CN1057621A

CN1057621A - Elevator control gear

Info

Publication number: CN1057621A
Application number: CN91104339A
Authority: CN
Inventors: 辻伸太郎
Original assignee: Mitsubishi Electric Corp
Current assignee: Mitsubishi Electric Corp
Priority date: 1990-06-19
Filing date: 1991-06-19
Publication date: 1992-01-08
Anticipated expiration: 2006-06-19
Also published as: GB2246214A; KR940009412B1; GB9113294D0; GB2246214B; JP2573726B2; CN1021700C; JPH0449183A; KR920000604A

Abstract

A kind of elevator control gear comprises: the input data converting apparatus that traffic state data is transformed into the form that can be used as neural network input data; The prediction that constitutes described neural network arrives delays degree arithmetical device, includes the input layer that is taken into described input data, to be equivalent to data that described prediction arrives the delay degree as the output layer of importing data be in the interlayer of having set coefficient of weight between described input layer and the output layer; Described output data is transformed into the output data converting means that to control the form of using in the purpose action in regulation; Study forms device with data, and the input data that store, prediction arrival delay degree and the described actual delay degree that arrives are exported with data as one group of study; Utilize described study to arrive the correcting device of the coefficient of weight of delay degree arithmetical device with the data correction prediction.

Description

Elevator control gear

The present invention relates to a kind of elevator control gear, it can predict the delay degree (late degree) of each floor of elevator car arrives accurately.

In the past, in being set up in parallel the lift appliance of many cars, carried out group's management running usually,, allocation scheme was for example arranged as one of each population management drive manner.So-called allocation scheme, record waiting space call out after immediately to each car computing evaluation number, the conduct of selection evaluation number the best should provide the distribution car of service, only allows distribute car to go to reply above-mentioned waiting space calling, tries to achieve the raising of running efficiency and the shortening of wait time like this.

In nearest group management elevator, generally carry out so-called " advance notice shows ", promptly before car arrives, turn on the arrival advance notice lamp that distributes car, with the notice person of waiting.The person of waiting shows according to this advance notice can learn in advance which car will provide service, thereby can wait securely before this car.

But the car of sometimes doing the advance notice demonstration can respond car call, and arrives (this is called " predicting error ") earlier than the advance notice car.Because this advance notice error causes confusion for the person of waiting, do not wish to occur, so in order to prevent this situation, the past people have proposed all schemes.

For example, in the elevator group manage apparatus on being recorded in special public clear 62-47787 communique, after recording the waiting space calling, the summation of the prediction latency time square value that all waiting spaces were called out when this waiting space calling hypothesis was distributed to each car, obtained as the wait time evaluation number, equally, be equipped with the value that obtains after the weight of regulation for the summation that waiting space is called out the hypothesis advance notice probability of failure (expression is made the car of advance notice than the index of the possibility of arrival earlier that advance notice is arranged) that all waiting spaces are called out when distributing to each car, as advance notice error evaluation number, these wait time evaluation numbers and advance notice error evaluation number are added up, obtain comprehensive evaluation value, select this comprehensive evaluation value for minimum car as distributing car.At this moment predict probability of failure according to the following formula computing:

Advance notice probability of failure=elder generation arrives probability * car call probability of occurrence

Here, the so-called probability that arrives earlier is to consider that the car prediction has deviation the time of advent, representing that other car shows that than advance notice the car of (distributions) arrives the index that advance notice shows the possibility of waiting space (with whether stop at this waiting space irrelevant) earlier, is poor (elder generation's arrival time difference) computing of predicting the time of advent according to the car of the overlapping degree of the car of advance notice and the car probability distribution time of advent without prior notice and advance notice and car without prior notice.In addition, so-called car call probability of occurrence is the index that expression car without prior notice has the possibility of the car call that goes to advance notice demonstration waiting space, when car without prior notice had the car call that goes to the waiting space of predicting demonstration, the car call probability of occurrence was " 1.0 "; When not having, then consider the situation that floor passenger on the way may enter elevator, wait according to the statistics of last descending stair number and set the car call probability of occurrence.

Open in the lift group management control method in the clear 58-125580 communique being recorded in the spy, with car call elder generation's arrival time, prediction time of advent of prediction time of advent of give distributing car and the unallocated car that arrives because of car call poor (car call arrives the time difference earlier) adds weight, as a key element of waiting space calling evaluation number.

In addition, open in the lift group management control method in the clear 58-36865 communique being recorded in the spy, arrive at distance (number of floor levels) till the floor that car call elder generation arrival situation takes place with the car that car call elder generation arrival situation takes place, arrive time weight earlier for car call, call out a key element of evaluation number as waiting space with this.

Like this, the predictor that postpones the arrival degree according to car in the near future, that is, predict arrival delay degree (being to add value that obtains after the weight of regulation or the like) to advance notice probability of failure, elder generation's arrival probability, elder generation's arrival time difference, car call elder generation's arrival time, first arrival time of car call, carry out the waiting space call distribution, by this, when shortening the waiting space calling waiting time, reduce the generation of advance notice fault phenomenon.

In addition, the advance notice probability of failure called out of people's waiting space of having proposed that also the waiting space call distribution is arrived to new record is the mode (special public clear 58-56708 communique) of minimum car; When new record to the advance notice probability of failure called out of waiting space car distributed when surpassing specified value and postpones till this probability of failure and reduce to mode (special public clear 58-56708 communique) below the specified value; Not only obtain the advance notice probability of failure that new waiting space is called out, also obtain the advance notice probability of failure of the waiting space calling that has distributed, and be the mode (special public clear 62-45151 communique) of the car distribution waiting space calling of minimum to summation.

