JP7327895B2 - cargo handling system - Google Patents

Description

The present invention relates to cargo handling systems.

As described in Patent Document 1, a battery-powered unmanned forklift that operates autonomously is known. This unmanned forklift performs cargo handling work inside and outside the facility. Loading and unloading operations within the facility include placing and unloading items on racks. An unmanned forklift usually has to turn once when entering the aisle between shelves and placing a load on the shelf. Forklift power is consumed at each turn. On the other hand, as described in Patent Document 2, there is known a three-way forklift that can change only the direction of the fork by 90 degrees to the left and right without changing the direction itself after entering the aisle. This forklift can perform cargo handling work without turning from the direction of travel to the direction of the shelf, so it has good power efficiency. Among these three-way forklifts, forklifts in which the drive system for fork shift (left and right slide) and rotation (turning) are electrically driven are particularly excellent in power efficiency.

Also, as in Patent Document 3, a wireless power feeding system is known that includes a power transmitting device that transmits microwave power to an automatic guided vehicle having a power receiving unit. According to this wireless power supply system, the unmanned guided vehicle can delay the timing of battery depletion by supplying power during cargo handling and movement. Therefore, according to this wireless power supply system, the operating rate of the forklift can be increased.

By the way, in the case of a cargo handling system equipped with a plurality of forklifts, forklifts with different power efficiencies may be included, like the above-mentioned forklifts. In that case, it is preferable to allocate cargo handling work and supply power in consideration of the power efficiency of each forklift. However, there has never been such a cargo handling system. For this reason, conventional cargo handling systems may not achieve efficiency in cargo handling work for the entire cargo handling vehicle.

JP 2021-59435 A Japanese Patent Application Laid-Open No. 2020-52630 JP 2017-055621 A

Accordingly, the problem to be solved by the present invention is to provide a cargo handling system capable of wirelessly supplying power to an unmanned forklift and further improving the efficiency of cargo handling work in the entire cargo handling vehicle by considering the power efficiency of the unmanned forklift. is to provide

In order to solve the above problems, the cargo handling system according to the present invention includes:
A cargo handling system comprising a plurality of unmanned forklifts having power receiving units, a power transmission device for wirelessly transmitting power to the unmanned forklifts for charging, and a management device,
The management device
a first neural network pre-learned the correlation between the identifier of the unmanned forklift, the cargo handling operation and the weight of the cargo in each cargo handling work, and the teacher data having the cargo handling power consumption as the output data;
a power consumption prediction unit;
a cargo handling vehicle schedule output unit;
an inter-loading power consumption calculator that calculates traveling power consumption of an unmanned forklift between cargo handling tasks based on the coordinates of the load placement position in the previous cargo handling task and the coordinates of the load picking position in the next cargo handling task;
a second neural network;
a cargo handling addition section;
The power consumption prediction unit inputs the identifier of the unmanned forklift, the cargo handling operation, and the weight of the cargo to the first neural network, and outputs predicted cargo handling power consumption of each unmanned forklift in each cargo handling work,
The second neural network includes the battery capacity of the battery mounted on each unmanned forklift, the cargo handling prediction power consumption of each unmanned forklift in each cargo handling work, the coordinates of the cargo pick-up position and cargo placement position of each cargo handling work, and the coordinates of each cargo handling work. Based on the traveling power consumption of the unmanned forklift during work, an unmanned forklift with a higher power efficiency rank outputs a cargo handling schedule that can handle more cargo than an unmanned forklift with a lower power efficiency rank. have been taught,
The cargo handling vehicle schedule output unit inputs the battery capacity of the battery mounted on each unmanned forklift and the cargo handling predicted power consumption of each unmanned forklift in each cargo handling work to a second neural network to output a cargo handling schedule,
Furthermore, when the power-efficient unmanned forklift can be made to carry out further cargo handling work by charging by wireless power transmission, the second neural network determines the power-efficient remaining battery level of the power-efficient unmanned forklift and the power-efficient power efficiency in each cargo handling task. An unmanned forklift with good power efficiency based on the predicted cargo handling power consumption of the unmanned forklift, the coordinates of the loading and unloading positions of each cargo handling work, and the running power consumption of the unmanned forklift with good power efficiency between each cargo handling work. is trained to output additional cargo handling schedules that can do more cargo handling work,
The cargo handling work addition unit inputs the remaining battery power of the unmanned forklift with good power efficiency and the predicted cargo handling power consumption of the unmanned forklift in each cargo handling work to the second neural network, and performs additional cargo handling work to the second neural network. It is characterized by outputting.

