CN108535566A - Electric appliance sorter - Google Patents

Abstract

A device for classifying electrical appliances, including an information collection module, an information processing module, and a communication module, while using electrical appliance startup current characteristics including startup impulse current, startup average current, startup current impulse, and electrical appliance load current spectrum characteristics as identification Classification features, rich feature information; using a combination classifier including BP neural network classifier and Bayesian classifier for judgment and classification, taking into account the characteristics of BP neural network classifier and Bayesian classifier for comprehensive judgment, high judgment accuracy ; The method for acquiring the characteristics of the starting current and the frequency spectrum of the load current provided is simple and reliable. The device can be used in collective public places such as student dormitories and large-scale bazaars that require management of electrical appliances, and can also be used in other occasions that require electrical equipment management that requires identification, classification and statistics of electrical appliances.

Description

Electrical classification device

The patent application for this invention is a divisional application, the original application number is 201610213353.9, the application date is April 8, 2016, and the invention name is a method for judging the type of electrical appliances.

technical field

The invention relates to an equipment judging and classifying device, in particular to an electrical appliance classifying device.

Background technique

At present, the mainstream methods for judging the properties of electrical appliances include the electrical load judging method based on the load power comprehensive coefficient algorithm, the electrical load judging method based on electromagnetic induction, the electrical load judging method based on neural network algorithm, and the electrical load judging method based on periodic discrete transform algorithm Judgment method, etc. Various methods can realize the judgment of the nature of electrical load to a certain extent, but due to the single characteristic and single judgment method, there are generally problems of insufficient generalization ability and incomplete and accurate judgment.

Contents of the invention

The object of the present invention is to provide an electrical appliance classification device capable of realizing efficient judgment, aiming at the defects of the prior art. The device includes an information collection module, an information processing module, and a communication module; a combined classifier is used to identify and classify electrical appliances, and the input characteristics of the combined classifier include the starting current characteristics of the electrical appliances and the load current spectrum characteristics of the electrical appliances; the starting current Features include starting impulse current, starting average current, and starting current impulse.

The starting current characteristics are obtained by the following methods:

Step 1. Wait until it is judged that the electrical appliance starts to start, then turn to step 2;

Step 2. Continuously sample the load current of the electrical appliance, calculate and save the effective value of the load current in units of power frequency cycles, and turn to step 3 until it is judged that the electrical load has entered a stable state;

Step 3. Take the average value of the effective value of the load current within the last N power frequency cycles as the steady-state current of the electrical appliance load; take the time between the start of the electrical appliance and the start of the last N power frequency cycles as the start-up process time ;Calculate the ratio between the average value of the electrical appliance load current effective value and the electrical appliance load steady-state current within L power frequency cycles after the appliance starts, and use the ratio as the starting impulse current of the appliance; calculate the starting process time of the appliance The ratio between the average value of the effective value of the electrical appliance load current and the steady-state current of the electrical appliance load, the ratio is taken as the average startup current of the appliance; the product of the average startup current of the appliance and the startup process time is calculated, and the product is As the starting current impulse of the electrical appliance; the value range of N is 50-500, and the value range of L is 1-5.

The method for judging that the electric appliance starts is as follows: before the electric appliance is started, start to continuously sample the load current of the electric appliance and judge the magnitude of the load current; when the effective value of the load current is greater than ε, it is determined that the electric appliance starts; value.

The method for judging that the electrical load enters a stable state is: calculating the average value of the effective value of the load current of the last N power frequency cycles; When the average value of the load current effective value of N power frequency cycles is compared, and the fluctuation range is smaller than the set relative error range E, it is determined that the electrical load enters a stable state; the value range of E is 2%-20%.

The load current spectrum feature is obtained by the following method:

Step 1. Obtain the steady-state current signal of the electrical load and convert it into a corresponding steady-state current digital signal;

Step 2, performing Fourier transform on the steady-state current digital signal to obtain the load current spectrum characteristic;

Step 3: Use the relative amplitude of the nth harmonic signal in the load current spectrum characteristic as the load current spectrum characteristic, where n=1, 2, . . . The relative amplitude of the harmonic signal is the ratio of the amplitude of the harmonic signal to the effective value of the steady-state current of the electrical appliance load.

The combination classifier includes a BP neural network classifier and a Bayesian classifier, wherein the BP neural network classifier is a main classifier, and the Bayesian classifier is an auxiliary classifier. The method for the combined classifier to identify and classify electrical appliances is as follows: when the main classifier successfully realizes the identification and classification of electrical appliances, the result of the electrical identification and classification of the main classifier is the judgment result of the combined classifier; when the main classifier fails to realize the identification and classification of electrical appliances, and The judgment result of the main classifier is 2 or more types of electrical appliances, and among the 2 or more types of electrical identification and classification results output by the main classifier, the electrical appliance type with the highest probability in the output of the auxiliary classifier is used as the electrical appliance of the combined classifier. Recognition and classification results; when the main classifier fails to realize electrical appliance identification and classification, and the judgment result of the main classifier fails to give the type of electrical appliance judged, the electrical appliance type with the highest probability in the output of the auxiliary classifier is used as the electrical appliance identification of the combined classifier classification results.

The input characteristics of the combined classifier also include the steady state current of the electrical appliance load.

The information collection module is used to collect the load current information of the electrical appliance and send it to the information processing module; the information processing module performs electrical identification and classification according to the input information; the communication module is used to send the electrical identification and classification results of the information processing module to PC.

The information acquisition module includes a current sensor, a preamplifier, a filter, and an A/D converter; the core of the information processing module is DSP, or ARM, or a single-chip microcomputer, or FPGA.

The A/D converter may adopt an A/D converter included in the core of the information processing module.