But, if arriving the delay degree, prediction loses correctness, and so, the evaluation number of being tried to achieve has just lost the meaning that is used as a reference value of selecting the distribution car, and the result just becomes and can not suppress to predict the generation of fault phenomenon.Like this, prediction arrives the performance generation big influence of the correctness of delay degree to group's management.

In addition, the mode that the somebody has proposed to utilize waiting space to distribute the means outside the control to prevent to predict fault phenomenon, for example, if detect the car without prior notice of estimating to respond and arrive earlier the calling wait place than the advance notice car, then output accelerates to arrive instruction to the advance notice car, it is reduced on the way stop and walk the needed time (spy opens clear 63-12577 communique) on the floor; Perhaps in contrast, output delay arrives instruction and gives car without prior notice, it is increased on the way stop and walk the needed time (spy opens clear 63-8180 communique) on the floor; The advance notice lamp that distributes car is lighted in postponement, till the advance notice probability of failure satisfies defined terms (spy opens clear 53-2850 communique).

In addition, to prevent to predict the method for fault phenomenon purpose, people have also proposed the group management control method that various employing predictions arrive the delay degree except that above-mentioned.For example, in order to prevent the person's of waiting confusion, when as if the advance notice error will take place, the car (spy opens clear 52-153551 communique) that arrives to waiting space announcement elder generation, owing to car call is estimated the car that arrives earlier near waiting space during to predetermined distance, will distribute and the advance notice object changes this expectations car (spy opens clear 52-29057 communique, the spy opens clear 58-188275 communique) of arrival earlier into.In such group's management control, the correctness that prediction arrives the delay degree still has big influence to the performance of group's management.

But, arrive delays degree if obtain accurately, just be necessary that the time of advent is predicted in computing correctly, car call probability of occurrence, the probability distribution of the time of advent etc.Before, the computing of predicting the time of advent is basically as putting down in writing in the special public clear 54-20742 communique, obtain walking required time (travel time) according to the spacing of car position and object floor, the number of times that stops from floor is on the way obtained and is stopped required time (length of the halt), again these times are added up, as predicting the time of advent.In order to improve this prediction precision of prediction of the time of advent, must take variety of way, as revise the mode (special public clear 57-40074 communique) of length of the halt predictor according to the state of car on the car position floor, revise the mode (special public clear 57-40072 communique) of length of the halt predictor according to the kind of last descending stair number and answering call, according to the mode (special public clear 63-34111 communique) of the prediction that car call takes place being revised the length of the halt predictor, consider that car reverses on the floor on the way, high call counter-rotating, lowest call is reversed) thus mode (special public clear 54-16293 communique) of travel time predictor or the like revised.

And for example the spy opens put down in writing in the flat 1-275381 communique such, and the someone proposes a kind of group management control apparatus, adopts and the cooresponding neural network of people's brain neuron, selects to call out cooresponding distribution car with waiting space according to the computing of this network.But, reckon without and how to improve the operational precision that prediction arrives the delay degree.

As mentioned above, elevator control gear in the past arrives the delay degree for the prediction of computing correctly, various key elements have been considered, that is, the kind of the predetermined state that stops floor, car status, answering call, predetermined stop in the floor on the descending stair number prediction, the prediction of car call takes place, the prediction of the distribution of calling out for new waiting space, prediction, the present key elements such as traffic behavior of each floor of counter-rotating floor, they are carried out computing as a key element in the calculating formula respectively.But, if on the computing basis that is added with whole these key elements, predict, want computing correctly prediction arrive the delay degree with corresponding to the time be engraved in the words of the traffic behavior that intricately changing, then the arithmetic expression of prediction arrival delay degree becomes more complicated, since also there is boundary in people's ability, so, be that the new arithmetic expression of target developing also is difficult to improve operational precision.On the other hand, if carry out detailed prediction computing, the problem that causes increasing operation time, can not being implemented in the function of determining to distribute car when the record waiting space is called out and predicting is arranged then.

The present invention makes for addressing the above problem a little.Its purpose is to provide a kind of elevator control gear, by the flexibly prediction approximate with actual traffic state and volume of traffic, can calculate the accurate prediction arrival delay degree that approaches the actual arrival of car delay degree.

Elevator control gear of the present invention comprises: will comprise that the traffic state data of car position, service direction and this calling of replying is transformed into the input data converting apparatus of the form that can be used as the use of neural network input data; The prediction that constitutes neural network arrives delays degree arithmetical device, it comprise the input layer that is taken into the input data, with the data that are equivalent to predict arrival delay degree as the output layer of output data and be in input layer and output layer between set the interlayer of coefficient of weight; Output data is transformed into the output data converting means of the form in the action that can be used for regulation control purpose.

Elevator control gear of the present invention also can comprise: study forms device with data, when elevator enters predetermined period at work, the prediction arrival delay degree and the input data storage at this moment of regulation waiting space are got up, simultaneously, car stayed in or the arrival delay degree during by the regulation waiting space arrives delays degree as reality and stores, the input data of being stored, prediction are arrived delays degree and reality arrival delay degree is exported with data as one group of study; Utilize study to arrive the correcting device of the coefficient of weight of delay degree arithmetical device with the data correction prediction.

Among the present invention, traffic state data is taken into neural network, obtains prediction by the computing that approaches the actual arrival of car delay degree and arrive the delay degree, utilize this prediction to arrive the delay degree, abide by the regulation purpose, carry out the control of elevator action.

In addition, in the present invention, according to predicting the outcome of calculating and at that time traffic state data and measured data, form the study data, and automatically revise prediction according to study with data and arrive coefficient of weight in the delay degree arithmetical device (neural network), by this, carry out and actual traffic state and the approximate computing of prediction flexibly of traffic demand.