The cargo handling system preferably includes:
A power transmission device preferentially transmits power to an unmanned forklift having a higher power efficiency rank than to an unmanned forklift having a lower power efficiency rank.

For example, the cargo handling system
Multiple unmanned forklifts include 3-way forklifts and 1-way forklifts, where 3-way forklifts are more power efficient than 1-way forklifts.

The cargo handling system preferably includes:
A second neural network is trained by reinforcement learning.

The cargo handling system preferably includes:
further comprising a reward generation unit that generates a reward to be given to the second neural network;
The more unmanned forklifts with good power efficiency are assigned to the cargo handling schedule output by the second neural network, the more cargo handling work is assigned by the reward generation unit until power is supplied at the power supply location, the more reward is given. to generate

The cargo handling system preferably includes:
The remuneration generation unit generates more remuneration for the cargo handling schedule output by the second neural network as the power consumption during cargo handling of each unmanned forklift is smaller.

The cargo handling system preferably includes:
The remuneration generation unit determines whether one or both of the loading and unloading positions of the loading and unloading work allocated to the unmanned forklift with good power efficiency is within the distance of the power transmission unit for the cargo handling schedule output by the second neural network. The closer you are, the more rewards you generate.

The cargo handling system preferably includes:
A power transmission device is provided on a shelf having a pickup position and a storage position.

The battery management system of the present invention can improve the efficiency of cargo handling work in the entire cargo handling vehicle by considering the power efficiency of the unmanned forklift.

It is a top view showing the whole cargo handling system concerning one embodiment of the present invention. FIG. 2 is a side view showing the configuration of the unmanned forklift shown in FIG. 1; FIG. 3 is a side view showing the configuration of another unmanned forklift shown in FIG. 1; FIG. 2 is a block of a management unit shown in FIG. 1; FIG. 5 is a diagram showing functions of a cargo handling operation output unit shown in FIG. 4; FIG. FIG. 5 shows the first neural network shown in FIG. 4, A is a diagram showing machine learning, and B is a diagram showing a function of outputting predicted cargo handling power. FIG. 5 illustrates the operation of the second neural network shown in FIG. 4; FIG. 5 illustrates another operation of the second neural network shown in FIG. 4; FIG. 2 is a flow chart showing the operation of the cargo handling system shown in FIG. 1;

An embodiment of a cargo handling system according to the present invention will be described below with reference to the accompanying drawings. In the drawings, a double-headed arrow X indicates the front-rear direction, and a double-headed arrow Z indicates the up-down direction.

As shown in FIG. 1, the cargo handling system S includes a battery-powered unmanned forklift 1 (hereinafter simply referred to as a "forklift") 1, a power transmission device 3, a management device 4, a shelf 6, and a power supply device 7. I have. The forklift 1 is composed of four forklifts 1 a , 1 b 1 , 1 b 2 , 1 b 3 , and the power transmission device 3 is composed of four power transmission devices 3 . Also, the shelf 6 is composed of four shelves 61 , 62 , 63 and 64 . In the following description, the forklift as a whole may be referred to as "forklift 1", and the rack as a whole may be referred to as "shelf 6". A plurality of forklifts 1 and management devices 4 are configured to be able to communicate with each other. The plurality of forklifts 1 according to the present embodiment includes two types of forklifts 1a and 1b, but this is merely an example. may be included. Each component of the cargo handling system S will be described below.

<Forklift>
As shown in FIG. 2, the forklift 1a includes front and rear wheels 10, a vehicle body 11, a laser scanner 12 disposed above the vehicle body 11, a battery 20, and a remaining battery level detection unit 13. . The laser scanner 12 irradiates a laser in the horizontal direction while rotating, and detects the current position of the forklift 1a by receiving reflected light from a reflector arranged at a predetermined location in the facility.

The battery remaining amount detection unit 13 detects the remaining amount of the battery 20 . The detected remaining amount of the battery 20 is transmitted to the management device 4 .

