CN106940806A - A kind of method and system for the use electricity condition for recognizing electrical equipment - Google Patents

Abstract

The present invention relates to a kind of method and system for the use electricity condition for recognizing electrical equipment, methods described includes：Step 1, the voltage x current data based on electrical equipment calculate characteristic parameter, build electric appliance working condition characteristic data set；Step 2, electric appliance working condition hidden Markov model is built, carry out model training using obtained electricity consumption working condition characteristic data set is calculated in step 1 and obtain optimal model parameters；Step 3, with reference to the state transition equation and observational equation obtained based on optimal model parameters, utilize Evolutionary particle filter, go to calculate the posterior probability density function of the status switch occurred under this observation sequence by known observation sequence, so as to realize the identification to electrical equipment complexity electricity condition.The system includes data acquisition unit, feature extraction unit, hidden Markov model construction unit and electric appliance state recognition unit.

Description

A kind of method and system for the use electricity condition for recognizing electrical equipment

Technical field

The present invention relates to intelligent electric meter field, and more particularly, to a kind of method for the use electricity condition for recognizing electrical equipment And system.

Background technology

With the problems occurred in traditional ammeter, the development of intelligent electric meter is extremely urgent.Implement intelligent electric meter development Strategy can make the supply of electric power of user's acquisition high security, high reliability, high-quality property, high efficiency and reasonable price.Intelligence electricity Table is an important link in wired home, and intelligent electric meter can monitor the power information of different electrical equipment in family, can be with The accurate electricity consumption situation for understanding family is enabled users to, can so allow user more rationally, effectively can use in family Electrical equipment.For most electrical equipment, general only two states：On-state and off-state, such as electric light, television set.But some Electrical equipment has many different conditions, as electric fan is except there is an on off state, several states of also different gears, different conditions it Electricity consumption situation be different.If we can only recognize power information total in one family, but electrical equipment is not known just Which type of uses electricity condition in, this is accomplished by intelligent electric meter and can identified according to the power information of electrical equipment at electrical equipment this moment In use electricity condition.

Present invention mainly solves the problem of be：For the Intelligent Recognition of the electrical equipment progress electricity condition of some complex states.

The content of the invention

In order to solve the problem of complexity to electrical equipment of background technology presence is identified with electricity condition, the present invention provides one The method of kind：Analysis is modeled with electricity condition to electrical equipment by building hidden Markov model first, then with evolution particle Filtering algorithm, by known observation sequence, goes to calculate the posteriority of the status switch most possibly occurred under this observation sequence Probability density function, so as to realize the identification to household electrical appliance complexity electricity condition.

The method of the use electricity condition of identification electrical equipment of the present invention includes：

Step 1, the voltage x current data based on electrical equipment calculate characteristic parameter, build electric appliance working condition characteristic Collection, wherein, the characteristic parameter includes active power, reactive power, frequency, the root mean square of electric current, the root mean square harmony of voltage Wave component；

Step 2, structure electric appliance working condition hidden Markov model, electrician is used using what calculating in step 1 was obtained Make state characteristic data set to carry out model training and obtain optimal model parameters；

Step 3, with reference to the state transition equation and observational equation obtained based on optimal model parameters, utilize the filter of evolution particle Ripple algorithm, goes to calculate the posterior probability density letter of the status switch occurred under this observation sequence by known observation sequence Number, so as to realize the identification to electrical equipment complexity electricity condition.

Further, the voltage x current data of the electrical equipment are voltage sensor to be respectively adopted and current sensor is adopted in real time The voltage and current signal of current-collector, the analog signal of voltage, electric current is then converted into by A/D modular converters by data signal and Obtain.

Further, the computing formula of characteristic parameter is as follows described in electricity consumption working condition characteristic data set：

Active-power P=UI cos α；

Reactive power Q=UI sin α；

The root mean square of electric current

The root mean square of voltage

Wherein P is active power, and Q is reactive power, I_rmsFor the root mean square of electric current, U_rmsFor the root mean square of voltage.

Further, the hidden Markov model of electric appliance working condition is built, using calculating obtained use in step 1 Electric working condition characteristic data set carries out model training and obtains optimal model parameters including：

Set up electricity consumption status of processes set Q={ q₁,q₂,...,q_NAnd observation set V={ v₁,v₂,...,v_M, its In, set Q={ q₁,q₂,...,q_NBe used to describe the working condition of electrical equipment, N is the number of state, V={ v₁,v₂,...,v_M} It is all possible observation set, M is possible observation number；

The discrete state sequence S={ s of foundation electric process₁,s₂,...,s_t,...,s_TAnd observation sequence O={ o₁, o₂,...,o_t,...,o_T, wherein s_t∈ Q represent state of the electrical equipment in t, o_tRepresent the spy that electrical equipment is exported in t electrical equipment Parameter value is levied, T represents run time section；

Hidden Markov mould is determined based on initial state probability vector π, state transition equation coefficient A and observational equation coefficient Type, wherein, state transition equation coefficient A=[a_ij]_N×N,a_ij=P (s_t+1=q_j|s_t=q_i) represent to be in state q in moment t_i Under conditions of moment t+1 be transferred to state q_jProbability, observational equation coefficient B=[b_sk]_N×M, b_sk=P (o_t=v_k|s_t=q_i) table Show and be in state q in moment t_iUnder conditions of generation observation v_kProbability, initial state probability vector π=(π_i), π_i=P (s₁= q_i) it is that moment t=1 is in state q_iProbability；

It is determined that hidden Markov model in, by way of machine learning, with the characteristic ginseng value observed to mould Shape parameter λ=(A, B, π) is trained, and after training n times, the parameter value λ that n times are drawn=(A, B, π) averages, described The average value of model parameterFor optimal model parameter, A, B in optimal model parameters are optimal state The coefficient of equation of transfer and observational equation.

Further, it is determined that hidden Markov model in, by way of machine learning, joined with the feature that observes Times N=8 that numerical value is trained to model parameter λ=(A, B, π).

Further, with reference to the state transition equation and observational equation obtained based on optimal model parameters, evolution grain is utilized Sub- filtering algorithm, goes to calculate the posterior probability density of the status switch occurred under this observation sequence by known observation sequence Function, so as to realize that the identification to electrical equipment complexity electricity condition includes：

Step 1, sample initialization, i.e., at the k moment, sample out N number of sample according to known priori probability density function, make For initial antibody, the antibody is the identification electric appliance working condition particle to be used, wherein, each particle is usedRepresent, make k=1；

Step 2, sample is updated, i.e., when being hidden markov process with electric process, seen according to noisy Measured value, when N number of particle by initialization being updated into K using the observational equation and state transition equation of hidden Markov model The particle at quarter, calculates the weight of each particleAnd the weight of N number of particle is normalized, then will be N number of after renewal Particle is preserved as initial antibodies group, wherein：

Observational equation：

State transition equation：

v_kIt is the process noise of system, n_kThe observation noise of system, both random errors, they it is separate and It is each independent, x_kIt is the state at the k moment that will be calculated, z_kIt is the observation at k moment；

The affinity aff of step 3, each antibody calculated in N number of antibody_k(i) repulsive force between antibody and antibody rep_k(i, j), computing formula is as follows：

aff_k(i) it is smaller, show that antibody matches better with antigen, rep_k(i, j) is smaller, shows i-th of antibody and jth Individual antibody is more similar；

Step 4, clonal antibody, i.e., clone according to the affinity of each antibody to antibody, and the calculating of colony counts is public Formula is as follows：

Wherein round { } represents to round to nearest integer, can make the small particle colony counts of affinity from cos functions It is larger, and be distributed between 0-1, the principle of clone is the antibody for promoting affinity small, and suppresses the big antibody of affinity, with Enable the system to quickly converge on globally optimal solution；

Step 5, antibody made a variation, the formula of variation is as follows：

Wherein, randn () is the random number that one randomly selected meets N (0,1) distributions, and the principle of variation is to affine The small antibody variation amount of power is small, and big to the big antibody variation amount of affinity；