Fig. 1 is the functional block diagram that shows the unitary construction of one embodiment of the invention;

Fig. 2 is the block diagram of the general configuration of group manage apparatus in the displayed map 1;

Fig. 3 specifically illustrates the data converting apparatus among Fig. 1 and predicts the block diagram that arrives delay degree arithmetical device;

Fig. 4 is the diagram of circuit that generally shows the group's management program among the ROM that is stored in Fig. 2;

Fig. 5 is that the hypothesis of No. 1 machine in the concrete displayed map 4 divides timing to arrive the diagram of circuit of delay degree predictor;

Fig. 6 is that the diagram of circuit that data form program is commonly used in concrete displayed map 4 middle schools;

Fig. 7 is the diagram of circuit of revision program in the concrete displayed map 4;

Fig. 8 shows that specifically the used study of another embodiment of the present invention forms the diagram of circuit of program with data.

In the accompanying drawing, same-sign is represented identical or cooresponding part.

Below, with reference to description of drawings one embodiment of the present of invention.Fig. 1 is the functional block diagram that shows the integral body formation of one embodiment of the invention, and Fig. 2 is the block diagram of group manage apparatus general configuration in the displayed map 1.

Among Fig. 1, group manage apparatus 10 is made of following array apparatus 10A-10D, 10F and 10G on function, for example controls a plurality of car control setup 11-13(, is used for machine-No. 3 machine No. 1).

Waiting space call recording device for electric 10A carries out record and the elimination that each floor waiting space is called out (waiting space on the uplink and downlink direction is called out), simultaneously, the elapsed time in computing is recorded from the waiting space calling (that is, time length).

Prediction latency time and prediction arrival delay degree (advance notice probability of failure) that the distribution device 10B that selects best car to distribute in order to serve the waiting space calling calls out according to waiting space calculate evaluation number, distribute this evaluation number to be minimum car.

Data converting apparatus 10C comprises: the input data converting apparatus that car position, service direction, car load, this calling of replying traffic state datas such as (waiting space after car call or the distribution are called out) is transformed into the form of the input data use that can be used as neural network; The output data (being equivalent to predict the data that arrive the delay degree) of neural network is transformed into the output data converting means of the form in the action (for example, comprising the computing of the evaluation number of prediction latency time) that can be used for regulation control purpose.

Carry out the prediction that the prediction of each car arrives the computing of delays degree according to time range and arrive delays degree arithmetical device 10D as described later, comprise by be taken into the input layer of importing data, with the data that are equivalent to predict arrival delay degree be output data output layer, be in the neural network that the interlayer of having set coefficient of weight between input layer and the output layer constitutes.

Study with data form prediction that device 10F stores each car arrive delays degree and the input data (traffic state data) of this moment and with the relevant measured data (teacher's data) of arrival delay degree (advance notice fault rate) of each car after this, they are exported with data as learning.

Correcting device 10G utilizes study to arrive the function of neural network among the delay degree arithmetical device 10D with data study and correction prediction.

The car control setup 11-13 that is used for No. 1 machine-No. 3 machine is a same structure, and for example, as described later, the car control setup 11 that No. 1 machine is used is made of well-known device 11A-11E.

Waiting space is called out cancellation element 11A and is called out output waiting space calling erasure signal for the waiting space of each floor.Car call recording device 11B writes down the car call of each floor.Arrive advance notice lamp control device 11C and control lighting of each floor arrival advance notice lamp (not shown).Operation controller 11D is in order to determine the service direction of car, or the response car call and the waiting space of dispensing call out and the walking and the stop of control car, controlling device for doors 11E controls closing of car gangway door.

Among Fig. 2, group manage apparatus 10 is made of well-known microcomputer, promptly by the MPU(microprocessing unit) or CPU101, ROM102, RAM103, input circuit 104 and output circuit 105 formations.

The assignment key signal 14 that comes from the assignment key of each floor, be input to the input circuit 104 from No. 1 machine-No. 3 machine status signal of car control setup 11-13.Export to modulating signal 15 of ensconcing the assignment key lamp each assignment key and the command signal of giving car control setup 11-13 from output circuit 105.

Fig. 3 is data converting apparatus 10C and the functional block diagram of predicting the relation that arrives delay degree arithmetical device 10D in the concrete displayed map 1.

Among Fig. 3, the input data converting apparatus is promptly imported data varitron unit 10CA and output data converting means, and promptly output data varitron unit 10CB has constituted the data converting apparatus 10C among Fig. 1.The prediction arrival delay degree arithmetic element 10DA that is inserted between input data varitron unit 10CA and the output data varitron unit 10CB is formed by neural network, and the prediction that it constructs Fig. 1 arrives the prediction interpretative subroutine that adopts among the delay degree arithmetical device 10D.

The form of traffic state datas such as input data varitron unit 10CA conversion car position, service direction, car load, this calling of replying (waiting space that is car call and dispensing is called out), traffic flow statistics feature (in 5 minutes using escalator number, in 5 minutes descending stair number) makes it to be used as the input data of neural network 1 0DA.

The form of the output data of output data varitron unit 10CB conversion neural network 1 0DA (being equivalent to arrive the delay degree) makes it to be used for the evaluation number computing of waiting space call distribution action.

Neural network, that is prediction arrive delays degree arithmetic element 10DA by be taken into input layer 10DA1 from the input data of input data varitron unit 10CA, with the data that are equivalent to predict arrival delay degree be the output layer 10DA3 of output data and be in input layer 10DA1 and output layer 10DA3 between set coefficient of weight interlayer 10DA2 constitute.These a few layers of 10DA1-10DA3 connect with network each other, all are made of a plurality of nodes.