The forklift 1a further includes a pair of left and right masts 14 extending vertically, a lift bracket 15 connected to the left and right masts 14 and vertically movable, and a pair of upper and lower side rails 16 provided on the lift bracket 15 and extending in the horizontal direction. , and a shift carriage 17 that shifts left and right along side rails 16 . A finger bar 18 is provided on the shift carriage 17 , and a pair of left and right forks 19 are provided on the finger bar 18 . The finger bar 18 is rotatably mounted on the shift carriage 17 . With these configurations, the forklift 1a can change the orientation of the fork 19 in three directions without changing the orientation of the vehicle body 11. - 特許庁

The forklift 1a further includes a power receiving section 21 (see FIG. 4). The power reception unit 21 has a pilot signal transmission unit (not shown) and a rectenna (not shown), and is electrically connected to the battery 20 . The pilot signal transmission unit transmits a pilot signal to the power transmission device 3 . When the rectenna receives the microwaves, the power receiving unit 21 converts the microwaves into charging power and supplies the charging power to the battery 20 to charge the battery 20 . The rectenna may be arranged on the outer surface of the vehicle body 11, for example. The pilot signal transmission unit is configured to transmit a pilot signal to the power transmission device 3 not only during cargo handling work but also during non-cargo handling work.

As shown in FIG. 3, the forklift 1b includes front and rear wheels 10, a vehicle body 11, and a laser scanner 12 disposed above the vehicle body 11. As described above, the laser scanner 12 detects the current position. to get The forklift 1b also includes a battery remaining amount detection unit 13 that detects the remaining amount of the battery 20, like the forklift 1a. Further, the forklift 1b includes a pair of left and right masts 14 extending vertically, a lift bracket 15 connected to the masts 14 and vertically movable, and a pair of left and right forks 19 connected to the lift bracket 15. The forklift 1b changes the direction of the forks 19 by changing the direction of the vehicle body 11. - 特許庁

The forklift 1b further includes a power receiving unit 21 (see FIG. 4), like the forklift 1a. The configuration of the power receiving unit 21 of the forklift 1b is the same as the configuration of the power receiving unit 21 of the forklift 1a, so description thereof is omitted.

The forklift 1 is configured to autonomously transport a load W from a loading position to a loading position. The operation of the forklift 1 to transport the load W from the loading position to the loading position is hereinafter referred to as "loading work". Upon receiving a return command from the management device 4, the forklift 1 returns to the power supply location CP and is supplied with power by the power supply device 7, as will be described later.

The three-way forklift 1a in this embodiment can perform cargo handling work by changing only the orientation of the forks 19 without changing the orientation of the vehicle body 11, so the power efficiency is higher than that of the one-way forklift 1b. is good. Also, the forklift 1b is not the same in power efficiency due to the model, model year, and deterioration over time, even if it is the same one-way forklift 1b. In this embodiment, the power efficiency of the one-way forklift 1b is assumed to be higher in the order of 1b1, 1b2, and 1b3.

<Power transmission device>
Power transmission device 3 is arranged on the upper surface of shelf 61 and shelf 62 , and has pilot signal receiving section 30 and power transmission section 31 . Upon receiving the pilot signal, the power transmission device 3 estimates the position of the forklift 1 and transmits microwaves to the position of the forklift 1 by the power transmission unit 31 . As a result, the forklift 1 is charged with this transmitted power during cargo handling work and non-cargo handling work, thereby reducing the decrease in remaining battery capacity and delaying the return timing. In this embodiment, the retrodirective method is used as the wireless power transmission method, but this is merely an example, and the wireless power transmission method according to the present invention is not limited to this. Since the power transmission device 3 is arranged on the upper surface of the shelf 61 and the shelf 62 , the forklift 1 passing through the aisle between the shelf 61 and the shelf 62 can concentrate and wirelessly transmit power. Further, the power transmission device 3 may determine whether or not to perform wireless power transmission according to the distance from the forklift 1 .

The power transmission device 3 is configured to wirelessly transmit power to the forklift 1 with high power efficiency. In this embodiment, the power transmission device 3 wirelessly transmits power to the forklift 1a over the forklift 1b. Therefore, upon receiving a pilot signal from the forklift 1a during wireless power transmission to the forklift 1b, the power transmission device 3 stops wireless power transmission to the forklift 1b and starts wireless power transmission to the forklift 1a.