All or part of the functions of the information collection module, information processing module and communication module are integrated on one SoC.

The communication module also receives relevant work instructions from the host computer; the communication mode between the communication module and the host computer includes a wireless communication mode and a wired communication mode; the wireless communication mode includes ZigBee, Bluetooth, WiFi, and 433MHz digital transmission mode ; The wired communication methods include 485 bus, CAN bus, Internet, and power carrier.

The beneficial effect of the present invention is: adopt the combination classifier that comprises BP neural network classifier and Bayesian classifier to carry out judgment classification, take into account the characteristics of BP neural network classifier and Bayesian classifier to carry out comprehensive judgment, generalization ability and The judgment accuracy is high; at the same time, the starting current characteristics of electrical appliances, the load current spectrum characteristics of electrical appliances, and the effective value of steady-state current of electrical appliances are used as the input features of the combined classifier, and the feature information is rich; the provided information includes starting impulse current and starting average current. The acquisition method of starting current characteristics including starting current impulse and the acquisition method of load current spectrum characteristics are simple and reliable.

Description of drawings

Fig. 1 is the structural representation of the embodiment of electric appliance classification device;

Fig. 2 is the current waveform of the starting process of the incandescent desk lamp;

Figure 3 is the current waveform of the start-up process of resistive loads such as resistance furnaces;

Figure 4 is the current waveform in the starting process of a single-phase motor load;

Fig. 5 is the starting process current waveform of computer and switching power supply load;

Fig. 6 is a flow chart of the electrical appliance identification and classification method of the electrical appliance classification device.

Detailed ways

The present invention will be further described below in conjunction with accompanying drawing.

FIG. 1 is a schematic structural diagram of an embodiment of an electrical appliance classification device, including an information collection module 101 , an information processing module 102 , and a communication module 103 .

The information collection module 102 is used to collect the load current of the electrical appliance and convert the load current into a current digital signal, and the current digital signal is sent to the information processing module 102 . The information acquisition module includes current sensors, preamplifiers, filters, A/D converters and other components, which respectively complete the sensing, amplification, filtering and analog-to-digital conversion functions of the load current signal. When the load current range is large, you can choose a preamplifier with program-controlled function, or add an independent program-controlled amplifier before the A/D converter, and implement segmented control amplification for a large range of load current, so that the input The voltage signal range to the A/D converter is kept within a reasonable range to ensure conversion accuracy. Filters are used to filter out high frequency components to avoid spectral aliasing.

The information processing module 102 uses a combined classifier including a BP neural network classifier and a Bayesian classifier to realize electrical appliance identification and classification according to the input current digital signal. The input features of the combined classifier include the starting current features of electrical appliances and the load current spectrum features of electrical appliances. The core of the information processing module 102 is DSP, ARM, single-chip microcomputer, or FPGA. When the core of the information processing module includes an A/D converter and the A/D converter meets the requirements, the A/D converter in the information collection module 101 can adopt the A/D converter included in the core of the information processing module 102. converter.

The communication module 103 is used to realize the communication with the upper computer, and send the judgment result to the upper computer. The communication mode between the communication module 102 and the upper computer includes wireless communication mode and wired communication mode. The wireless communication mode that can be used includes ZigBee, Bluetooth, WiFi, 433MHz data transmission, etc. The wired communication mode that can be used includes 485 bus, CAN Bus, Internet, power carrier, etc. The communication module 103 can also receive relevant work instructions from the host computer to complete specified work tasks. The upper computer can be the server of the management department, various workstations, or various mobile terminals.

All or part of the functions of the information collection module 101, the information processing module 102, and the communication module 103 can be integrated on one SoC, which reduces the size of the device and facilitates installation.

Different electrical equipment has different starting current characteristics. As shown in Figure 2, it is the current waveform of the starting process of the incandescent lamp. An incandescent lamp is an electric light source that heats the filament to an incandescent state and emits visible light by thermal radiation. The filament of an incandescent lamp is usually made of high temperature resistant metal tungsten, but the resistance of metal tungsten varies greatly with temperature. Let R _t represent the resistance of the tungsten wire at t°C, and R ₀ represent the resistance of the tungsten wire at 0°C, then The two have the following relationship

R _t =R ₀ (1+0.0045t)

For example, if the temperature of the filament (tungsten filament) of an incandescent lamp is 2000°C during normal operation, the resistance of the filament of a "220V 100W" incandescent lamp at 2000°C is

Its resistance at 0°C when no power is applied is

Its resistance at 20°C when no power is applied is

R ₂₀ ＝R ₀ (1+0.0045t)＝52.8Ω

That is, the current of the incandescent lamp at the moment of starting and energizing exceeds 9 times its rated current, and the maximum starting current occurs at the starting moment. As the temperature of the tungsten filament of the incandescent lamp increases, the load current of the incandescent lamp decreases exponentially and then enters a steady state.

Assuming that the effective value of the steady-state current of the electrical appliance load is I _W , and it is defined that the effective value of the electrical appliance load current enters a set relative error range of the steady-state current effective value of the electrical appliance load and is stable within this relative error range, then the electrical appliance load into a steady state. The relative error range can be set at 10%, or at a value between 2%, 5%, 15%, 20%, etc., between 2% and 20%. In Figure 2, the set relative error range is 10%. When the load current of the incandescent lamp decreases to the 10% error range of its I _W according to the exponential law, as shown in the time T _S in Figure 2, the start-up process ends. The starting process time of the incandescent lamp is T _S . I _W is a valid value.