Here, be made as N respectively as if node number with input layer 10DA1, interlayer 10DA2 and output layer 10DA3 ₁, N ₂, N ₃, then the node of output layer 10DA3 is counted N ₃Be expressed as: N ₃=2(FL-1)

Wherein, FL: the number of floor levels in building

The node of input layer 10DA1 and interlayer 10DA2 is counted N ₁And N ₂Determine according to the kind and the car platform number of the number of floor levels FL in building, employed input data respectively.

If setting variable i, j, k are:

i＝1，2，…，N ₁

j＝1，2，…，N ₂

k＝1，2，…，N ₃

Then the input value and the output value table of i node of input layer 10DA1 are shown xa1(i) and ya1(i), the input value and the output valve of j node of interlayer 10DA2 are xa2(j) and ya2(j), the input value and the output value table of k node of output layer 10DA3 are shown xa3(k) and ya3(k).

If i node of an input layer 10DA1 and interlayer 10DA2 j internodal coefficient of weight is made as wal(i, j), j node of interlayer 10DA2 and output layer 10DA3 k internodal coefficient of weight is made as wa2(j, and k), then the relation table of the input value of each node and output valve is shown:

ya1（i）＝1/[1+exp{-xa1（i）｝] …（1）

xa2（j）＝∑{wa1（i，j）×ya1（i）} …（2）

(i=1-N ₁The summation formula)

ya2（j）＝1/[1+exp{-xa2（j）}] …（3）

xa3（k）＝∑{wa2（j，k）×ya2（j）} …（4）

(j=1-N ₂The summation formula)

ya3（k）＝1/[1+exp{-xa3（k）}] …（5）

Wherein: 0≤wa1(i, j)≤1

0≤wa2（j，k）≤1

Below, with reference to Fig. 4, group's management activities of the one embodiment of the invention shown in Fig. 1-Fig. 3 is described.

At first, group manage apparatus 10 is according to well-known input routine (step 31), is taken into assignment key signal 14 and from the status signal of car control setup 11-13.Here, include erasure signal that car position, direction of travel, stop or walking states, door opening and closing state, car load, car call, waiting space call out etc. in the status signal of input.

Then, judge that according to well-known waiting space call record program (step 32) the waiting space call record still eliminates, and the assignment key lamp lights or extinguishes, simultaneously, the time length that the computing waiting space is called out.

Then, judge whether to record new waiting space and call out C(step 33), if record, then divide the arrival delay degree predictor (step 34) of timing to calculate hypothesis when new waiting space calling C is distributed to No. 1 machine by the hypothesis that is used for No. 1 machine, No. 1 each waiting space k(k=1 of machine, 2 ..., N ₃) pairing prediction arrival delay degree Tal(k).

Similarly, divide arrival the delays degree predictor (step 35) of timing to calculate hypothesis when waiting space calling C is distributed to No. 2 machines by No. 2 machines hypothesis, No. 2 each waiting space k(k=1 of machine, 2 ..., N ₃) pairing prediction arrival delay degree Ta2(k).In addition, arrival the delays degree predictor (step 36) when distributing No. 3 machines with hypothesis calculates hypothesis No. 3 each waiting space k(k=1 of machine when waiting space calling C is distributed to No. 3 machines, and 2 ..., N ₃) pairing prediction arrival delay degree Ta3(k).

In addition, call out C ignoring new waiting space, it is not distributed in No. 1 machine-No. 3 machine under any one situation, arrival delay degree predictor (step 37-39) when operation is not done to suppose to distribute, union goes out each waiting space k(k=1 of machine-No. 3 machine, 2 No. 1,, N ₃) pairing prediction arrival delay degree Tb1(k)-Tb3(k).

After this, utilize allocator (step 40), according to each the waiting space k(k=1 that in step 34-39, has calculated, 2 ..., N ₃) prediction arrive delay degree Ta1(k)-Ta3(k) and Tb1(k)-Tb3(k), calculate evaluation number W ₁-W ₃, the car of getting this evaluation number minimum is as formal distribution car.Set the advance notice instruction of calling out the cooresponding assignment command of C with waiting space for the car that distributes like this.In the public clear 58-48464 communique of spy, put down in writing evaluation number W ₁-W ₃An example of compute mode.

Below, utilize output program (step 41) that above-mentioned such assignment key modulating signal of setting 15 is sent to waiting space, simultaneously, distributed intelligence and approaching signal etc. is sent to car control setup 11-13.

Form in the program (step 42) with data in study, will arrive as the prediction of the traffic state data after the input data conversion, each waiting space delays degree and thereafter each car measured data of arriving the delay degree store, export with data as learning.

Use the study data in the revision program (step 43), revise the network coefficient of weight that prediction arrives delay degree arithmetical device 10D.

Like this, group manage apparatus 10 is execution in step 31-43 repeatedly, and a plurality of lift cars are carried out group's management control.

Below, with reference to Fig. 5, be example with step 34, specify the action of the arrival delay degree predictor of each step 34-39.

At first, suppose that new waiting space is called out C distributes to machine No. 1, be formed for being input to the distribution waiting space call data (step 50) among the input data varitron unit 10CA.

In

step

35,36, suppose that waiting space is called out C distributes to No. 2 machines, No. 3 machines, form and distribute the waiting space call data, in step 37-39, will not suppose that the distribution waiting space call data under the distribution condition directly are used for input as distribution waiting space call data.