Step 6, the aff for selecting all antibody after optimal antibody, i.e. calculating variation_kAnd rep (i)_k(i, j), abandons rep_k (i,j)<An antibody in 0.0001, and by aff_k(i) it is ranked up, the N number of antibody selected above is preserved, returns to step Rapid 3, as the repulsive force rep of the antibody of gained_kStep 7 is carried out during (i, j) >=0.0001；

Step 7, according to the particle of N number of antibody of preservation as a new generation, calculate the status switch of electric appliance state in k The posterior probability density function at+1 moment, realizes the identification to electrical equipment complexity electricity condition, and computing formula is as follows：

Further, the calculating of the status switch of electric appliance state in the posterior probability density function at k+1 moment is obtained Formula includes：

According to noisy observation, worked with reference to observational equation (1) and state transition equation (2) recursive estimation electricity consumption The posterior probability density P (x of state_0:k|z_1:k), i.e., by known observation sequence, go to calculate what is under this observation sequence occurred The posterior probability density function of status switch, wherein, x_0:k={ x₀,x₁,...,x_kRepresent the shape produced by etching system when 0 to k State sequence, z_1:k={ z₁,z₂,...,z_kThe sequence of observations is represented, observation sequence is the spy of the output observed at each moment Parameter is levied, status switch is then the state at each moment and the one-to-one electrical equipment of observation sequence；

The set constituted using N number of particleTo represent the posterior probability density P (x of system_0:k|z_1:k), its In, x_0:k={ x₀,x₁,...,x_kThe status switch produced by etching system when 0 to k is represented,To be corresponding Weight, andThen the posterior probability density at K moment can be approximately with discrete weighted sample：

Wherein, δ is the Dirac function quoted；

Weight is selected by importance samplingWork as sample setBy importance sampling function q (x_0:k|z_1:k) sampling is when obtaining, then the weight of i-th of sample is：

Important sampling function is decomposed into：

q(x_0:k|z_1:k)=q (x_k|x_0:k-1,z_1:k)q(x_0:k-1|z_1:k-1) (11)

Will be by q (x_0:k-1|z_1:k-1) obtained sample setWith by q (x_k|x_0:k-1,z_1:k) obtained sample setMerge, Obtain new sample set

In order to obtain the renewal equation of weight, posterior probability density is expressed as：

(11) (12) formula is updated in (10) formula, obtaining weight more new formula is：

As q (x_k|x_0:k-1,z_1:k)=q (x_k|x_k-1,z_k), then importance sampling function relies only on x_k-1And z_k, weight amendment For：

Posterior probability density P (x_k|z_1:k) be：

Wherein, δ is the Dirac function quoted.

According to another aspect of the present invention, the present invention provides a kind of system for the use electricity condition for recognizing electrical equipment, the system Including：

Data acquisition unit, the voltage x current letter that voltage sensor and current sensor gather electrical equipment in real time is respectively adopted in it Number, the analog signal of voltage, electric current is then converted into by A/D modular converters by data signal；

Feature extraction unit, the voltage x current data of its electrical equipment gathered based on data acquisition unit calculate characteristic parameter, Electric appliance working condition characteristic data set is built, wherein, the characteristic parameter includes active power, reactive power, frequency, electricity The harmonious wave component of the root mean square of stream, the root mean square of voltage；

Hidden Markov model construction unit, it is used to build electric appliance working condition hidden Markov model, utilizes The electricity consumption working condition characteristic data set of feature extraction unit carries out model training and obtains optimal model parameters；

Electric appliance state recognition unit, it is combined based on the optimal model parameters in hidden Markov model construction unit Obtained state transition equation and observational equation, using Evolutionary particle filter, goes to calculate by known observation sequence The posterior probability density function of status switch under this observation sequence, so as to realize the identification to electrical equipment complexity electricity condition.

Further, the computing formula of the characteristic parameter described in the feature extraction unit of system is as follows：

Active-power P=UI cos α；

Reactive power Q=UI sin α；

The root mean square of electric current

The root mean square of voltage

Wherein P is active power, and Q is reactive power, I_rmsFor the root mean square of electric current, U_rmsFor the root mean square of voltage.

Further, hidden Markov configuration construction unit builds electric appliance working condition hidden Markov model, profit The electricity consumption working condition characteristic data set obtained with being calculated in feature extraction unit carries out model training and obtains optimal models ginseng Number includes：

Set up electricity consumption status of processes set Q={ q₁,q₂,...,q_NAnd observation set V={ v₁,v₂,...,v_M, its In, set Q={ q₁,q₂,...,q_NBe used to describe the working condition of electrical equipment, N is the number of state, V={ v₁,v₂,...,v_M} It is all possible observation set, M is possible observation number；

The discrete state sequence S={ s of foundation electric process₁,s₂,...,s_t,...,s_TAnd observation sequence O={ o₁, o₂,...,o_t,...,o_T, wherein s_t∈ Q represent state of the electrical equipment in t, o_tRepresent the spy that electrical equipment is exported in t electrical equipment Parameter value is levied, T represents run time section；

Hidden Markov mould is determined based on initial state probability vector π, state transition equation coefficient A and observational equation coefficient Type, wherein, state transition equation coefficient A=[a_ij]_N×N,a_ij=P (s_t+1=q_j|s_t=q_i) represent to be in state q in moment t_i Under conditions of moment t+1 be transferred to state q_jProbability, observational equation coefficient B=[b_sk]_N×M, b_sk=P (o_t=v_k|s_t=q_i) table Show and be in state q in moment t_iUnder conditions of generation observation v_kProbability, initial state probability vector π=(π_i), π_i=P (s₁= q_i) it is that moment t=1 is in state q_iProbability；

It is determined that hidden Markov model in, by way of machine learning, with the characteristic ginseng value observed to mould Shape parameter λ=(A, B, π) is trained, and after training n times, the parameter value λ that n times are drawn=(A, B, π) averages, described The average value of model parameterFor optimal model parameter, A, B in optimal model parameters are optimal state The coefficient of equation of transfer and observational equation.

Further, in the hidden Markov model that hidden Markov model construction unit is determined, machine learning is passed through Mode, times N=8 being trained with the characteristic ginseng value observed to model parameter λ=(A, B, π).

Further, the electric appliance state recognition unit combine based in hidden Markov model construction unit most State transition equation and observational equation that excellent model parameter is obtained, using Evolutionary particle filter, pass through known observation sequence Row go to calculate the posterior probability density function of the status switch under this observation sequence, so as to realize to electrical equipment complexity electricity condition Identification include：

Step 1, sample initialization, i.e., at the k moment, sample out N number of sample according to known priori probability density function, make For initial antibody, the antibody is the identification electric appliance working condition particle to be used, wherein, each particle is usedRepresent, make k=1；

Step 2, sample is updated, i.e., when being hidden markov process with electric process, seen according to noisy Measured value, when N number of particle by initialization being updated into K using the observational equation and state transition equation of hidden Markov model The particle at quarter, calculates the weight of each particleAnd the weight of N number of particle is normalized, then will be N number of after renewal Particle is preserved as initial antibodies group, wherein：

Observational equation：

State transition equation：

v_kIt is the process noise of system, n_kThe observation noise of system, both random errors, they it is separate and It is each independent；x_kIt is the state at the k moment that will be estimated；z_kIt is the observation at k moment；

The affinity aff of step 3, each antibody calculated in N number of antibody_k(i) repulsive force between antibody and antibody rep_k(i, j), computing formula is as follows：

aff_k(i) it is smaller, show that antibody matches better with antigen, rep_k(i, j) is smaller, shows i-th of antibody and jth Individual antibody is more similar；

Step 4, clonal antibody, i.e., clone according to the affinity of each antibody to antibody, and the calculating of colony counts is public Formula is as follows：