The starting impulse current I _G , the starting average current _ID and the starting current impulse Q _I are selected as the starting current characteristics of the electrical appliance; the starting impulse current I _G and the starting average current _ID are all standard unit values. The specific definition is: the starting impulse current I _G is the ratio of the average value of the electrical load current to the steady-state current I _W of the electrical load within T ₂ time after the start of the electrical appliance; the starting average current I _D is within the starting time T _S of the electrical appliance The ratio of the average value of the load current of the electrical appliance to the steady-state current I _W of the electrical appliance load; the starting current impulse Q _I is the product of the average starting current _ID and the starting process time T _S , and the dimension is ms. Electrical load current and electrical load steady-state current are effective values. The value range of T ₂ is 20-100 ms, or 1-5 power frequency cycles; for example, the value of T ₂ is 40 ms, that is, 2 power frequency cycles. The starting impulse current _IG reflects the magnitude of the current impulse within a short time after the electrical load is started. In the starting process of some electrical appliances, when the actual starting process time T _S of some electrical appliances is less than the set T ₂ , make the starting process time T _S of the electrical appliances equal to T ₂ . The starting average current _ID reflects the overall size of the current during the starting process of the electrical load. The starting current impulse Q _I reflects the overall strength of the electrical load starting.

In Fig. 2, the starting impulse current I _G of the incandescent lamp is the average value of the current of the incandescent lamp between T ₀ (the starting time of the incandescent lamp, the current is I ₀ ) and T ₂ (the setting time, the current is I ₂ ) and The ratio of the steady-state current I _W of an incandescent lamp. The starting average current _ID is the ratio of the average current of the incandescent lamp between T ₀ (starting time of the incandescent lamp) and T _S (end time of the starting process of the incandescent lamp) and the steady-state current I _W of the incandescent lamp. The starting current impulse Q _I is the product of the average starting current _ID of the incandescent lamp and the starting process time T _S .

Figure 3 shows the current waveform of the start-up process of resistive loads such as resistance furnaces. Resistive loads such as resistance furnaces usually use electrothermal alloy wires such as nickel-chromium, iron-chromium-aluminum, etc., and their common characteristics are small resistance temperature correction coefficient and stable resistance value. Taking the nickel-chromium heating wire with the brand name Cr20Ni80 as an example, its resistance correction coefficient at 1000°C is 1.014, that is, at 1000°C compared to 20°C, the resistance of the nickel-chromium heating wire with the brand name Cr20Ni80 only increases by 1.4%. Resistive loads such as resistance furnaces enter a stable state when they are powered on, and the actual start-up time of resistive loads such as resistance furnaces T _S = 0. Therefore, the actual start-up time of resistive loads such as resistance furnaces T _S = T ₂ ; For example, when T ₂ is set to 40ms, the start-up process time T _S is also 40ms. Since the current I ₀ at the moment T ₀ of the resistive load and the current I ₂ at the moment T ₂ are equal to the steady-state current I _W of the resistive load, the starting impulse current I _G of the resistive load = 1, and the starting average current I _D = 1.

Figure 4 shows the current waveform of the single-phase motor load during startup. Single-phase motor loads have both inductive load characteristics and counter electromotive force load characteristics. At the starting moment, due to the effect of the inductance, the starting current I ₀ at the starting moment is 0; then the current rises rapidly, and reaches the current peak value I _M before the back electromotive force of the motor is established; after that, the motor speed increases, and the motor load current gradually decreases, until it reaches a steady state. In Figure 4, the starting impulse current I _G of a single-phase motor load is a single-phase current between T ₀ (when the single-phase motor load starts, the current is I ₀ ) and T ₂ (the setting time, the current is I ₂ ). The ratio of the average current of the phase motor load to the steady-state current I _W. The starting average current I _D is the ratio of the average current of the single-phase motor load to the steady-state current I _W between T ₀ (starting moment of the single-phase motor load) and T _S (the end time of the single-phase motor load start-up process) . The starting current impulse Q _I is the product of the average starting current _ID of a single-phase motor load and the starting process time T _S.

Figure 5 shows the current waveforms of the computer and switching power supply loads during startup. Computer and switching power supply loads will generate a large surge current at the moment of startup due to the impact on capacitor charging, and its peak value can reach several times to ten times the steady-state current effective value I _W , and the time is 1 to 2 frequency cycle. Due to the short start-up time of computers and switching power supply loads, the start-up process time T _S may be less than the set T ₂ ; when the start-up process time T _S is less than the set T ₂ , set T _S equal to T ₂ . In Fig. 5, the start-up impulse current I _G of computer and switching power supply loads is between T ₀ (when the computer and switching power supply loads start, the current is I ₀ ) to T ₂ (the setting time, the current is I ₂ ) The ratio of the average current of computer and switching power supply loads to the steady-state current I _W. Starting average current I _D is the average current and steady-state current I of computer and switching power loads between T ₀ (starting time of computer and switching power loads) and T _S (end time of starting process of computer and switching power loads) Ratio of _W. The starting current impulse Q _I is the product of the average starting current _ID of computer and switching power supply loads and the starting process time T _S.

The method to obtain the starting current characteristics of the electrical appliance is:

Before the electrical appliance starts, when the load current value is 0 (not powered on) or very small (in standby state), the information processing module 102 starts to continuously sample the load current; when the sampled effective value of the load current starts to be greater than 0 or is When it starts to be greater than the standby current of the electrical appliance, it is judged that the electrical appliance has been started, and this moment is recorded as T ₀ . A small non-negative threshold ε is used to distinguish the load current value before and after the electrical appliance is started. When the value of ε is particularly small, for example, when the value of ε is 1mA, the device does not consider the standby situation, that is, the standby is also the starting state of the electrical appliance ; When the value of ε is small but greater than the standby current of the electrical appliance, for example, when the value of ε is 20mA, the device will consider the standby state of the electrical appliance as an inactive state, but at the same time, some electrical appliances with particularly small power will cause Leaked judgment.