Below, from the traffic state data of input, take out the data (car position, service direction, car load, car call, distribution waiting space calling) relevant and represent the data (in 5 minutes using escalator number, in 5 minutes descending stair number) of present traffic flow statistics feature, they are transformed into and predict the pairing input data of each node xa1(1 of the input layer 10DA1 of arrival delay degree arithmetic element 10DA with playing the car that should the computing prediction arrives delays degree now)-xa1(N ₁) (step 51).So-called car load is the ratio of relative rated load, nominal load.

Here, if building number of floor levels FL is made as 12 layers, for waiting space number f, f=1,2 ... 11 represent 1,2 respectively ..., the 11st buildings the up direction waiting space, f=12,13 ... 22 represent 12,11 respectively ..., 2 waiting spaces of line direction downstairs, so, such as " the car position floor is f, and service direction is upwards " and so on car status be with the value representation that is normalized to 0-1:

xa1（f）＝1

xa1（i）＝0

（i＝1，2，…，22，i≠f）

Car load xa1(23) divided by the maxim NT that can get _Max(for example, 120%), thus be normalized to the value of 0-1.The car call xa1(24 in building, the 1st building-12)-xa1(35) write down, with " 1 " expression, Unrecorded words, with " 0 " expression.Xa1(36 is called out in the 1st building-11 upstairs distribution waiting space of line direction)-xa1(46) as distributing,, then use " 0 " expression with " 1 " expression, unappropriated words.Xa1(47 is called out in the 12nd buildings-2 distribution waiting space of line direction downstairs)-xa1(57) distribute, with " 1 " expression, unappropriated words, with " 0 " expression.The number that takes a lift in will try to achieve according to the volume of traffic in past statistics 5 minutes is divided by the maxim NN that can get _Max(for example, 100 people), thereby with the 1st building-11 5 minutes using escalator number xa1(58 of line direction upstairs)-xa1(68) be normalized to the value of 0-1.Equally, use maxim NN _MaxAfter removing, with the 12nd buildings-2 using escalator number xa1(69 in 5 minutes of line direction downstairs)-xa1(79), the 1st building-11 goes out terraced number xa1(80 in 5 minutes of line direction upstairs)-xa1(90) and the 12nd buildings-2 downstairs line direction go out terraced number xa1(91 in last 5 minute)-xa1(101) normalization method.

The method of input data normalization is not limited to said method, also can represents car position and service direction respectively.For example, when the car position floor is f, the input value xa1(1 of the 1st node of expression car position floor) be expressed as:

xa1（1）＝f/FL

To represent the input value xa1(2 of the 2nd node of cage operation direction), up direction represents with " 1 " that with "+1 " expression, down direction directionless usefulness " 0 " is represented.

Like this, set input data by step 51 to input layer 10DA1 after, carry out network operations by step 52-56 again, be used to predict the arrival delay degree when supposition is distributed to No. 1 machine to new waiting space calling C.

At first, utilize input data xa1(i), by the output valve ya1(i of (1) formula computing input layer 10DA1) (step 52).

Then, at the output valve ya1(i that tries to achieve by (1) formula) on multiply by coefficient of weight wa1(i, j), and, with regard to i=1-N ₁Obtain summation, again by the input value xa2(j of (2) formula computing interlayer 10DA2) (step 53).

Follow again, utilize the input value xa2(j that tries to achieve by (2) formula), by the output valve ya2(j of (3) formula computing interlayer 10DA2) (step 54).

Then, at the output valve ya2(j that tries to achieve by (3) formula) on multiply by coefficient of weight wa2(j, k), and, with regard to j=1-N ₂Obtain summation, with the input value xa3(k of (4) formula computing output layer 10DA3) (step 55).

And then, utilize the input value xa3(k that tries to achieve by (4) formula) calculate the output valve ya3(k of output layer 10DA3 from (5) formula) (step 56).

As mentioned above, after prediction arrives the network operations end of delay degree, by the output data varitron unit 10CB conversion output valve ya3(i among Fig. 1)-ya3(N ₃) form, determine that final prediction arrives delay degree (advance notice probability of failure) (step 57).

At this moment, each node of output layer 10DA3 is corresponding to the waiting space on all directions, the output valve ya3(1 of 1-Section 11 point)-ya3(11) be used for respectively determining 1,2 ..., 11 predictions of line direction waiting space upstairs arrive delay degree operation values.The output valve ya3(12 of 12-the 22nd node)-ya3(22) be respectively applied for and determine that the prediction of down direction waiting space arrives the operation values of delay degree.

In addition, because the output valve ya3(k of k node) (k=1,2 ..., N ₃) be normalized in the scope of 0-1, so, can be directly used in the evaluation number computing of waiting space call distribution.Therefore, the prediction of waiting space k arrives delay degree T(k) be expressed as:

T（k）＝ya3（k） …（6）

Like this, because in arriving delay degree predictor (step 34-39), embody traffic behavior and predict the causal relationship that arrives the delay degree with network, traffic state data is taken in the neural network, calculate prediction and arrive the delay degree, so, can obtain the prediction arrival delay degree that approaches actual advance notice probability of failure with the inaccessiable precision of mode institute in the past.Arrive the delay degree according to this prediction again, select the distribution car corresponding to the waiting space calling, therefore, can suppress to predict the generation of fault phenomenon reliably, can shorten wait time, simultaneously, confusion is avoided.

But this network depends on the coefficient of weight wa1(i that connects each node in the neural network 1 0DA, j) and wa2(j, k) changes, therefore, by study, suitably change, revise coefficient of weight wa1(i, j) and wa2(j, k), thereby can determine that more suitable prediction arrives the delay degree.

Below, with reference to Fig. 6 and Fig. 7, illustrate forming under program (step 42) and revision program (step 43) situation one embodiment of the present of invention with data by learning to form device 10F and correcting device 10G execution study with data.Study in this case (correction of network) utilizes contrary biography method to carry out effectively.So-called contrary biography method, be utilize the output data of network with result from measured data and the desirable output data (teacher's data) of control target etc. between error, revise the coefficient of weight of connection network.