<Management device>
As shown in FIG. 4, the management device 4 includes a facility storage unit 40, a battery information storage unit 41, a traveling power consumption storage unit 42, a cargo handling storage unit 43, a cargo handling operation output unit 46, and a power consumption calculation unit. 47, a first neural network 48, a power consumption prediction unit 49, a power consumption calculation unit 51, a second neural network 53, a reward generation unit 54, a cargo handling vehicle schedule output unit 55, and a cargo handling command unit. 56 , a cargo handling work addition unit 57 , and a return command unit 58 .

The facility storage unit 40 stores the coordinates of the shelf 6, passageway, and power supply point CP in the facility related to the cargo handling work.

The battery information storage unit 41 stores information of the battery 20 including the battery capacity for each battery 20 mounted on the forklift 1 .

The travel power consumption storage unit 42 stores the power consumption per travel distance of each forklift 1 during non-cargo handling. Traveling during non-loading means traveling when the forklift 1 does not load the load W. As shown in FIG. That is, when heading from the loading position of the previous cargo handling work to the loading position of the next cargo handling work, when heading from the power supply position CP to the next loading position, and when heading from the loading position to the power feeding position CP. Traveling during non-cargo handling. The travel power consumption per travel distance during non-cargo handling may be obtained from the specifications of each forklift 1, or may be obtained from the actual power consumption of the forklift 1 during travel. Note that this running power consumption is the power consumption when wireless power transmission is not performed. In addition, hereinafter, from the loading position in the previous cargo handling work to the loading position in the next cargo handling work, from the power supply position CP to the next loading position, and from the loading position to the power supply position CP, or any of them This is sometimes referred to as "between cargo handling operations".

The cargo handling storage unit 43 stores cargo handling information. The cargo handling information includes the coordinates of the cargo pick-up position and cargo placement position for each cargo handling work and the weight of the cargo W. The coordinates of the loading position and the loading position include horizontal coordinates and height coordinates.

As shown in FIG. 5, based on the type of forklift truck 1, the information in the facility, the pick-up position, and the coordinates of the pick-up position, the cargo handling operation output unit 46 outputs the traveling distance of each forklift 1 in each cargo handling work, Fork lifting height, fork lifting height at the time of loading, and cargo handling operation including the number of turns of the vehicle body 11 are output. Note that the cargo handling operation output unit 46 outputs not only the cargo handling operations of the forklift 1 in the "past cargo handling operations" but also the cargo handling operations of the forklift 1 in the "previous cargo handling operations" that have not yet been performed.

"Travel distance of each forklift 1" is the travel distance from the pick-up position to the dump position. The cargo handling operation output unit 46 acquires the route from the facility information and the coordinates of the pickup position and the cargo placement position, and then outputs the distance traveled during cargo handling based on the distance of the route.

The "fork lifting height during unloading and unloading" refers to the lift of the forks 19 during unloading and unloading operations. The cargo handling operation output unit 46 outputs the lifting height of the forks 19 of the forklift 1 during cargo pickup and cargo placement based on the height coordinates of the cargo pickup position and cargo placement position and the specifications of the forklift 1 .

The "number of turns of the vehicle body 11" is the number of turns of the vehicle body 11 of the forklift 1 from the pickup position to the loading position. The cargo handling operation output unit 46 outputs how many times the forklift 1 needs to change direction from the cargo pickup position to the cargo placement position based on the type of the forklift 1, the information in the facility, and the coordinates of the cargo pickup position and the cargo placement position. Naturally, the number of times of direction change of the forklift 1a is output less than the number of times of direction change of the forklift 1b.

The power consumption calculating unit 47 receives the remaining amount of the battery 20 of the forklift 1 from the remaining battery amount detecting unit 13 through communication. The power consumption calculator 47 calculates the cargo handling power consumption of the forklift 1 in each cargo handling work based on the remaining capacity of the battery 20 at the start of cargo handling work and the remaining capacity of the battery 20 at the completion of the cargo handling work. Note that this cargo handling power consumption is the power consumption when wireless power transmission is not performed.