Wherein round { } represents to round to nearest integer, can make the small particle colony counts of affinity from cos functions It is larger, and be distributed between 0-1, the principle of clone is the antibody for promoting affinity small, and suppresses the big antibody of affinity, with Enable the system to quickly converge on globally optimal solution；

Step 5, antibody made a variation, the formula of variation is as follows：

Wherein, randn () is the random number that one randomly selected meets N (0,1) distributions, and the principle of variation is to affine The small antibody variation amount of power is small, and big to the big antibody variation amount of affinity；

Step 6, the aff for selecting all antibody after optimal antibody, i.e. calculating variation_kAnd rep (i)_k(i, j), abandons rep_k (i,j)<An antibody in 0.0001, and by aff_k(i) it is ranked up, the N number of antibody selected above is preserved, returns to step Rapid 3, as the repulsive force rep of the antibody of gained_kStep 7 is carried out during (i, j) >=0.0001；

Step 7, according to the particle of N number of antibody of preservation as a new generation, calculate the status switch of electric appliance state in k The posterior probability density function at+1 moment, realizes the identification to electrical equipment complexity electricity condition, and computing formula is as follows：

Further, electric appliance state recognition unit obtains the status switch of electric appliance state after the k+1 moment Testing the computing formula of probability density function includes：

According to noisy observation, worked with reference to observational equation (1) and state transition equation (2) recursive estimation electricity consumption The posterior probability density P (x of state_0:k|z_1:k), i.e., by known observation sequence, go to calculate what is under this observation sequence occurred The posterior probability density function of status switch, wherein, x_0:k={ x₀,x₁,...,x_kRepresent the shape produced by etching system when 0 to k State sequence, z_1:k={ z₁,z₂,...,z_kThe sequence of observations is represented, observation sequence is the spy of the output observed at each moment Parameter is levied, status switch is then the state at each moment and the one-to-one electrical equipment of observation sequence；

The set constituted using N number of particleTo represent the posterior probability density P (x of system_0:k|z_1:k), wherein, x_0:k={ x₀,x₁,...,x_kThe status switch produced by etching system when 0 to k is represented,For corresponding power Weight, andThen the posterior probability density at K moment can be approximately with discrete weighted sample：

Wherein, δ is the Dirac function quoted；

Weight is selected by importance samplingWork as sample setBy importance sampling function q (x_0:k|z_1:k) sampling is when obtaining, then the weight of i-th of sample is：

Important sampling function is decomposed into：

q(x_0:k|z_1:k)=q (x_k|x_0:k-1,z_1:k)q(x_0:k-1|z_1:k-1) (11)

Will be by q (x_0:k-1|z_1:k-1) obtained sample setWith by q (x_k|x_0:k-1,z_1:k) obtained sample setMerge, Obtain new sample set

In order to obtain the renewal equation of weight, posterior probability density is expressed as：

(11) (12) formula is updated in (10) formula, obtaining weight more new formula is：

As q (x_k|x_0:k-1,z_1:k)=q (x_k|x_k-1,z_k), then importance sampling function relies only on x_k-1And z_k, weight amendment For：

Posterior probability density P (x_k|z_1:k) be：

Wherein, δ is the Dirac function quoted.

The method and system of the use electricity condition of identification electrical equipment of the present invention is by building hidden Markov model, preferably Ground describes the use electric process of electrical equipment, passes through the training and study of the observation sequence to structure, it is determined that the optimized parameter of model, So as to construct identification electric appliance state optimum state equation of transfer and observational equation, on this basis, according to it is described most Excellent state transition equation and observational equation, using evolution particle algorithm, by known observation sequence, go to calculate and observe sequence herein The posterior probability density function of status switch under row, solves the problem of sample is degenerated in general particle algorithm, it will manually exempt from Epidemic disease algorithm is combined with particle filter algorithm, and the degradation phenomena of particle filter is preferably alleviated simultaneously with the diversity for increasing particle collection The problem of particle exhausts is solved, best particle can be used for the estimation of posterior probability density, identification is drastically increased The accuracy of the use electricity condition of electrical equipment, is the distribution that user reasonably plans electricity consumption, mitigates the waste of electricity consumption there is provided possible.

Brief description of the drawings

By reference to the following drawings, the illustrative embodiments of the present invention can be more fully understood by：

Fig. 1 is the flow chart of the method for the use electricity condition of the identification electrical equipment of the specific embodiment of the invention；

Fig. 2 is the flow chart of the evolution particle algorithm of the specific embodiment of the invention；And

Fig. 3 is the structure chart of the system of the use electricity condition of the identification electrical equipment of the specific embodiment of the invention.

Embodiment

The illustrative embodiments of the present invention are introduced with reference now to accompanying drawing, however, the present invention can use many different shapes Formula is implemented, and it is to disclose at large and fully there is provided these embodiments to be not limited to embodiment described herein The present invention, and fully pass on the scope of the present invention to person of ordinary skill in the field.For showing for being illustrated in the accompanying drawings Term in example property embodiment is not limitation of the invention.In the accompanying drawings, identical cells/elements are attached using identical Icon is remembered.

Unless otherwise indicated, term (including scientific and technical terminology) used herein has to person of ordinary skill in the field It is common to understand implication.Further it will be understood that the term limited with usually used dictionary, is appreciated that and it The linguistic context of association area has consistent implication, and is not construed as Utopian or excessively formal meaning.

Fig. 1 is the flow chart of the method for the use electricity condition of the identification electrical equipment of the specific embodiment of the invention.As shown in figure 1, The method 100 of the use electricity condition of identification electrical equipment of the present invention is since step 101.

In step 101, the voltage and current signal that voltage sensor and current sensor gather electrical equipment in real time is respectively adopted, so The voltage x current number of electrical equipment obtained from the analog signal of voltage, electric current is converted into data signal by A/D modular converters afterwards According to.

In step 102, the voltage x current data based on electrical equipment calculate characteristic parameter, build electric appliance working condition feature Data set, wherein, the characteristic parameter includes active power, reactive power, frequency, the root mean square of electric current, the root mean square of voltage Harmonious wave component.

Preferably, the computing formula of characteristic parameter is as follows described in electricity consumption working condition characteristic data set：

Active-power P=UI cos α；

Reactive power Q=UI sin α；

The root mean square of electric current

The root mean square of voltage

Wherein P is active power, and Q is reactive power, I_rmsFor the root mean square of electric current, U_rmsFor the root mean square of voltage.

In step 103, electric appliance working condition hidden Markov model is built, is obtained using calculating in step 102 Electricity consumption working condition characteristic data set carries out model training and obtains optimal model parameters.

Preferably, the hidden Markov model of electric appliance working condition is built, using calculating obtained use in step 102 Electric working condition characteristic data set carries out model training and obtains optimal model parameters including：

Set up electricity consumption status of processes set Q={ q₁,q₂,...,q_NAnd observation set V={ v₁,v₂,...,v_M, its In, set Q={ q₁,q₂,...,q_NBe used to describe the working condition of electrical equipment, N is the number of state, V={ v₁,v₂,...,v_M} It is all possible observation set, M is possible observation number；

The discrete state sequence S={ s of foundation electric process₁,s₂,...,s_t,...,s_TAnd observation sequence O={ o₁, o₂,...,o_t,...,o_T, wherein s_t∈ Q represent state of the electrical equipment in t, o_tRepresent the spy that electrical equipment is exported in t electrical equipment Parameter value is levied, T represents run time section；

Hidden Markov mould is determined based on initial state probability vector π, state transition equation coefficient A and observational equation coefficient Type, wherein, state transition equation coefficient A=[a_ij]_N×N,a_ij=P (s_t+1=q_j|s_t=q_i) represent to be in state q in moment t_i Under conditions of moment t+1 be transferred to state q_jProbability, observational equation coefficient B=[b_sk]_N×M, b_sk=P (o_t=v_k|s_t=q_i) table Show and be in state q in moment t_iUnder conditions of generation observation v_kProbability, initial state probability vector π=(π_i), π_i=P (s₁= q_i) it is that moment t=1 is in state q_iProbability；

It is determined that hidden Markov model in, by way of machine learning, with the characteristic ginseng value observed to mould Shape parameter λ=(A, B, π) is trained, and after training n times, the parameter value λ that n times are drawn=(A, B, π) averages, described The average value of model parameterFor optimal model parameter, A, B in optimal model parameters are optimal state The coefficient of equation of transfer and observational equation；

Preferably, it is determined that hidden Markov model in, by way of machine learning, with the characteristic parameter observed It is worth times N=8 being trained to model parameter λ=(A, B, π).