Form with data among Fig. 6 of program (step 42) being shown specifically study, at first, new study is set forms permission with data, and, judge whether just finish the distribution (step 61) of new waiting space calling C.

Form permission if be provided with study with data, and, carried out the distribution that waiting space is called out C, distribute the traffic state data xa1(1 of car in the time of then will distributing)-xa1(N ₁) and be equivalent to the output data ya3(1 that the prediction of each waiting space at this moment arrives the delay degree)-ya3(N ₃) store, as the part (teacher data) of m study with data.In addition, remove the formation permission of new study, the actual measurement instruction of actual arrival delay degree is set with data, simultaneously, with arrival (car) the platform counting number zero clearing of whole waiting spaces.(step 62).

Like this, in the step 61 of next execution cycle, because judge that new study is not set permits with the formation of data, so, enter step 63.In step 63, judge whether to be provided with the actual measurement instruction that arrives the delay degree, because in step 62, be provided with actual measurement instruction,, judge that distributing car whether waiting space to be called out C has made and reply (waiting space that stays in or call out by waiting space C) so carry out step 64.

If the unsettled waiting space that fixes on is called out the waiting space of C and is stopped, then enter step 65, judge whether removing that waiting space calls out that car the distribution car of C responds car call at least after determining stop.

In for several times later execution cycle, if detecting the car that distributes outside the car just responds car call and determines to stop, then will carry out increment (step 66) with the position of this car and arrival (car) the platform number of the cooresponding waiting space of direction, enter step 67, judge and distribute the car position f of car whether to change.In addition, in step 65, not after car decision outside distributing car stops, just directly to enter step 67 if judge.

In for several times later execution cycle,, then enter step 68, judge at this time to have or not the advance notice error, and store as the part of m study with data from step 67 if detect the variation of car position f.This is original teacher's data, is expressed as the actual arrival delay degree TA(f of the waiting space of car position f).Arrive (car) platform counting number value representation and call out after C distributes, response car call and arrive the car platform number of the waiting space that distributes car position f in advance at waiting space.Therefore, arrive platform counting number value and be " 1 " when above, expression " is called out if record waiting space at this waiting space, and is distributed to this car, so probably have the advance notice error to take place ", therefore, reality arrival delay degree TA(f) be set at " 1 ".In addition, when arrival platform counting number value is " 0 ", set the actual delay degree TA(f that arrives) be " 0 ".

In the step 65 of for several times later execution cycle, if detect the stop decision of calling out the waiting space of C to waiting space, then enter step 69 and step 68 in the same manner at this moment actual arrival delay degree TA(c) store as the part of m study with data.Then, remove the actual actual measurement instruction that arrives the delay degree, simultaneously,, new study is set again forms permission (step 70) with data with the number m increment of study with data.

Like this, with carry out period that waiting space distributes consistently, produce input data and the output data relevant repeatedly with the car that is assigned with, and after this to distribute car reply waiting space call out car till the C by or pairing each actual advance notice of each waiting space of floor midway of stopping slip up (reality arrival delay degree), use data as study, and store.

Then, correcting device 10G utilizes the study data in the revision program (step 43) of Fig. 4, revise neural network 1 0DA.

Below, with reference to Fig. 7, illustrate in greater detail this corrective action.

At first, judged whether that this does the period (step 71) of network correction, if correction period, the step 72-78 below then carrying out.

When here, the study of storage at present being counted m and reaches more than S (for example, 500) with the group of data as network correction period.Study is counted S with the judgment standard of data can be according to the number of floor levels FL that platform number, building are set of elevator and waiting space calls or the like network size and setting at random.

In step 71, if judge that it is more than S that study is counted m with the group of data, then will learn to be initially set " 1 " (step 72) with the count number n of data, then, learn with taking out the actual delay degree TA(1 that arrives the data from n)-TA(N ₃), obtain value with the cooresponding node of these waiting spaces from following formula, that is, and teacher's data da(k) (k=1,2 ..., N ₃) (step 73):

da（k）＝TA（k）/NT _max…（7）

Then, will be from n study with the output valve ya3(1 of the output layer 10DA3 that takes out the data)-ya3(N ₃) and teacher's data da(1)-da(N ₃) squared difference, and, to k=1-N ₃Summation, thus obtain both error E a:

Ea＝∑[{da（k）-ya3（k）} ²]/2 …（8）

（k＝1-N ₃）

And then, utilize the error E a that obtains from (8) formula, as described below, revise the coefficient of weight wa2(j between interlayer 10DA2 and the output layer 10DA3, k) (j=1,2 ..., N ₂, k=1,2 ..., N ₃) (step 74).

At first,, k), put in order with aforementioned (1)-(5) formula to the error E a differential of (8) formula with wa2(j, coefficient of weight wa2(j, variable quantity △ wa2(j k) k) just is expressed as:

△wa2（j，k）＝-α｛

Ea/

wa2（j，k）｝

＝-α·δa2（k）·ya2（j） …（9）

Wherein, α is the parameter of expression pace of learning, is chosen as the arbitrary value in the 0-1 scope.In addition, in (9) formula,

δa2（k）＝{ya3（k）-da（k）}ya3（k）{1-ya3（k）}

Calculate coefficient of weight wa2(j like this, variable quantity △ wa2(j k), k) after, revise coefficient of weight wa2(j according to following formula (10), k).

wa2（j，k）←wa2（j，k）+△wa2（j，k） …（10）

Similarly,, revise the coefficient of weight wa1(i between input layer 10DA1 and the interlayer 10DA2 according to following (11) and (12) formula, j) (i=1,2 ..., N ₁, j=1,2 ..., N ₂) (step 75).