The information processing module 102 continuously samples the load current, and calculates and saves the effective value of the load current in units of power frequency cycles; when the electrical appliance has been started and the continuous sampling reaches N power frequency cycles, the latest N values are continuously calculated while sampling. The average value _IV of the load current effective value of the power frequency cycle; the information processing module 102 compares the load current effective value of each power frequency cycle within the latest N power frequency cycles with the load current effective value of the N power frequency cycles Compared with the average value, when the error (or fluctuation) range is less than the set relative error range E, it is judged that the electrical load has entered a stable state, and the start time of the latest N power frequency cycles is the end time of the start-up process, record this time is T ₁ (as shown in Fig. 2-Fig. 5).

The average value of the effective value of the load current within the latest N power frequency cycles is taken as the steady-state current _{I W} of _the electrical appliance load; Start process time T _S . Calculate the ratio of the average value of the load current to the steady-state current I _W between T ₀ and the set T ₂ (that is, within 1-5 power frequency cycles after the electrical appliance starts to start), and use this ratio as the starting impulse of the electrical appliance current I _G . Calculate the ratio of the average load current between T ₀ and T _S to the steady-state current I _W , and use this ratio as the starting average current _ID of the electrical appliance. Calculate the product of the starting average current _ID of the electrical appliance and the starting process time T _S , and use this product as the starting current impulse Q _I of the electrical appliance.

Since the effective value of the steady-state current I _W of the electrical appliance load is not known in advance, the average value of the effective value of the load current when the fluctuation range is less than the set relative error range E within N power frequency cycles, that is, a period of time T _P As the effective value of the steady-state current of the electrical appliance load I _W . Due to the fast start-up process of ordinary electrical loads, the value range of T _P is 1-10s, the typical value is 2s, and the corresponding power frequency cycle number N ranges from 50-500, the typical value of N is 100. The value range of the relative error range E is 2%-20%, and the typical value of E is 10%.

The input features of the combination classifier also include the load current spectrum features of electrical appliances. The load current spectrum feature of the electrical appliance is obtained by the information processing module 102 controlling the information collection module 101 through the following steps:

Step 1. After the electrical load enters a steady state, the steady-state current signal of the electrical load is obtained and converted into a corresponding steady-state current digital signal.

Step 2, performing Fourier transform on the steady-state current digital signal to obtain the load current spectrum characteristic. In order to ensure the smooth progress of Fourier transform, in the process of obtaining the steady-state current signal of the electrical load and converting it into the corresponding steady-state current digital signal, the accuracy and speed of the A/D converter need to meet the requirements of Fourier transform , the sampling frequency can be set to 10kHz, or other numerical values; the information processing module 102 performs FFT operation on the collected steady-state current digital signal, and calculates its frequency spectrum.

Step 3, use the relative amplitude of the nth harmonic signal in the load current spectral characteristic as the load current spectral characteristic, wherein, n=1, 2, ..., M; when forming the input feature vector of the combined classifier, the nth harmonic The relative amplitude of the wave signal is arranged in the order of 1, 2, ..., M in the input feature vector. Since the load current spectral characteristics are mainly composed of odd harmonics, except for a few electrical equipment, the even harmonic components are almost 0, therefore, the odd harmonic signal whose harmonic order is n in the load current spectral characteristics can also be The relative amplitude is used as the characteristic of the load current spectrum, where, n=1, 3, . . . , M. The relative amplitude of the harmonic signal is the ratio of the amplitude of the harmonic signal to the effective value I _W of the steady-state current of the electrical appliance load. The 1st harmonic when n=1 is the power frequency fundamental wave. The M represents the highest order of the harmonic, and in general, M is greater than or equal to 5.

In the combined classifiers, the BP neural network classifier is the main classifier, and the Bayesian classifier is the auxiliary classifier. The input features of the combined classifier include the aforementioned start-up current features and load current spectrum features, and the input features of the combined classifier are both the input features of the BP neural network classifier and the input features of the Bayesian classifier.

As shown in Figure 6, it is a flow chart of the electrical appliance identification and classification method of the electrical appliance classification device, and the specific steps are:

Step A, wait for the electrical appliance to start;

Step B, collecting and saving the starting current data of the electric appliance until the end of the starting process of the electric appliance;

Step C, analyzing the collected starting current data of the electric appliance to obtain the starting current characteristics of the electric appliance;

Step D, collect and save the data when the electric appliance works in a steady state;

Step E, analyzing the collected data of the electric appliance during steady state operation, and obtaining the load current spectrum characteristics of the electric appliance;

Step F, using the starting current feature and the load current spectrum feature as the input features of the combined classifier; the combined classifier performs electrical identification and classification;

Step G, outputting the electrical appliance identification and classification results.

The method for the combined classifier to identify and classify electrical appliances is as follows: when the main classifier successfully implements electrical identification and classification, that is, when the judgment result output by the main classifier is the only electrical appliance type, that is, when the only electrical appliance type in the judgment result is yes, The electrical appliance type judged by the main classifier is used as the electrical appliance identification and classification result of the combined classifier; When 2 or more types of electrical appliances are yes, among the 2 or more types of electrical identification and classification results output by the main classifier, the electrical appliance type with the highest probability in the output of the auxiliary classifier is used as the electrical identification and classification result of the combined classifier; When the main classifier fails to realize electrical identification and classification, and the judgment result of the main classifier fails to give the judged electrical type, that is, if there is no electrical type in the judgment result, the electrical type with the highest probability in the output of the auxiliary classifier is taken as Combined classifier's appliance recognition classification results.