As shown in FIG. 6A, the first neural network 48 uses the identifier of the forklift 1, the cargo handling operation of each forklift 1, and the weight of the cargo W in each cargo handling work performed in the past as input data, and outputs the cargo handling power consumption. Supervised learning by deep learning is performed in advance by teacher data used as data. Thereby, the first neural network 48 has previously learned the correlation between each forklift 1, the cargo handling operation of each forklift 1, the weight of the cargo W, and cargo handling power consumption.

As shown in FIG. 6B , the power consumption prediction unit 49 uses the identifier of the forklift 1, the cargo handling operation of each forklift 1, and the cargo handling operation of each forklift 1 in each previous cargo handling work in order to output the cargo handling prediction power consumption of each forklift 1 in each cargo handling work. and the weight of the cargo W are input to the first neural network 48 to output the predicted cargo handling power consumption of each forklift 1 in the previous cargo handling work. The first neural network 48 outputs cargo handling prediction power consumption based on the correlation obtained by machine learning.

The inter-cargo handling power consumption calculation unit 51 calculates the traveling of the forklift 1 between each cargo handling work based on the coordinates of the cargo placement position in the previous cargo handling work, the coordinates of the cargo pick-up position in the next cargo handling work, and the facility information. Calculate the power consumption RCP. For example, the inter-cargo handling power consumption calculation unit 51 calculates the traveling route between cargo handling operations from the coordinates of the cargo placement position in the previous cargo handling work, the coordinates of the cargo pickup position in the next cargo handling work, and the facility information. , the travel distance of the travel route (that is, the travel distance between cargo handling operations) is calculated. Next, the power consumption calculation unit 51 calculates the traveling power consumption RCP of the forklift 1 during each cargo handling operation by adding the power consumption per travel distance of each forklift 1 during non-cargo handling to the travel distance between cargo handling operations. do.

The second neural network 53 includes the battery capacity of the battery 20 mounted on each forklift 1, the cargo handling predicted power consumption of each forklift 1 in each cargo handling work, the coordinates of the cargo pick-up position and cargo placement position of each cargo handling work, Based on the traveling power consumption of the forklift 1 between each cargo handling work, a cargo handling schedule is output in which the forklift 1 with the higher power efficiency can perform more cargo handling work than the forklift 1 with the lower power efficiency. Reinforcement learning is performed by deep learning so that In addition, although the second neural network 53 employs reinforcement learning in the present embodiment, the machine learning of the second neural network 53 is such that the forklift 1 with higher power efficiency ranks lower in power efficiency. If machine learning is performed to output a cargo handling schedule that allows more cargo handling work than the forklift 1, the invention is not limited to this.

The reward generator 54 generates rewards to be assigned to the second neural network 53 . In the cargo handling schedule output by the second neural network 53, the remuneration generation unit 54 increases the remuneration as the forklift 1 with good power efficiency is assigned more cargo handling work until power supply at the power supply location CP. to generate Furthermore, the remuneration generation unit 54 generates more remuneration for the cargo handling schedule output by the second neural network 53 as the traveling distance between cargo handling operations becomes shorter. Further, the remuneration generation unit 54 generates a remuneration as the pickup position and/or both of the pickup position and the pickup position are closer to the power transmission device 3 .

The operation of the second neural network 53 will be described with reference to FIG. FIG. 7 visualizes the operation of the second neural network 53. FIG. "jn" in FIG. 7 indicates the power consumption of each cargo handling work, and the area of the square surrounding "jn" visually indicates the power consumption. Since the heights of the squares are all the same, the difference in the width of the squares indicates the difference in the power consumption of each forklift 1 in each cargo handling operation. That is, the shorter the width of the square portion, the less the power consumption of the forklift 1 in the cargo handling work. For example, as indicated by the reference numbers of the forklifts 1a, 1b1, 1b2, and 1b3, in the case of cargo handling work j3, the forklift 1a consumes less power than the other forklifts 1b1, 1b2, and 1b3. showing.

Each rightward arrow in the upper part of FIG. 7 indicates the battery capacity of the battery 20 of each forklift 1 . "CC" indicates the full battery capacity of each battery 20, and "CT" indicates power supply at the power supply location CP. A difference in the length of each CC indicates a difference in battery capacity. As shown in FIG. 7, the forklift 1a is allocated with the battery 20 having a large battery capacity, so the number of times of power supply at the power supply location CP is smaller than that of the other forklifts 1b1, 1b2, and 1b3. As a result, the forklift 1a has a longer operating time than the other forklifts 1b1, 1b2, and 1b3, and as a result, can handle more cargo.