In step 104, with reference to the state transition equation and observational equation obtained based on optimal model parameters, evolution grain is utilized Sub- filtering algorithm, goes to calculate the posterior probability density of the status switch occurred under this observation sequence by known observation sequence Function, so as to realize the identification to electrical equipment complexity electricity condition.

Fig. 2 is the flow chart of the evolution particle algorithm of the specific embodiment of the invention.As shown in Fig. 2 combining based on optimal State transition equation and observational equation that model parameter is obtained, using Evolutionary particle filter, pass through known observation sequence Go to calculate the posterior probability density function of status switch occurred under this observation sequence, so as to realize to the complicated electricity consumption shape of electrical equipment The knowledge method for distinguishing 200 of state is since step 201.

In step 201, sample initialization, that is, at the k moment, is sampled out N number of sample according to known priori probability density function This, as initial antibody, the antibody is the identification electric appliance working condition particle to be used, wherein, each grain Son is usedRepresent, make k=1.

In step 202, sample is updated, i.e., when being hidden markov process with electric process, according to noise Observation, using hidden Markov model observational equation and state transition equation will by initialization N number of particle update For the particle at K moment, the weight of each particle is calculatedAnd the weight of N number of particle is normalized, after then updating N number of particle as initial antibodies group preserved, wherein：

Observational equation：

State transition equation：

v_kIt is the process noise of system, n_kThe observation noise of system, both random errors, they it is separate and It is each independent, x_kIt is the state at the k moment that will be calculated, z_kIt is the observation at k moment.

In step 203, the affinity aff of each antibody in N number of antibody is calculated_k(i) row between antibody and antibody Repulsion rep_k(i, j), computing formula is as follows：

aff_k(i) it is smaller, show that antibody matches better with antigen, rep_k(i, j) is smaller, shows i-th of antibody and jth Individual antibody is more similar；

In step 204, clonal antibody clones according to the affinity of each antibody to antibody, the meter of colony counts Calculate formula as follows：

Wherein round { } represents to round to nearest integer, can make the small particle colony counts of affinity from cos functions It is larger, and be distributed between 0-1, the principle of clone is the antibody for promoting affinity small, and suppresses the big antibody of affinity, with Enable the system to quickly converge on globally optimal solution；

In step 205, antibody is made a variation, the formula of variation is as follows：

Wherein, randn () is the random number that one randomly selected meets N (0,1) distributions, and the principle of variation is to affine The small antibody variation amount of power is small, and big to the big antibody variation amount of affinity；

In step 206, optimal antibody is selected, that is, calculates the aff of all antibody after variation_kAnd rep (i)_k(i, j), is abandoned rep_k(i,j)<An antibody in 0.0001, and by aff_k(i) it is ranked up, the N number of antibody selected above is preserved, and is returned Step 3 is returned, as the repulsive force rep of the antibody of gained_kStep 207 is carried out during (i, j) >=0.0001.

In step 207, according to particle of the N number of antibody of preservation as a new generation, the state sequence of electric appliance state is calculated The posterior probability density function at k+1 moment is listed in, the identification to electrical equipment complexity electricity condition is realized, computing formula is as follows：

Preferably, the status switch for obtaining electric appliance state is public in the calculating of the posterior probability density function at k+1 moment Formula includes：

According to noisy observation, worked with reference to observational equation (1) and state transition equation (2) recursive estimation electricity consumption The posterior probability density P (x of state_0:k|z_1:k), i.e., by known observation sequence, go to calculate what is under this observation sequence occurred The posterior probability density function of status switch, wherein, x_0:k={ x₀,x₁,...,x_kRepresent the shape produced by etching system when 0 to k State sequence, z_1:k={ z₁,z₂,...,z_kThe sequence of observations is represented, observation sequence is the spy of the output observed at each moment Parameter is levied, status switch is then the state at each moment and the one-to-one electrical equipment of observation sequence；

The set constituted using N number of particleTo represent the posterior probability density P (x of system_0:k|z_1:k), wherein, x_0:k={ x₀,x₁,...,x_kThe status switch produced by etching system when 0 to k is represented,For corresponding power Weight, andThen the posterior probability density at K moment can be approximately with discrete weighted sample：

Wherein, δ is the Dirac function quoted；

Weight is selected by importance samplingWork as sample setBy importance sampling function q (x_0:k|z_1:k) sampling is when obtaining, then the weight of i-th of sample is：

Important sampling function is decomposed into：

q(x_0:k|z_1:k)=q (x_k|x_0:k-1,z_1:k)q(x_0:k-1|z_1:k-1) (11)

Will be by q (x_0:k-1|z_1:k-1) obtained sample setWith by q (x_k|x_0:k-1,z_1:k) obtained sample setMerge, Obtain new sample set

In order to obtain the renewal equation of weight, posterior probability density is expressed as：

(11) (12) formula is updated in (10) formula, obtaining weight more new formula is：

As q (x_k|x_0:k-1,z_1:k)=q (x_k|x_k-1,z_k), then importance sampling function relies only on x_k-1And z_k, weight amendment For：

Posterior probability density P (x_k|z_1:k) be：

Wherein, δ is the Dirac function quoted.

Fig. 3 is the structure chart of the system of the use electricity condition of the identification electrical equipment of the specific embodiment of the invention.As shown in figure 3, It is provided by the present invention identification electrical equipment use electricity condition system 300 include data acquisition unit 301, feature extraction unit 302, Hidden Markov model construction unit 303 and electric appliance state recognition unit 304.

Data acquisition unit 301, the voltage electricity that voltage sensor and current sensor gather electrical equipment in real time is respectively adopted in it Signal is flowed, the analog signal of voltage, electric current is then converted into by A/D modular converters by data signal；

Feature extraction unit 302, the voltage x current data of its electrical equipment gathered based on data acquisition unit 301 calculate feature Parameter, builds electric appliance working condition characteristic data set, wherein, the characteristic parameter includes active power, reactive power, frequency Rate, the root mean square of electric current, the harmonious wave component of root mean square of voltage；

Hidden Markov model construction unit 303, it is used to build electric appliance working condition hidden Markov model, profit Model training is carried out with the electricity consumption working condition characteristic data set of feature extraction unit 302 and obtains optimal model parameters；

Electric appliance state recognition unit 304, it is combined based on the optimal mould in hidden Markov model construction unit 303 State transition equation and observational equation that shape parameter is obtained, using Evolutionary particle filter, are gone by known observation sequence The posterior probability density function of the status switch under this observation sequence is calculated, so as to realize the knowledge to electrical equipment complexity electricity condition Not.

Preferably, the computing formula of the characteristic parameter described in the feature extraction unit 302 of system is as follows：

Active-power P=UI cos α；

Reactive power Q=UI sin α；

The root mean square of electric current

The root mean square of voltage

Wherein P is active power, and Q is reactive power, I_rmsFor the root mean square of electric current, U_rmsFor the root mean square of voltage.