At first, obtain coefficient of weight wa1(i from following (11) formula, variable quantity △ wa1(i j), j):

△wa1（i，j）＝-α·δa1（j）·ya1（i） …（11）

In (11) formula, δ al(j) be following to k=1-N ₃The summation formula, be expressed as:

δa1（j）＝∑{δa2（k）·wa2（j，k）·ya2（j）×[1-ya2（j）]}

Utilization j), is weighted coefficient wa1(i, correction j) by the variable quantity △ wa1(i that (11) formula obtains as 12 formulas.

wa1（i，j）←wa1（i，j）+△wa1（i，j） …（12）

In the superincumbent step 74 and 75, have only the coefficient of weight relevant to obtain revising with the waiting space that has teacher's data.That is, as study is formed the explanation that program (Fig. 6) is done with data, only stored actual arrival delay degree for floor waiting space in the way between the waiting space of car position that is in the branch timing and waiting space calling C, as teacher's data, so, do not revise the coefficient of weight relevant with other waiting space.

Like this, after revising step 73-75 with data according to n study, study with the number n increment (step 76) of data, is carried out the processing of step 73-76 repeatedly, made correction (n 〉=m) with data until in step 77, judging with regard to global learning.

And then, after global learning having been done with data revise,, j) and wa2(j, k) be recorded among the prediction arrival delay degree arithmetical device 10D (step 78) the coefficient of weight wa1(i that finishes correction.

At this moment, in order to store up-to-date study data once more, the study data full scale clearance with using in revising is initially set " 1 " to the number m that learns with data.The network correction (study) of this neural network 1 0DA that just is through with.

Like this, produce study according to measured value and use data, rely on these study to revise the coefficient of weight wa1(i that prediction arrives delay degree arithmetical device 10D respectively with data, j) and wa2(j, k), therefore, even the flow of traffic in the building changes, also can be automatically corresponding with it.

In addition, because adopted using escalator number and descending stair number in 5 minutes of each waiting space of counting in the past, as the input data that embody traffic flow character, so, only compare as the input data conditions with car position, service direction, car load and this calling of replying, for the flow of traffic that changes constantly, the present invention can realize correct prediction computing more neatly.

In the above-described embodiments, the advance notice probability of failure is arrived the delay degree as prediction, but, also can arrive sequential prediction value or representative and arrive from the prediction that arrives arrival predictor delay time that car starts at first and arrive the delay degree as prediction delay time with car.So long as embody the index that car arrives delay degree, can arrive the delay degree as prediction.

For example, arrive with car under the situation of sequential prediction value as prediction arrival delay degree, predicting the output valve ya3(1 of the output layer 10DA3 that arrives delay degree arithmetic element 10DA at Fig. 3)-ya3(N ₃) in, can make the output valve ya3(1 of 1-Section 11 point)-ya3(11) correspond respectively to 1,2 ..., the 11 arrival orders of line direction waiting space upstairs, make the output valve ya3(12 of 12-22 node)-ya3(22) correspond respectively to 12,11 ..., the 2 arrival orders of line direction waiting space downstairs.The output valve ya3(k of each node k) (k=1,2 ..., N ₃) prediction that is transformed into waiting space k arrives delay degree T(k), be expressed as:

T（k）＝ya3（k）×NR _max…（13）

In this case, because the output valve ya3(k of node k) be normalized in the scope of 0-1, so, by multiply by maxim NR _Max(for example, carrying out the car platform number of group's management) is transformed into the form in the evaluation number computing that can be used for the waiting space call distribution.

Study when spending as prediction arrival delay to arrive the sequential prediction value forms as shown in Figure 8 with data.

The study of Fig. 8 with data formation program in, the step except that step 68' and 69' is identical with Fig. 6's.In addition, the arrival platform counting number value of each waiting space is illustrated in waiting space and calls out after the C distribution as hereinbefore, arrives the car platform number of the waiting space of this distribution car position f in advance because of car call.Therefore, in this case, in step 68' and 69', the value (arrival order) after adding " 1 " on the arrival platform counting number value is as original teacher's data TA(f) store.

Subsequent, according to the study data that form as shown in Figure 8, by the revision program correction coefficient of weight of Fig. 7.At this moment, abide by following formula:

da（k）＝TA（k）/NR _max…（14）

Carry out conversion to teacher's data.

Arrive as prediction under the situation of delay degree arriving the delay time predictor with car, can be at the output valve ya3(1 of output layer 10DA3)-ya3(N ₃) in (with reference to Fig. 3), make the output valve ya3(1 of 1-Section 11 point)-ya3(11) correspond respectively to 1,2 ..., 11 arrival delay times of line direction waiting space upstairs, make the output valve ya3(12 of 12-the 22nd node)-ya3(22) correspond respectively to 12,11 ..., 2 arrival delay times of line direction waiting space downstairs.The output valve ya3(k of each node k) prediction that is transformed into waiting space k by following formula arrives delay degree T(k):

T（k）＝ya3（k）×NT _max…（15）

Wherein, NT _MaxBe arrival peaked steady state value delay time that an expression can be expected, for example, be made as 100 seconds.

Like this, arrive as prediction under the situation of delays degree to arrive the delay time predictor, study can be carried out with Fig. 6 or Fig. 8 in the same manner with the formation of data, and measurement is from the delay time that the car of arrival is at first started at, with it as the study data.In this case, according to:

da（k）＝TA（k）/NT _max…（16）

Be transformed into teacher's data.

