the load operation is a continuous process from a transient state to a steady state, and the artificial segmentation characteristic interval does not accord with the self characteristics of the load, so that the error of load identification is increased.

The prior art cannot solve the contradiction between the measurement quantity of the electrical parameters, the accuracy of the algorithm and the performance of hardware equipment. When the quantity of the collected electrical parameters is small, the identification result is very inaccurate. When the number of the collected electrical parameters is large, the algorithm overhead is huge, and the capability of performing edge computing equipment is exceeded. More importantly, the data are used indiscriminately, the algorithm has poor convergence and is easy to fall into local optimization, and invalid data increases result errors, which all cause load identification errors.

Disclosure of Invention

Aiming at the current market blank, the invention provides a full load type identification technology for identifying the charging of an electric vehicle battery.

The invention provides a unique comprehensive identification algorithm with optimized fusion of a characteristic enhancement algorithm and a sliding window method, and whether an electric bicycle, including a lithium battery and a lead-acid battery, is charged by an electric power user can be accurately identified by calculating the acquired current or power data of an electric power inlet. The invention can accurately identify the types and the quantity of the loads, and solves the problems of few load identification products and low accuracy of the existing products.

The electric bicycle is also a power load when being charged, and when the electric bicycle charging identification is described by the following algorithm, the description and the algorithm of the load include the electric bicycle charging, which is not separately described and explained together with the power load in use.

The invention provides a characteristic enhancement algorithm, and adopts a three-order construction function method to enhance the distinguishable characteristic points of the load and inhibit the interference of the non-distinguishable characteristic points in the data on the result. Through a feature enhancement algorithm, the distinguishing feature points of the load directly determine the final load identification result. The algorithm selects current or power data during load operation.

The third-order constructor calculation process of the feature enhancement algorithm comprises the following steps:

1. the original data sequence a (n) = { a1, a 2., aN },

calculate b (N) = { b1, b 2.., bN }, b (N) = a (N) ^3, N =1, 2.., N.

2. Setting an observation constant k, k = 100, a slope width constant d, d =10, a sequence length N, N = 1000

3. Calculating the slopes m (1) of points a (1) and a (10), the slopes m (2) of points a (2) and a (11), the slopes m (3) of points a (3) and a (12), and.

4. The standard deviation σ of m (n) is calculated.

5. And | m (n) -m (n-d + 1) | <3 σ and | m (n) -m (n + d-1) | <3 σ, a (n) are marked as characteristic points, and other points are interference points and are removed.

6. The data points are moved back 100 and the feature points are repeatedly calculated until all feature points are calculated.

The invention provides a sliding window method for searching an optimal identification data interval. The specific method comprises the following steps:

and acquiring current or power data in real time at a sampling rate of 1k/s, and circularly storing for 6 minutes. Identification calculations were performed every 6 minutes. The identification windows defining each identification window length 30s, 30s comprise both transient and steady state characteristics of the load.

Identification step 1: the identification is started, a first identification window comprises a data sequence which is 30s backward from a starting point, feature enhancement algorithm calculation is firstly carried out, a feature value is enhanced, the feature value sequence is firstly subjected to paste progress calculation with the feature value containing the electric bicycle in the load matrix library, then the paste progress calculation is sequentially carried out with the load group in the load matrix library, and the obtained highest paste degree is stored as a matching result K1 of the window;

and (2) identification: and (4) moving the recognition window backwards for 1s and repeating the calculation of the recognition step 1, wherein the obtained highest pasting speed is stored as a matching result K2 of the window. The highest pasting speed is calculated after the recognition window is moved each time and is stored as K3 and K4..

And (3) identification: and in the matching results of all the windows, if the matching result of the load group containing the electric bicycle is the highest, identifying that the electric bicycle is charged, otherwise, identifying the load group corresponding to the Kn value of the highest matching result as all the currently used load types.

In the sliding window method, the characteristic value of the load group in the load matrix library is calculated according to the following process:

1. fixing the type and the total amount of the load, and acquiring a current or power data sequence;

2. carrying out data cleaning and interpolation;

3. obtaining a characteristic value of the sample at the time through a characteristic enhancement algorithm;

4. performing mass training, and storing characteristic values of the samples;

5. calculating the eigenvalue with the optimal similarity with all the eigenvalue sequences by using a machine learning algorithm, and taking the eigenvalue as the eigenvalue of the load matrix library;

6. setting new load types and total quantities, and calculating the characteristic value of the full combined load group.

Drawings

Fig. 1 is a schematic diagram of the technical application of the present invention, and mainly illustrates the work flow.

Fig. 2 is a diagram of an application system composition of the present invention, which mainly illustrates a structural composition of the system and a connection flow between various components such as hardware, software, and human-computer interaction in a typical application scenario.

Detailed Description

The most typical application of the invention is to monitor whether a power consumer violates charging an electric bicycle.

The technology of the invention is simply implanted into the DSP and the MCU, no additional data analysis technology is needed, and the electric bicycle can be immediately charged by finding that the electric power user violates the regulations through the current or power data monitored in real time, and the electric power user can be sent to a supervision department to immediately stop the charging, and the electric power user is allowed to be informed to immediately stop the charging at the same time.

The equipment is installed in a non-invasive mode, namely the equipment is not connected to the home, is not connected to a charging wire of an electric vehicle and a power supply wire of an electric appliance, and is not connected to a wire to be tested to the home. The installation position is an electric power entrance and is generally placed in an electric meter box.

Application scenarios reference is made to the technical application diagram of fig. 1 and the application system composition diagram of fig. 2.

Another typical use is appliance identification. The DSP and the MCU are installed at the power inlet, all the electric appliance types newly used and used by a power user can be found immediately, including the power consumed by each electric appliance, and the electric appliance types and the power are sent in the modes of APP or computer clients and the like for power-saving management of the power user.

In a third typical application, as a system component, the equipment can be communicated with a property department and an electric power department, and when illegal electric bicycles are detected to be charged, the equipment is immediately reported to related departments.

The fourth typical application is as a system component, is integrated into an intelligent household product, realizes intelligent control of household appliances, and gives a power saving prompt by combining the characteristics of the appliances.

The algorithm can be independently used, and the method for processing the current and the power is also suitable for processing the power parameters such as voltage, harmonic waves and the like, so that the characteristic points are strengthened, the optimal identification interval is searched, and the data with the highest matching degree is determined.

The above description is only a preferred embodiment of the present invention, and various modifications and changes may be made by those skilled in the art. These examples are intended to illustrate the invention only and do not limit the scope of the invention in any way.

Claims

1. A non-invasive full load recognition technology for accurately recognizing the charge of electric bicycle features that a unique characteristic strengthening algorithm and a sliding window method are used to perform the discernable characteristic point strengthening on the current or power data of non-invasive collected power inlet and search the optimal data recognition region for recognizing the charge and load type of electric bicycle.

2. A non-invasive full load identification technique for accurately identifying electric bicycle charging as claimed in claim 1 is characterized by providing a method of calculating characteristic values of load groups in the load matrix library as data references for charging electric bicycles and appliance load identification based on AI or other methods.