Preferably, hidden Markov configuration construction unit 303 builds electric appliance working condition hidden Markov model, profit The electricity consumption working condition characteristic data set obtained with being calculated in feature extraction unit carries out model training and obtains optimal models ginseng Number includes：

Set up electricity consumption status of processes set Q={ q₁,q₂,...,q_NAnd observation set V={ v₁,v₂,...,v_M, its In, set Q={ q₁,q₂,...,q_NBe used to describe the working condition of electrical equipment, N is the number of state, V={ v₁,v₂,...,v_M} It is all possible observation set, M is possible observation number；

The discrete state sequence S={ s of foundation electric process₁,s₂,...,s_t,...,s_TAnd observation sequence O={ o₁, o₂,...,o_t,...,o_T, wherein s_t∈ Q represent state of the electrical equipment in t, o_tRepresent the spy that electrical equipment is exported in t electrical equipment Parameter value is levied, T represents run time section；

Hidden Markov mould is determined based on initial state probability vector π, state transition equation coefficient A and observational equation coefficient Type, wherein, state transition equation coefficient A=[a_ij]_N×N,a_ij=P (s_t+1=q_j|s_t=q_i) represent to be in state q in moment t_i Under conditions of moment t+1 be transferred to state q_jProbability, observational equation coefficient B=[b_sk]_N×M, b_sk=P (o_t=v_k|s_t=q_i) table Show and be in state q in moment t_iUnder conditions of generation observation v_kProbability, initial state probability vector π=(π_i), π_i=P (s₁= q_i) it is that moment t=1 is in state q_iProbability；

It is determined that hidden Markov model in, by way of machine learning, with the characteristic ginseng value observed to mould Shape parameter λ=(A, B, π) is trained, and after training n times, the parameter value λ that n times are drawn=(A, B, π) averages, described The average value of model parameterFor optimal model parameter, A, B in optimal model parameters are optimal state The coefficient of equation of transfer and observational equation.

Preferably, in the hidden Markov model that hidden Markov model construction unit 303 is determined, machine learning is passed through Mode, times N=8 being trained with the characteristic ginseng value observed to model parameter λ=(A, B, π).

Preferably, the electric appliance state recognition unit 304, which is combined, is based in hidden Markov model construction unit 303 Optimal model parameters obtained state transition equation and observational equation, using Evolutionary particle filter, pass through known see Sequencing row go to calculate the posterior probability density function of the status switch under this observation sequence, so as to realize to the complicated electricity consumption of electrical equipment The knowledge method for distinguishing 200 of state is since step 201.

In step 201, sample initialization, that is, at the k moment, is sampled out N number of sample according to known priori probability density function This, as initial antibody, the antibody is the identification electric appliance working condition particle to be used, wherein, each grain Son is usedRepresent, make k=1.

In step 202, sample is updated, i.e., when being hidden markov process with electric process, according to noise Observation, using hidden Markov model observational equation and state transition equation will by initialization N number of particle update For the particle at K moment, the weight of each particle is calculatedAnd the weight of N number of particle is normalized, after then updating N number of particle as initial antibodies group preserved, wherein：

Observational equation：

State transition equation：

v_kIt is the process noise of system, n_kThe observation noise of system, both random errors, they it is separate and It is each independent, x_kIt is the state at the k moment that will be calculated, z_kIt is the observation at k moment.

In step 203, the affinity aff of each antibody in N number of antibody is calculated_k(i) row between antibody and antibody Repulsion rep_k(i, j), computing formula is as follows：

aff_k(i) it is smaller, show that antibody matches better with antigen, rep_k(i, j) is smaller, shows i-th of antibody and jth Individual antibody is more similar；

In step 204, clonal antibody clones according to the affinity of each antibody to antibody, the meter of colony counts Calculate formula as follows：

Wherein round { } represents to round to nearest integer, can make the small particle colony counts of affinity from cos functions It is larger, and be distributed between 0-1, the principle of clone is the antibody for promoting affinity small, and suppresses the big antibody of affinity, with Enable the system to quickly converge on globally optimal solution；

In step 205, antibody is made a variation, the formula of variation is as follows：

Wherein, randn () is the random number that one randomly selected meets N (0,1) distributions, and the principle of variation is to affine The small antibody variation amount of power is small, and big to the big antibody variation amount of affinity；

In step 206, optimal antibody is selected, that is, calculates the aff of all antibody after variation_kAnd rep (i)_k(i, j), is abandoned rep_k(i,j)<An antibody in 0.0001, and by aff_k(i) it is ranked up, the N number of antibody selected above is preserved, and is returned Step 3 is returned, as the repulsive force rep of the antibody of gained_kStep 207 is carried out during (i, j) >=0.0001.

In step 207, according to particle of the N number of antibody of preservation as a new generation, the state sequence of electric appliance state is calculated The posterior probability density function at k+1 moment is listed in, the identification to electrical equipment complexity electricity condition is realized, computing formula is as follows：

Preferably, the status switch for obtaining electric appliance state is public in the calculating of the posterior probability density function at k+1 moment Formula includes：

According to noisy observation, worked with reference to observational equation (1) and state transition equation (2) recursive estimation electricity consumption The posterior probability density P (x of state_0:k|z_1:k), i.e., by known observation sequence, go to calculate what is under this observation sequence occurred The posterior probability density function of status switch, wherein, x_0:k={ x₀,x₁,...,x_kRepresent the shape produced by etching system when 0 to k State sequence, z₁:_k={ z₁,z₂,...,z_kThe sequence of observations is represented, observation sequence is the spy of the output observed at each moment Parameter is levied, status switch is then the state at each moment and the one-to-one electrical equipment of observation sequence；

The set constituted using N number of particleTo represent the posterior probability density P (x of system₀:_k|z₁:_k), its In, x₀:_k={ x₀,x₁,...,x_kThe status switch produced by etching system when 0 to k is represented,To be corresponding Weight, andThen the posterior probability density at K moment can be approximately with discrete weighted sample：

Wherein, δ is the Dirac function quoted；

Weight is selected by importance samplingWork as sample setBy importance sampling function q (x₀:_k|z₁:_k) sampling is when obtaining, then the weight of i-th of sample is：

Important sampling function is decomposed into：

q(x_0:k|z_1:k)=q (x_k|x_0:k-1,z_1:k)q(x_0:k-1|z_1:k-1) (11)

Will be by q (x_0:k-1|z_1:k-1) obtained sample setWith by q (x_k|x_0:k-1,z_1:k) obtained sample setMerge, Obtain new sample set

In order to obtain the renewal equation of weight, posterior probability density is expressed as：

(11) (12) formula is updated in (10) formula, obtaining weight more new formula is：

As q (x_k|x_0:k-1,z_1:k)=q (x_k|x_k-1,z_k), then importance sampling function relies only on x_k-1And z_k, weight amendment For：

Posterior probability density P (x_k|z_1:k) be：

Wherein, δ is the Dirac function quoted.

Normally, all terms used in the claims are all solved according to them in the usual implication of technical field Release, unless clearly defined in addition wherein.All references " one/described/be somebody's turn to do【Device, component etc.】" all it is opened ground At least one example in described device, component etc. is construed to, unless otherwise expressly specified.Any method disclosed herein Step need not all be run with disclosed accurate order, unless explicitly stated otherwise.

Claims (13)

1. a kind of method for the use electricity condition for recognizing electrical equipment, it is characterised in that methods described includes：

Step 1, the voltage x current data based on electrical equipment calculate characteristic parameter, build electric appliance working condition characteristic data set, Wherein, the characteristic parameter includes active power, reactive power, frequency, the root mean square of electric current, the harmonious wavelength-division of root mean square of voltage Amount；

Step 2, electric appliance working condition hidden Markov model is built, worked shape using obtained electricity consumption is calculated in step 1 State characteristic data set carries out model training and obtains optimal model parameters；

Step 3, with reference to the state transition equation and observational equation obtained based on optimal model parameters, utilize evolution particle filter to calculate Method, goes to calculate the posterior probability density function of the status switch occurred under this observation sequence by known observation sequence, from And realize the identification to electrical equipment complexity electricity condition.

2. according to the method described in claim 1, it is characterised in that the voltage x current data of electrical equipment described in step 1 are difference The voltage and current signal of electrical equipment is gathered in real time using voltage sensor and current sensor, then by A/D modular converters electricity Pressure, the analog signal of electric current are converted into obtained from data signal.