In above-mentioned each embodiment, the input data converting apparatus carries out conversion with car position, service direction, car load and this calling of replying as the input data, and still, the traffic state data that can be used as the use of input data is not limited to this.For example, can be used as the input data to car status (in the deceleration, in the action of opening the door, door opening, closing the door in the action, door is closed stand, walking is medium) waiting space call duration, car call time length, car load, the car platform number etc. that carries out group management.In addition, not only present traffic state data, traffic state data in the near future (experience of the experience of car movement and call answering state etc.) also is used as the input data, by this, just may carry out more accurate prediction and arrive the computing of delay degree.

In the distribution of carrying out the waiting space calling, study with data form device 10F distribute the cooresponding prediction of each waiting space of car arrive delays degree and the input data of this moment and with after this distribute car to reply waiting space call out till stop once or the waiting space that passes through be that the actual arrival delay degree of object stores, as one group of study data, but, form and be not limited to this period of learning with data.For example, can import when playing elapsed time in the data storage and exceed schedule time (for example, 1 minute) with the last time and form period as study with data, (for example, every 1 minute) study that also can specified period forms period with data.Because study under various conditions must be many more with data gathering, then condition for study just improves more, so, can pre-determine representative state, for example, this state can consider it is in car stops on the regulation floor, perhaps, car is in the specified states (in the deceleration, stop medium), or the like, when detecting this state, form the study data.

It is the actual arrival delay degree of object that study is only stored with the waiting space that once stopped till distributing car to call out to the waiting space of replying dispensing or pass through with data formation device 10F, as teacher's data, when being weighted the correction of coefficient with correcting device 10G, only revise the relevant coefficient of weight of teacher's data with storage, but the abstracting method of teacher's data is not limited to this.For example, also can store the actual arrival delay degree that the prediction relevant with whole waiting spaces arrives the delay degree and can measure in cage operation, only revise the coefficient of weight relevant with the waiting space that has teacher's data.Here, the so-called waiting space that can not measure actual arrival delay degree, for example under the situation of car floor place reverse directions in the way, be equivalent to waiting space in a counter-rotating floor distant place, and in the way, become on the floor under the sky car situation of (not distributing the car of calling out) at car, then be equivalent to the waiting space in a floor distant place that changes into the sky car and in the input data storage car position floor behind side waiting space (for example, upwards in the operational process, be in the waiting space of current position below).

In addition, prediction arrives delay degree arithmetical device 10D and revise coefficient of weight when the study of storage reaches defined amount with data bulk, and still, the coefficient of weight correction is not limited to this period.For example, also can the moment that is predetermined (for example, every 1 hour) utilize the study of storage thitherto to revise coefficient of weight with data, can also become idle in traffic, when the prediction arrival delay degree computing frequency that prediction arrival delay degree arithmetical device 10D carries out tails off, revise coefficient of weight.

Have again, also can repeatedly carry out the correction step (for example, carrying out 500 times) of coefficient of weight repeatedly, make the coefficient of weight convergence, to obtain desirable approximate output for 500 data.

As mentioned above, the present invention possesses: will include car position, service direction and this calling of replying are transformed into the form that can be used as the use of neural network input data at interior traffic state data input data converting apparatus; Include the input layer that is taken into the input data, be equivalent to predict the data that arrive delays degree be the output layer of output data and be in input layer and output layer between set the interlayer of coefficient of weight, and constitute the prediction arrival delay degree arithmetical device of neural network; Output data is transformed into the output data converting means of the form that can be used for regulation control purpose.The present invention is taken into neural network with traffic state data, the car prediction that calculates corresponding to waiting space arrives the delay degree, therefore, can obtain prediction by the computing that approaches actual arrival delay degree and arrive the delay degree, simultaneously, arrive the delay degree according to this correct prediction, the performance of group's management is improved.

In addition, the present invention also possesses: study forms device with data, when elevator work entered the period that is predetermined, the prediction of store predetermined car arrived the actual arrival delay degree of delay degree and the input data of this moment and regulation car, and they are exported with data as one group of study; Utilize study to revise the correcting device that prediction arrives the coefficient of weight of delay degree arithmetical device with data.Elevator control gear of the present invention is according to the result and traffic state data at that time and the measured data that calculate, automatically revise the coefficient of weight in the neural network, therefore, can be automatically corresponding to the variation of flow of traffic in the building under the actual conditions, and the precision of prediction that arrives the delay degree is higher.

Claims

1, a kind of elevator control gear, the time delay degree that its prediction calculates the elevator car arrives waiting space arrives the delay degree as prediction, utilizes described prediction to arrive the action that the delay degree is controlled described car, it is characterized in that it also comprises:

Position, service direction and this calling that responses that will include described car are transformed at interior traffic state data and can be used as the input data converting apparatus that neural network is imported the form of data,

The prediction that constitutes described neural network arrives delay degree arithmetical device, include the input layer that is taken into described input data, to be equivalent to data that described prediction arrives the delay degree as the output layer of output data be in the interlayer of having set coefficient of weight between described input layer and the output layer

Described output data is transformed into the output data converting means that to control the form of using in the purpose action in regulation.

2, elevator control gear as claimed in claim 1 is characterized in that, it also comprises:

Study forms device with data, when elevator enters after date when being predetermined at work, the prediction of store predetermined waiting space arrives delay degree and input data at that time, simultaneously, car stayed in or during by described regulation waiting space the arrival delay degree of described car store as the actual delay degree that arrives, the described input data that store, described prediction arrival delay degree and the described actual delay degree that arrives are exported with data as one group of study

Utilize described study to arrive the correcting device of the coefficient of weight of delay degree arithmetical device with the data correction prediction.