A simple embodiment 1 is taken as an example to illustrate the method of combining classifiers to identify and classify electrical appliances. There is a combined classifier whose input feature is x={I _G , I _D , Q _I , A ₁ , A ₂ , A ₃ , A ₄ , A ₅ }, where I _G is the starting impulse current; I _D is the starting average current; Q _I is the starting current impulse; A ₁ , A ₂ , A ₃ , A ₄ , and A ₅ are the relative amplitudes of the 1-5 harmonic signals in the load current spectrum characteristics. The output of the combined classifier is {B ₁ , B ₂ , B ₃ , B ₄ }, B ₁ , B ₂ , B ₃ , and B ₄ respectively represent the output of the combined classifier’s judgment results for incandescent lamps, resistance furnaces, hair dryers, and computers , the values of the judgment results B ₁ , B ₂ , B ₃ , and B ₄ are all binary classification marks. The input features of the main classifier are also x={I _G , I _D , Q _I , A ₁ , A ₂ , A ₃ , A ₄ , A ₅ }, and its output is {F ₁ , F ₂ , F ₃ , F ₄ }, F ₁ , F ₂ , F ₃ , and F ₄ respectively represent the output of the judgment results of the main classifier for incandescent lamps, resistance furnaces, hair dryers, and computers, and the values of the judgment results F ₁ , F ₂ , F ₃ , and F ₄ are also Both are binary classification labels. The input feature of the auxiliary classifier is also x={I _G , _ID , Q _I , A ₁ , A ₂ , A ₃ , A ₄ , A ₅ }, and its output is {P(y ₁ |x), P( y ₂ |x), P(y ₃ |x), P(y ₄ |x)}, P(y ₁ |x), P(y ₂ |x), P(y ₃ |x), P(y ₄ |x) is the posterior probability output by the auxiliary classifier, and the mutual size between P(y ₁ |x), P(y ₂ |x), P(y ₃ |x), and P(y ₄ |x) Indicates that the current input features of the auxiliary classifier represent the possibility that the judged electrical appliances belong to incandescent lamps, resistance stoves, hair dryers, and computers.

In Example 1, the classification marks of B ₁ , B ₂ , B ₃ , and B ₄ and the classification marks of F ₁ , F ₂ , F ₃ , and F ₄ are all 1 and 0. When the classification mark is 1, the corresponding electrical appliance type matches the current input feature, which is a confirmed judgment result, or the corresponding electrical appliance identification and classification result is yes; when the classification mark is 0, the corresponding electrical appliance type does not match the input feature, not The judgment result that can be confirmed, or the corresponding electric appliance identification and classification result is no.

In Embodiment 1, if the classification mark of the judgment result of a certain main classifier is F ₁ F ₂ F ₃ F ₄ =0100, then it is considered that the main classifier successfully realizes the identification and classification of electrical appliances, so the judgment of the auxiliary classifier is not considered As a result, directly set B ₁ B ₂ B ₃ B ₄ =0100, that is, the judgment result of the combination classifier is: the judged electric appliance is a resistance furnace.

In Embodiment 1, if the classification mark of the judgment result of a certain main classifier is F ₁ F ₂ F ₃ F ₄ =1010, then it is considered that the main classifier failed to realize the electrical identification classification, and the judgment result of the main classifier is 2 or more than 2 types of electrical appliances; and assuming that the judgment result of the auxiliary classifier satisfies P(y ₁ |x)<P(y ₃ |x), then let B ₁ B ₂ B ₃ B ₄ =0010, that is The judging result of the combined classifier is: the judged electrical appliance is a hair dryer.

In Embodiment 1, if the classification mark of the judgment result of a certain main classifier is F ₁ F ₂ F ₃ F ₄ =0000, then it is considered that the main classifier has failed to realize electrical appliance identification and classification, and the judgment result of the main classifier is The type of electrical appliance that can not be judged is given; and the judgment result of the auxiliary classifier at this time is set to satisfy P(y ₁ |x)>P(y ₂ |x) and P(y ₁ |x)>P(y ₃ |x ) and P(y ₁ |x)>P(y ₄ |x), then set B ₁ B ₂ B ₃ B ₄ =1000, that is, the judgment result of the combination classifier is: the electric appliance judged is an incandescent lamp.

The classification marks of the judgment results of the combined classifier and the main classifier can also adopt other schemes, for example, use the classification marks 1, -1, or 0, 1, or -1, 1, and other schemes to represent the corresponding electrical appliances The judgment result is yes or no. The classification labeling scheme of the combined classifier and the main classifier can be the same or different.

The input features of the combined classifier may also include the effective value I _W of the steady-state current of the electrical appliance load. For example, there are two different electrical appliances, electric soldering iron and resistance furnace, which need to be judged. Both electric soldering iron and resistance furnace are pure resistance loads, and both have the common characteristics of small resistance temperature correction coefficient and stable resistance value. Therefore, they cannot be distinguished simply by relying on the aforementioned starting current characteristics and load current spectrum characteristics. After adding the effective value of the steady-state current I _W of the electric appliance load in the input feature, the power of the electric soldering iron is small, and the effective value I _W of the steady-state current of the electric appliance load is small; the power of the resistance furnace is large, and the effective value I _W of the steady-state current of the electric appliance load is large, and the characteristics are different. Combined classifiers can make and complete judgments.

The auxiliary classifier is a Bayesian classifier. You can choose among three Bayesian classifiers such as NBC classifier (Naive Bayesian classifier), TAN classifier (Tree-extended Naive Bayesian classifier), BAN classifier (Enhanced Bayesian classifier) A kind of as an auxiliary classifier.