3. according to the method described in claim 1, it is characterised in that the computing formula of characteristic parameter described in step 1 is as follows：

Active-power P=UI cos α；

Reactive power Q=UI sin α；

The root mean square of electric current

The root mean square of voltage

Wherein P is active power, and Q is reactive power, I_rmsFor the root mean square of electric current, U_rmsFor the root mean square of voltage.

4. according to the method described in claim 1, it is characterised in that build electric appliance working condition hidden Markov model, The electricity consumption working condition characteristic data set obtained using being calculated in step 1 carries out model training and obtains optimal model parameters bag Include：

Set up electricity consumption status of processes set Q={ q₁,q₂,...,q_NAnd observation set V={ v₁,v₂,...,v_M, wherein, collection Close Q={ q₁,q₂,...,q_NBe used to describe the working condition of electrical equipment, N is the number of state, V={ v₁,v₂,...,v_MIt is all Possible observation set, M is possible observation number；

The discrete state sequence S={ s of foundation electric process₁,s₂,...,s_t,...,s_TAnd observation sequence O={ o₁,o₂,..., o_t,...,o_T, wherein s_t∈ Q represent state of the electrical equipment in t, o_tRepresent the characteristic parameter that electrical equipment is exported in t electrical equipment Value, T represents run time section；

Hidden Markov model is determined based on initial state probability vector π, state transition equation coefficient A and observational equation coefficient, Wherein, state transition equation coefficient A=[a_ij]_N×N,a_ij=P (s_t+1=q_j|s_t=q_i) represent to be in state q in moment t_iBar The part lower moment, t+1 was transferred to state q_jProbability, observational equation coefficient B=[b_sk]_N×M, b_sk=P (o_t=v_k|s_t=q_i) represent Moment t is in state q_iUnder conditions of generation observation v_kProbability, initial state probability vector π=(π_i), π_i=P (s₁=q_i) be Moment t=1 is in state q_iProbability；

It is determined that hidden Markov model in, by way of machine learning, with the characteristic ginseng value observed to model join Number λ=(A, B, π) is trained, and after training n times, the parameter value λ that n times are drawn=(A, B, π) averages, the model The average value of parameterFor optimal model parameter, A, B in optimal model parameters are optimal state transfer The coefficient of equation and observational equation.

5. method according to claim 4, it is characterised in that it is determined that hidden Markov model in, pass through engineering The mode of habit, times N=8 being trained with the characteristic ginseng value observed to model parameter λ=(A, B, π).

6. according to the method described in claim 1, it is characterised in that with reference to the state transfer side obtained based on optimal model parameters Journey and observational equation, using Evolutionary particle filter, go to calculate the appearance under this observation sequence by known observation sequence Status switch posterior probability density function, so as to realize that the identification to electrical equipment complexity electricity condition includes：

Step 1, sample initialization, i.e., at the k moment, sample out N number of sample, as first according to known priori probability density function The antibody of beginning, the antibody is the identification electric appliance working condition particle to be used, wherein, each particle is usedRepresent, make k=1；

Step 2, sample is updated, i.e., when being hidden markov process with electric process, according to noisy observation, N number of particle by initialization is updated to the K moment using the observational equation and state transition equation of hidden Markov model Particle, calculates the weight of each particleAnd the weight of N number of particle is normalized, then by N number of particle after renewal As initial antibodies, group is preserved, wherein：

Observational equation：

State transition equation：

$x_k=A\&times;f_k(x_{k-1},v_{k-1})=A\&times;\&lsqb;\frac{1}{2}{x}_{k-1}+\frac{25x_{k-1}}{1+x_{k-1}^2}+8cos\left(\frac{6}{5}\right(k-1)+v_k\&rsqb;---\left(2\right)$

$w_k^i\&Proportional;w_{k-1}^i\frac{P\left(z_k\right|x_k^i)P\left(x_k^i\right|x_{k-1}^i)}{q\left(x_k^i\right|x_{k-1}^i,z_k)}---\left(3\right)$

${\hat{w}}_k^i=\frac{{\hat{w}}_k^i}{\underset{i=1}{\overset{N}{\&Sigma;}}{\hat{w}}_k^i}---\left(4\right)$

v_kIt is the process noise of system, n_kIt is the observation noise of system, both random errors, they are separate and respective It is independent, x_kIt is the state at the k moment that will be calculated, z_kIt is the observation at k moment；

The affinity aff of step 3, each antibody calculated in N number of antibody_k(i) the repulsive force rep between antibody and antibody_k (i, j), computing formula is as follows：

${\mathrm{aff}}_k\left(i\right)=1-w_i^k---\left(5\right)$

${\mathrm{rep}}_k(i,j)=|x_k^i-x_k^j|---\left(6\right)$

aff_k(i) it is smaller, show that antibody matches better with antigen, rep_k(i, j) is smaller, shows i-th of antibody and resists for j-th Body is more similar；

Step 4, clonal antibody, i.e., clone according to the affinity of each antibody to antibody, and the computing formula of colony counts is such as Under：

${\mathrm{Clo}}_k\left(i\right)=round\{N\&times;cos\&lsqb;\frac{\mathrm{\&pi;}}{2}\&times;{\mathrm{aff}}_k\left(i\right)\&rsqb;\}---\left(7\right)$

Wherein round { } represents to round to nearest integer, from cos functions can make the small particle colony counts of affinity compared with Greatly, and it is distributed between 0-1, the principle of clone is the antibody for promoting affinity small, and suppresses the big antibody of affinity, so that System can quickly converge on globally optimal solution；

Step 5, antibody made a variation, the formula of variation is as follows：

$x_k^i=x_k^i+{\mathrm{aff}}_k\left(i\right)randn\left(\right)---\left(7\right)$

Wherein, randn () is the random number that one randomly selected meets N (0,1) distributions, and the principle of variation is small to affinity Antibody variation amount it is small, it is and big to the big antibody variation amount of affinity；

Step 6, the aff for selecting all antibody after optimal antibody, i.e. calculating variation_kAnd rep (i)_k(i, j), abandons rep_k(i,j)< An antibody in 0.0001, and by aff_k(i) it is ranked up, the N number of antibody selected above is preserved, return to step 3, when The repulsive force rep of the antibody of gained_kStep 7 is carried out during (i, j) >=0.0001；

Step 7, according to the particle of N number of antibody of preservation as a new generation, calculate the status switch of electric appliance state in k+1 The posterior probability density function at quarter, realizes the identification to electrical equipment complexity electricity condition, and computing formula is as follows：

$P\left(x_k\right|z_{1:k})\&ap;\underset{i=1}{\overset{N}{\&Sigma;}}{w}_k^i\mathrm{\&delta;}(x_k-x_k^i)---\left(8\right)$

7. method according to claim 6, it is characterised in that obtain the state of the electric appliance state in the step 7 Sequence includes in the computing formula of the posterior probability density function at k+1 moment：

According to noisy observation, with reference to observational equation (1) and state transition equation (2) recursive estimation electricity consumption working condition Posterior probability density P (x_0:k|z_1:k), i.e., by known observation sequence, go to calculate the state occurred under this observation sequence The posterior probability density function of sequence, wherein, x_0:k={ x₀,x₁,...,x_kRepresent the state sequence produced by etching system when 0 to k Row, z_1:k={ z₁,z₂,...,z_kThe sequence of observations is represented, observation sequence is the feature ginseng of the output observed at each moment Number, status switch is then the state at each moment and the one-to-one electrical equipment of observation sequence；

The set constituted using N number of particleTo represent the posterior probability density P (x of system_0:k|z_1:k), wherein, x_0:k ={ x₀,x₁,...,x_kThe status switch produced by etching system when 0 to k is represented,For corresponding weight, AndThen the posterior probability density at K moment can be approximately with discrete weighted sample：

$P\left(x_{0:k}\right|z_{1:k})\&ap;\underset{i}{\overset{N}{\&Sigma;}}{w}_k^i\mathrm{\&delta;}(x_{0:k}-x_{0:k}^i)---\left(9\right)$

Wherein, δ is the Dirac function quoted；

Weight is selected by importance samplingWork as sample setBy importance sampling function q (x_0:k| z_1:k) sampling is when obtaining, then the weight of i-th of sample is：

$w_k^i\&Proportional;\frac{P\left(x_{0:k}^i\right|z_{1:k})}{q\left(x_{0:k}^i\right|z_{1:k})}---\left(10\right)$

Important sampling function is decomposed into：

q(x_0:k|z_1:k)=q (x_k|x_0:k-1,z_1:k)q(x_0:k-1|z_1:k-1) (11)

Will be by q (x_0:k-1|z_1:k-1) obtained sample setWith by q (x_k|x_0:k-1,z_1:k) obtained sample setMerge, obtain New sample set

In order to obtain the renewal equation of weight, posterior probability density is expressed as：

$\begin{array}{c}P\left(x_{0:k}|z_{1:k}\right)=\frac{P\left(z_k|x_{0:k},z_{1:k-1}\right)P\left(x_{0:k}|z_{1:k-1}\right)}{P\left(z_k|z_{1:k-1}\right)}\\ =\frac{P\left(z_k|x_{0:k},z_{1:k-1}\right)P\left(x_k|x_{0:k-1},z_{1:k-1}\right)P\left(x_{0:k-1}|z_{1:k-1}\right)}{P\left(z_k|z_{1:k-1}\right)}\\ =\frac{P\left(z_k|x_k\right)P\left(x_k|x_{k-1}\right)P\left(x_{0:k-1}|z_{1:k-1}\right)}{P\left(z_k|z_{1:k-1}\right)}\\ \&Proportional;P\left(z_k|x_k\right)P\left(x_k|x_{k-1}\right)P\left(x_{0:k-1}|z_{1:k-1}\right)\end{array}---\left(12\right)$

(11) (12) formula is updated in (10) formula, obtaining weight more new formula is：

$\begin{array}{c}{w}_k^i\&Proportional;\frac{P\left(z_k|x_k^i\right)P\left(x_k^i|x_{k-1}^i\right)P\left(x_{0:k-1}^i|z_{1:k-1}\right)}{q\left(x_k^i|x_{0:k-1}^i,z_{1:k}\right)q\left(x_{0:k-1}^i|z_{1:k-1}\right)}\\ =w_{k-1}^i\frac{P\left(z_k|x_k^i\right)P\left(x_k^i|x_{k-1}^i\right)}{q\left(x_k^i|x_{0:k-1}^i,z_{1:k}\right)}\end{array}---\left(13\right)$

As q (x_k|x_0:k-1,z_1:k)=q (x_k|x_k-1,z_k), then importance sampling function relies only on x_k-1And z_k, weight is modified to：

$w_k^i\&Proportional;w_{k-1}^i\frac{P\left(z_k\right|x_k^i)P\left(x_k^i\right|x_{k-1}^i)}{q\left(x_k^i\right|x_{k-1}^i,z_k)}---\left(14\right)$

Posterior probability density P (x_k|z_1:k) be：

$P\left(x_k\right|z_{1:k})\&ap;\underset{i=1}{\overset{N}{\&Sigma;}}{w}_k^i\mathrm{\&delta;}(x_k-x_k^i)---\left(8\right)$

Wherein, δ is the Dirac function quoted.

8. a kind of system for the use electricity condition for recognizing electrical equipment, it is characterised in that the system includes：

Data acquisition unit, the voltage and current signal that voltage sensor and current sensor gather electrical equipment in real time is respectively adopted in it, Then the analog signal of voltage, electric current is converted into by A/D modular converters by data signal；

Feature extraction unit, the voltage x current data of its electrical equipment gathered based on data acquisition unit calculate characteristic parameter, build Electric appliance working condition characteristic data set, wherein, the characteristic parameter includes active power, reactive power, frequency, electric current The harmonious wave component of root mean square of root mean square, voltage；

Hidden Markov model construction unit, it is used to build electric appliance working condition hidden Markov model, utilizes feature The electricity consumption working condition characteristic data set of extraction unit carries out model training and obtains optimal model parameters；

Electric appliance state recognition unit, it is combined and obtained based on the optimal model parameters in hidden Markov model construction unit State transition equation and observational equation, using Evolutionary particle filter, go to calculate by known observation sequence and see herein The posterior probability density function of status switch under sequencing row, so as to realize the identification to electrical equipment complexity electricity condition.

9. system according to claim 8, it is characterised in that the calculating of the characteristic parameter described in feature extraction unit is public Formula is as follows：

Active-power P=UI cos α；

Reactive power Q=UI sin α；

The root mean square of electric current

The root mean square of voltage

Wherein P is active power, and Q is reactive power, I_rmsFor the root mean square of electric current, U_rmsFor the root mean square of voltage.

10. system according to claim 8, it is characterised in that hidden Markov configuration construction unit builds electric appliance Working condition hidden Markov model, is carried out using obtained electricity consumption working condition characteristic data set is calculated in feature extraction unit Model training simultaneously obtains optimal model parameters and included：

Set up electricity consumption status of processes set Q={ q₁,q₂,...,q_NAnd observation set V={ v₁,v₂,...,v_M, wherein, collection Close Q={ q₁,q₂,...,q_NBe used to describe the working condition of electrical equipment, N is the number of state, V={ v₁,v₂,...,v_MIt is all Possible observation set, M is possible observation number；

The discrete state sequence S={ s of foundation electric process₁,s₂,...,s_t,...,s_TAnd observation sequence O={ o₁,o₂,..., o_t,...,o_T, wherein s_t∈ Q represent state of the electrical equipment in t, o_tRepresent the characteristic parameter that electrical equipment is exported in t electrical equipment Value, T represents run time section；

Hidden Markov model is determined based on initial state probability vector π, state transition equation coefficient A and observational equation coefficient, Wherein, state transition equation coefficient A=[a_ij]_N×N,a_ij=P (s_t+1=q_j|s_t=q_i) represent to be in state q in moment t_iBar The part lower moment, t+1 was transferred to state q_jProbability, observational equation coefficient B=[b_sk]_N×M, b_sk=P (o_t=v_k|s_t=q_i) represent Moment t is in state q_iUnder conditions of generation observation v_kProbability, initial state probability vector π=(π_i), π_i=P (s₁=q_i) be Moment t=1 is in state q_iProbability；

It is determined that hidden Markov model in, by way of machine learning, with the characteristic ginseng value observed to model join Number λ=(A, B, π) is trained, and after training n times, the parameter value λ that n times are drawn=(A, B, π) averages, the model The average value of parameterFor optimal model parameter, A, B in optimal model parameters are optimal state transfer The coefficient of equation and observational equation.

11. system according to claim 10, it is characterised in that the hidden horse determined in hidden Markov model construction unit In Er Kefu models, by way of machine learning, model parameter λ=(A, B, π) is carried out with the characteristic ginseng value observed Times N=8 of training.