Example 2 selects the NBC classifier as the auxiliary classifier. Naive Bayes classification is defined as follows:

(1) Let x={a ₁ , a ₂ ,..., a _m } be an item to be classified, and each a is a characteristic attribute of x;

⑵There is a category set C={y ₁ , y ₂ ,...,y _n };

⑶ Calculate P(y ₁ |x), P(y ₂ |x),..., P(y _n |x);

(4) If P(y _k |x)=max{P(y ₁ |x), P(y ₂ |x), ..., P(y _n |x)}, then x∈y _k .

The specific method of calculating each conditional probability in step (3) is:

① Find a set of items to be classified with known classification as the training sample set;

②Statistically obtain the conditional probability estimates of each feature attribute under each category;

P(a ₁ |y ₁ ), P(a ₂ |y ₁ ),..., P(a _m |y ₁ );

P(a ₁ |y ₂ ), P(a ₂ |y ₂ ),..., P(a _m |y ₂ );

...;

P(a ₁ |y _n ), P(a ₂ |y _n ),..., P(a _m |y _n ).

③According to Bayes' theorem, there are:

Because the denominator is constant for all categories, we only need to maximize the numerator; and because each feature attribute in Naive Bayes is conditionally independent, so there are:

In Example 2, the input features of the combined classifier are {I _G , _ID , Q _I , A ₁ , A ₃ , A ₅ , I _W }, where I _G is the starting impulse current; _ID is the starting average current; Q _I is the starting current impulse; A ₁ , A ₃ , A ₅ are the relative amplitudes of the 1st, 3rd, and 5th odd-order harmonic signals in the spectrum characteristics of the load current; I _W is the effective value of the steady-state current of the electrical appliance load, The unit is ampere. The electrical appliances that require judgment are incandescent lamps, resistance furnaces, electric fans, computers, and electric soldering irons. Let the elements in the feature attribute combination x={a ₁ , a ₂ , a ₃ , a ₄ , a ₅ , a ₆ , a ₇ } of the naive Bayesian classifier and the elements in the input feature set of the combined classifier press Sequence {I _G , I _D , Q _I , A ₁ , A ₃ , A ₅ , I _W } one-to-one correspondence; output category set C of Naive Bayesian classifier={y ₁ , y ₂ , y ₃ , y ₄ , y ₅ } are in one-to-one correspondence with electrical appliances such as incandescent lamps, resistance furnaces, electric fans, computers, and electric soldering irons.

The process of training an NBC classifier includes:

1. Divide the feature attributes into segments and perform discretization. In Embodiment 2, the feature attribute discretization method adopted is:

a ₁ : {a ₁ <2.8, 2.8≤a ₁ ≤5.2, a ₁ >5.2};

a ₂ : {a ₂ <1.31, 1.31≤a ₂ ≤2.6, a ₂ >2.6};

a ₃ : {a ₃ <128, 128≤a ₃ ≤550, a ₃ >550};

a ₄ : {a ₄ <0.7, 0.7≤a ₄ ≤0.9, a ₄ >0.9};

a ₅ : {a ₅ <0.02, 0.02≤a ₅ ≤0.05, a ₅ >0.05};

a ₆ : {a ₆ <0.01, 0.01≤a ₆ ≤0.035, a ₆ >0.035};

a ₇ : {a ₇ <0.45, a ₇ ≥0.45}.

2. Collect multiple sets of samples for each type of electrical appliances as training samples, and calculate the proportion of each type of electrical appliances in all electrical appliances, that is, calculate P(y ₁ ), P(y ₂ ), P (y ₃ ), P(y ₄ ), P(y ₅ ). When the same number of samples is collected for each type of electrical appliance, for example, each type of electrical appliance collects more than 100 groups of samples, among which 100 groups of samples are randomly selected for each type of electrical appliance as training samples, and the others are used as test samples. The total training samples are 500 group with

P(y ₁ )=P(y ₂ )=P(y ₃ )=P(y ₄ )=P(y ₅ )=0.2.

3. Calculate the frequency (proportion) of each feature attribute segment under each category condition of the training sample, and obtain the conditional probability estimate of each feature attribute under each category, that is, statistical calculation respectively

P(a ₁ <2.8|y ₁ ), P(2.8≤a ₁ ≤5.2|y ₁ ), P(a ₁ >5.2|y ₁ );

P(a ₁ <2.8|y ₂ ), P(2.8≤a ₁ ≤5.2|y ₂ ), P(a ₁ >5.2|y ₂ );

...;

P(a ₁ <2.8|y ₅ ), P(2.8≤a ₁ ≤5.2|y ₅ ), P(a ₁ >5.2|y ₅ );

P(a ₂ <1.31|y ₁ ), P(1.31≤a ₂ ≤2.6|y ₁ ), P(a ₂ >2.6|y ₁ );

P(a ₂ <1.31|y ₂ ), P(1.31≤a ₂ ≤2.6|y ₂ ), P(a ₂ >2.6|y ₂ );

...;

P) a ₂ <1.31|y ₅ ), P(1.31≤a ₂ ≤2.6|y ₅ ), P(a ₂ >2.6|y ₅ );

P(a ₃ <128|y ₁ ), P(128≤a ₃ ≤550|y ₁ ), P(a ₃ >550|y ₁ );

P(a ₃ <128|y ₂ ), P(128≤a ₃ ≤550|y ₂ ), P(a ₃ >550|y ₂ );

...;

P(a ₃ <128|y ₅ ), P(128≤a ₃ ≤550|y ₅ ), P(a ₃ >550|y ₅ );

P(a ₄ <0.7|y ₁ ), P(0.7≤a ₄ ≤0.9|y ₁ ), P(a ₄ >0.9|y ₁ );

P(α ₄ <0.7|y ₂ ), P(0.7≤a ₄ ≤0.9|y ₂ ), P(a ₄ >0.9|y ₂ );

...;