12. system according to claim 8, it is characterised in that the electric appliance state recognition unit is combined based on hidden State transition equation and observational equation that optimal model parameters in Markov model construction unit are obtained, utilize evolution particle Filtering algorithm, goes to calculate the posterior probability density function of the status switch under this observation sequence by known observation sequence, So as to realize that the identification to electrical equipment complexity electricity condition includes：

Step 1, sample initialization, i.e., at the k moment, sample out N number of sample, as first according to known priori probability density function The antibody of beginning, the antibody is the identification electric appliance working condition particle to be used, wherein, each particle is usedRepresent, make k=1；

Step 2, sample is updated, i.e., when being hidden markov process with electric process, according to noisy observation, N number of particle by initialization is updated to the K moment using the observational equation and state transition equation of hidden Markov model Particle, calculates the weight of each particleAnd the weight of N number of particle is normalized, then by N number of particle after renewal As initial antibodies, group is preserved, wherein：

Observational equation：

State transition equation：

$x_k=A\&times;f_k(x_{k-1},v_{k-1})=A\&times;\&lsqb;\frac{1}{2}{x}_{k-1}+\frac{25x_{k-1}}{1+x_{k-1}^2}+8cos\left(\frac{6}{5}\right(k-1)+v_k\&rsqb;---\left(2\right)$

$w_k^i\&Proportional;w_{k-1}^i\frac{P\left(z_k\right|x_k^i)P\left(x_k^i\right|x_{k-1}^i)}{q\left(x_k^i\right|x_{k-1}^i,z_k)}---\left(3\right)$

${\hat{w}}_k^i=\frac{{\hat{w}}_k^i}{\underset{i=1}{\overset{N}{\&Sigma;}}{\hat{w}}_k^i}---\left(4\right)$

v_kIt is the process noise of system, n_kIt is the observation noise of system, both random errors, they are separate and respective It is independent；x_kIt is the state at the k moment that will be estimated；z_kIt is the observation at k moment；

The affinity aff of step 3, each antibody calculated in N number of antibody_k(i) the repulsive force rep between antibody and antibody_k (i, j), computing formula is as follows：

${\mathrm{aff}}_k\left(i\right)=1-w_i^k---\left(5\right)$

${\mathrm{rep}}_k(i,j)=|x_k^i-x_k^j|---\left(6\right)$

aff_k(i) it is smaller, show that antibody matches better with antigen, rep_k(i, j) is smaller, shows i-th of antibody and resists for j-th Body is more similar；

Step 4, clonal antibody, i.e., clone according to the affinity of each antibody to antibody, and the computing formula of colony counts is such as Under：

${\mathrm{Clo}}_k\left(i\right)=round\{N\&times;cos\&lsqb;\frac{\mathrm{\&pi;}}{2}\&times;{\mathrm{aff}}_k\left(i\right)\&rsqb;\}---\left(7\right)$

Wherein round { } represents to round to nearest integer, from cos functions can make the small particle colony counts of affinity compared with Greatly, and it is distributed between 0-1, the principle of clone is the antibody for promoting affinity small, and suppresses the big antibody of affinity, so that System can quickly converge on globally optimal solution；

Step 5, antibody made a variation, the formula of variation is as follows：

$x_k^i=x_k^i+{\mathrm{aff}}_k\left(i\right)randn\left(\right)---\left(7\right)$

Wherein, randn () is the random number that one randomly selected meets N (0,1) distributions, and the principle of variation is small to affinity Antibody variation amount it is small, it is and big to the big antibody variation amount of affinity；

Step 6, the aff for selecting all antibody after optimal antibody, i.e. calculating variation_kAnd rep (i)_k(i, j), abandons rep_k(i,j)< An antibody in 0.0001, and by aff_k(i) it is ranked up, the N number of antibody selected above is preserved, return to step 3, when The repulsive force rep of the antibody of gained_kStep 7 is carried out during (i, j) >=0.0001；

Step 7, according to the particle of N number of antibody of preservation as a new generation, calculate the status switch of electric appliance state in k+1 The posterior probability density function at quarter, realizes the identification to electrical equipment complexity electricity condition, and computing formula is as follows：

$P\left(x_k\right|z_{1:k})\&ap;\underset{i=1}{\overset{N}{\&Sigma;}}{w}_k^i\mathrm{\&delta;}(x_k-x_k^i)---\left(8\right)$

13. system according to claim 12, it is characterised in that electric appliance state recognition unit obtains electric appliance shape The status switch of state includes in the computing formula of the posterior probability density function at k+1 moment：

According to noisy observation, with reference to observational equation (1) and state transition equation (2) recursive estimation electricity consumption working condition Posterior probability density P (x_0:k|z_1:k), i.e., by known observation sequence, go to calculate the state occurred under this observation sequence The posterior probability density function of sequence, wherein, x_0:k={ x₀,x₁,...,x_kRepresent the state sequence produced by etching system when 0 to k Row, z_1:k={ z₁,z₂,...,z_kThe sequence of observations is represented, observation sequence is the feature ginseng of the output observed at each moment Number, status switch is then the state at each moment and the one-to-one electrical equipment of observation sequence；

The set constituted using N number of particleTo represent the posterior probability density P (x of system_0:k|z_1:k), wherein, x_0:k ={ x₀,x₁,...,x_kThe status switch produced by etching system when 0 to k is represented,For corresponding weight, AndThen the posterior probability density at K moment can be approximately with discrete weighted sample：

$P\left(x_{0:k}\right|z_{1:k})\&ap;\underset{i}{\overset{N}{\&Sigma;}}{w}_k^i\mathrm{\&delta;}(x_{0:k}-x_{0:k}^i)---\left(9\right)$

Wherein, δ is the Dirac function quoted；

Weight is selected by importance samplingWork as sample setBy importance sampling function q (x_0:k| z_1:k) sampling is when obtaining, then the weight of i-th of sample is：

$w_k^i\&Proportional;\frac{P\left(x_{0:k}^i\right|z_{1:k})}{q\left(x_{0:k}^i\right|z_{1:k})}---\left(10\right)$

Important sampling function is decomposed into：

q(x_0:k|z_1:k)=q (x_k|x_0:k-1,z_1:k)q(x_0:k-1|z_1:k-1) (11)

Will be by q (x_0:k-1|z_1:k-1) obtained sample setWith by q (x_k|x_0:k-1,z_1:k) obtained sample set xⁱ _kMerge, obtain New sample set

In order to obtain the renewal equation of weight, posterior probability density is expressed as：

$\begin{array}{c}P\left(x_{0:k}|z_{1:k}\right)=\frac{P\left(z_k|x_{0:k},z_{1:k-1}\right)P\left(x_{0:k}|z_{1:k-1}\right)}{P\left(z_k|z_{1:k-1}\right)}\\ =\frac{P\left(z_k|x_{0:k},z_{1:k-1}\right)P\left(x_k|x_{0:k-1},z_{1:k-1}\right)P\left(x_{0:k-1}|z_{1:k-1}\right)}{P\left(z_k|z_{1:k-1}\right)}\\ =\frac{P\left(z_k|x_k\right)P\left(x_k|x_{k-1}\right)P\left(x_{0:k-1}|z_{1:k-1}\right)}{P\left(z_k|z_{1:k-1}\right)}\\ \&Proportional;P\left(z_k|x_k\right)P\left(x_k|x_{k-1}\right)P\left(x_{0:k-1}|z_{1:k-1}\right)\end{array}---\left(12\right)$

(11) (12) formula is updated in (10) formula, obtaining weight more new formula is：

$\begin{array}{c}{w}_k^i\&Proportional;\frac{P\left(z_k|x_k^i\right)P\left(x_k^i|x_{k-1}^i\right)P\left(x_{0:k-1}^i|z_{1:k-1}\right)}{q\left(x_k^i|x_{0:k-1}^i,z_{1:k}\right)q\left(x_{0:k-1}^i|z_{1:k-1}\right)}\\ =w_{k-1}^i\frac{P\left(z_k|x_k^i\right)P\left(x_k^i|x_{k-1}^i\right)}{q\left(x_k^i|x_{0:k-1}^i,z_{1:k}\right)}\end{array}---\left(13\right)$

As q (x_k|x_0:k-1,z_1:k)=q (x_k|x_k-1,z_k), then importance sampling function relies only on x_k-1And z_k, weight is modified to：

$w_k^i\&Proportional;w_{k-1}^i\frac{P\left(z_k\right|x_k^i)P\left(x_k^i\right|x_{k-1}^i)}{q\left(x_k^i\right|x_{k-1}^i,z_k)}---\left(14\right)$

Posterior probability density P (x_k|z_1:k) be：

$P\left(x_k\right|z_{1:k})\&ap;\underset{i=1}{\overset{N}{\&Sigma;}}{w}_k^i\mathrm{\&delta;}(x_k-x_k^i)---\left(8\right)$

Wherein, δ is the Dirac function quoted.