P(a ₄ <0.7|y ₅ ), P(0.7≤a ₄ ≤0.9|y ₅ ), P(a ₄ >0.9|y ₅ );

P(a ₅ <0.02|y ₁ ), P(0.02≤a ₅ ≤0.05|y ₁ ), P(a ₅ >0.05|y ₁ );

P(a ₅ <0.02|y ₂ ), P(0.02≤a ₅ ≤0.05|y ₂ ), P(a ₅ >0.05|y ₂ );

P(a ₅ <0.02|y ₅ ), P(0.02≤a ₅ ≤0.05|y ₅ ), P(a ₅ >0.05|y ₅ );

P(a ₆ <0.01|y ₁ ), P(0.01≤a ₆ ≤0.035|y ₁ ), P(a ₆ >0.035|y ₁ );

P(a ₆ <0.01|y ₂ ), P(0.01≤a ₆ ≤0.035|y ₂ ), P(a ₆ >0.035|y ₂ );

...;

P(a ₆ <0.01|y ₅ ), P(0.01≤a ₆ ≤0.035|y ₅ ), P(a ₆ >0.035|y ₅ );

P(a ₇ <0.45|y ₁ ), P(a ₇ ≥0.45|y ₁ );

P(a ₇ <0.45|y ₂ ), P(a ₇ ≥0.45|y ₂ );

...;

P(a ₇ <0.45|y ₅ ), P(a ₇ ≥0.45|y ₅ ).

After the above steps 1, 2 and 3, the training of the NBC classifier is completed. Among them, step 1 divides the feature attributes into segments manually determined, and when discretizing each input feature into segments, the number of segments is 2 segments or more. For example, in embodiment 2, feature a ₁ - a ₆ is divided into 3 sections, feature a ₇ is divided into 2 sections. How many segments each feature is divided into, and the selection of the segmentation threshold can be adjusted according to the results of the test sample after the training of the Bayesian classifier. Step 2 and Step 3 are completed by the information processing module 102 or computer calculation.

The method that adopts Bayesian classifier to classify among the present invention is:

1. Use the input features of the combined classifier as the input features of the Bayesian classifier. In Example 2, the input feature set {I _G , _ID , Q _I , A ₁ , A ₃ , A ₅ , I _W } of the combined classifier is used as the input feature x of the Bayesian classifier, and x = {a ₁ , a ₂ , a ₃ , a ₄ , a ₅ , a ₆ , a ₇ }.

2. According to the conditional probability estimation of each characteristic attribute under each category obtained through training, respectively determine the segmentation location of each input characteristic attribute and determine its probability P(a ₁ |y ₁ )～P(a _m |y _n ), where the set of electrical appliance categories is C={y ₁ , y ₂ ,...,y _n }. In Example 2, the electrical appliance category set C={y ₁ , y ₂ , y ₃ , y ₄ , y ₅ } corresponds to the representative electrical appliance categories including incandescent lamps, resistance furnaces, electric fans, computers, and electric soldering irons, and P(a ₁ |y ₁ )～P(a ₇ |y ₅ ) method is to use the conditional probability estimation of each feature attribute obtained in the process of training the NBC classifier.

3. According to the formula

Compute the posterior probability for each appliance class. Since the denominator P(x) is a constant for all electrical appliances, setting P(x)=1 to replace the actual P(x) value does not affect the mutual comparison between the posterior probabilities of each electrical appliance category. At this time, there is

In Example 2, there are

Use test samples to test the trained Bayesian classifier, and decide whether to adjust the discretization method for input features (that is, adjust the number of segments and thresholds) according to the test results, and retrain the Bayesian classifier.

The main classifier is a BP neural network classifier, and a 3-layer BP neural network classifier is selected as the main classifier. The number of elements in the BP neural network classifier input feature vector, that is, the number of input features, is used as the number of nodes in the input layer, for example, 8 in embodiment 1, or 7 in embodiment 2. The quantity of the electric appliance type that needs classification judgment is used as the output layer node number, for example, in embodiment 1, output layer node is 4, outputs and judges the result of incandescent lamp, resistance furnace, blower, computer respectively; In embodiment 2, output There are 5 layer nodes, which respectively output the results of judging incandescent lamps, resistance furnaces, electric fans, computers, and electric soldering irons. The number of nodes in the middle hidden layer is selected based on experience. For example, the number of nodes in the hidden layer in Embodiment 1 and Embodiment 2 can be selected within the range of 6-15. Collect multiple sets of samples for each type of electrical appliance, for example, 200 sets of samples are collected; some of them are randomly selected, for example, 150 sets of samples as training samples, and the rest are used as test samples to train and test the BP neural network classifier . Due to the coupling effect between multiple outputs, the multi-input and multi-output 3-layer BP neural network classifier may not be able to completely judge the samples during training or testing; even if it can completely judge the samples during training or testing, Restricted by the generalization ability, when judging a certain feature attribute of the new input, the main classifier may output the judgment result as the only electrical type, or the judgment result is 2 or more types of electrical appliances, or fail to Give the type of electrical appliances judged.

The main classifier can also be composed of multiple single-node output 3-layer BP neural network classifiers, and each single-node output 3-layer BP neural network classifier corresponds to judge a type of electrical appliance. For example, in embodiment 1, it can be used The 3-layer BP neural network classifier of 4 single-node outputs judges incandescent lamp, resistance furnace, hair dryer, computer respectively; Can adopt the 3-layer BP neural network classifier of 5 single-node outputs to judge respectively incandescent lamp, resistor in embodiment 2 Stove, electric fan, computer, electric soldering iron. When the main classifier selects multiple single-node output 3-layer BP neural network classifiers to form together, the number of input layer nodes of all single-node output 3-layer BP neural network classifiers is the number of elements in the main classifier input feature vector; The number of nodes in the middle hidden layer is based on experience. The number of nodes in the middle hidden layer of the 3-layer BP neural network classifier output by each single node can be the same or different, and can be selected according to their own needs. Like the neural network with non-single-node output, multiple sets of samples need to be collected for each type of electrical appliance, for example, 200 sets of samples are collected; some of them are randomly selected, for example, 150 sets of samples are used as training samples, and the rest are used as test samples , to train and test the BP neural network classifier output by each single node. When the main classifier is composed of multiple single-node output 3-layer BP neural network classifiers, each single-node output 3-layer BP neural network classifier only needs to complete the judgment of one electrical appliance type, and the training of each network is relatively Simple. Since the main classifier is composed of multiple 3-layer BP neural network classifiers output by single nodes at this time, and the 3-layer BP neural network classifiers output by each single node are independent of each other, therefore, when judging a certain characteristic attribute, The judgment result output by the main classifier may be the only electrical appliance type, or the judgment result may be two or more electrical appliance types, or the electrical appliance type for which classification judgment cannot be given.

The training method of the BP neural network classifier can adopt the gradient descent method, and can also adopt optimization methods such as particle swarm optimization and genetic algorithm. Sample collection adopts the aforementioned method of acquiring the starting current characteristics of the electrical appliance and the method of acquiring the load current spectrum characteristics of the electrical appliance.

Claims (9)

1. a kind of electric appliance sorter, which is characterized in that including information acquisition module, message processing module, communication module；Using Assembled classifier carries out electric appliance identification classification, and the input feature vector of the assembled classifier includes the starting current feature and electricity of electric appliance The load current spectrum signature of device；The starting current feature includes starting current rush, starting average current, starting current punching Amount.

2. electric appliance sorter as described in claim 1, which is characterized in that the starting current feature obtains by the following method ：

Step 1 waits for, until after judging that electric appliance starts startup, turns to step 2；

Step 2 carries out continuous sampling to the load current of electric appliance, as unit computational load current effective value and is protected using power frequency period It deposits, until after judging that electric appliance load enters stable state, turns to step 3；

Step 3, using the average value of the load current virtual value within nearest N number of power frequency period as electric appliance load steady-state current； Electric appliance is started into Startup time to the time between nearest N number of power frequency period initial time as the start-up course time；Calculate electricity Device starts after starting between the average value and electric appliance load steady-state current of the electric appliance load current effective value within L power frequency period Ratio, using the ratio as the startup current rush of electric appliance；Calculate the electric appliance load electricity within the start-up course time of electric appliance The ratio between the average value and electric appliance load steady-state current of virtual value is flowed, using the ratio as the startup average current of electric appliance； The startup average current for calculating electric appliance and the product between the start-up course time are rushed the product as the starting current of electric appliance Amount；The value range of the N is 50-500, and the value range of L is 1-5.

3. electric appliance sorter as claimed in claim 2, which is characterized in that it is described judge electric appliance start start method be： Before appliance starting, starts the load current continuous sampling to electric appliance and load current size is judged；When load current has When valid value is more than ε, judgement electric appliance starts to start；The ε is the numerical value more than 0.

4. electric appliance sorter as claimed in claim 3, which is characterized in that the judgement electric appliance load enters stable state Method is：Calculate the average value of the load current virtual value of N number of power frequency period recently；It is every within nearest N number of power frequency period The load current virtual value of a power frequency period is compared with the average value of the load current virtual value of N number of power frequency period, fluctuation When amplitude is respectively less than the relative error range E set, judgement electric appliance load enters stable state；The value range of the E is 2%-20%.

5. electric appliance sorter as described in claim 1, which is characterized in that the load current spectrum signature passes through with lower section Method obtains：

Step 1: obtaining the steady state current signals of electric appliance load, and it is converted into corresponding steady-state current digital signal；

Step 2: carrying out Fourier transform to steady-state current digital signal, load current spectral characteristic is obtained；

Step 3: using the nth harmonic signal relative magnitude in load current spectral characteristic as load current spectrum signature, In, n=1,2 ..., M；The M indicates harmonic wave highest number and M is more than or equal to 5.

6. electric appliance sorter as claimed in claim 5, which is characterized in that the harmonic signal relative magnitude is harmonic signal The ratio of amplitude and electric appliance load steady-state current virtual value.

7. the electric appliance sorter as described in any one of claim 1-6, which is characterized in that the assembled classifier includes BP Neural network classifier and Bayes classifier；In the assembled classifier, BP neural network grader is main grader, pattra leaves This grader is auxiliary grader.

8. electric appliance sorter as claimed in claim 7, which is characterized in that the assembled classifier carries out electric appliance identification classification Method be：When Main classification device successfully realizes electric appliance identification classification, the electric appliance identification classification results of Main classification device are combination point The judging result of class device；Classify when Main classification device fails realization electric appliance identification, and the judging result of Main classification device is 2 kinds or 2 Kind or more appliance type, in 2 kinds that Main classification device is exported or two or more electric appliance identification classification results, subsidiary classification device is defeated The electric appliance for going out the highest appliance type of middle probability as assembled classifier identifies classification results；When Main classification device fails to realize electric appliance Identification classification, and when failing to provide the appliance type of judgement in the judging result of Main classification device, will be general in the output of subsidiary classification device The highest appliance type of rate identifies classification results as the electric appliance of assembled classifier.

9. electric appliance sorter as described in claim 1, which is characterized in that described information acquisition module is for acquiring electric appliance Load current information is simultaneously sent to message processing module；Described information processing module carries out electric appliance identification point according to the information of input Class；The electric appliance that the communication module is used to send message processing module identifies classification results to host computer.