The second neural network 53 arranges cargo handling operations jn (square portions) on the full battery capacity CC of the battery 20 . Between the arranged cargo handling operations jn and the cargo handling operations jn, the travel power consumption RCP between the cargo handling operations is added and arranged. In addition, the power consumption during cargo handling RCP is added and arranged before the first cargo handling job jn and after the cargo handling job jn immediately before the power supply at the power supply place CP.

It should be noted that the second neural network 53 cannot arrange the same cargo handling work jn in the upper row in FIG. In addition, the second neural network 53 must arrange each cargo handling task jn such that each cargo handling task jn and each cargo handling power consumption RCP are within the full battery capacity CC. The second neural network 53 sequentially performs this work under these environments. Thereby, each cargo handling work jn is assigned to each forklift 1a, 1b1, 1b2, 1b3, and the order of cargo handling work for each forklift 1a, 1b1, 1b2, 1b3 is also determined. As a result, the second neural network 53 can output the cargo handling schedule of the entire forklift 1 .

The remuneration generation unit 54 generates a remuneration for the result of the second neural network 53 arranging the cargo handling work jn. That is, as described above, the remuneration generator 54 generates more remuneration as the forklift 1 with good power efficiency is assigned more cargo handling work until power supply at the power supply location CP. The second neural network 53 repeats the above work so as to maximize the reward. The second neural network 53 refers to the coordinates of the pickup position and the loading position so that the reward can be maximized, and the distance between the loading and unloading operations can be shortened. Learning is repeated many times so that one or both of the loading positions are close to the power transmission device 3.例文帳に追加The reward generation unit 54 may use the reward for shortening the travel distance between the cargo handling operations and for the cargo pickup position or the cargo storage position being closer to the power transmission device 3 as the “Q value” in Q learning of reinforcement learning. . In this case, even if the cargo pickup position or the cargo storage position is close to the power transmission device 3, if the distance between cargo handling operations is long, the efficiency of the cargo handling as a whole will be degraded. Therefore, when the distance between cargo handling operations is longer than a certain amount, the remuneration generation unit 54 may generate a negative remuneration according to the distance.

The second neural network 53 uses the reinforcement learning to determine the capacity of the battery 20 mounted on each of the forklifts 1a, 1b1, 1b2, and 1b3 and the cargo handling predicted power consumption of the forklifts 1a, 1b1, 1b2, and 1b3 in each cargo handling work. When input, the optimum cargo handling schedule of the forklift 1 can be output. The reinforcement learning of the second neural network 53 in this embodiment is merely an example, and the reinforcement learning of the second neural network 53 in the present invention is not limited to this.

The cargo handling vehicle schedule output unit 55 outputs the capacity of the battery 20 mounted on each of the forklifts 1a, 1b1, 1b2, and 1b3 and the estimated cargo handling power consumption of each of the forklifts 1a, 1b1, 1b2, and 1b3 in each cargo handling work to a second It is input to the neural network 53 and causes the second neural network 53 to output the cargo handling schedule.

The cargo handling command unit 56 commands the cargo handling work to each forklift 1 based on the cargo handling schedule output by the cargo handling vehicle schedule output unit 55 . The forklift 1 performs cargo handling work according to the received command.

The second neural network 53 further uses deep learning so that when the remaining battery level of the forklift 1a with good power efficiency increases due to wireless power transmission by the power transmission device 3, cargo handling work can be added based on the remaining battery level. Reinforcement learning is done.

A specific example will be described with reference to FIG. FIG. 8 shows that the remaining battery amount AP is added by charging by wireless power transmission from the start of cargo handling. The current battery remaining amount BR2 is obtained by adding the remaining battery amount AP and the remaining battery amount BR1 before charging by wireless power transmission. As shown in FIG. 8, when the second neural network 53 can allocate any of the unallocated cargo handling operations jn in the lower part of FIG. 8 to the remaining battery level BR2, The cargo handling work jn is additionally arranged. In addition, the power consumption RCP during cargo handling is added and arranged before and after the arranged cargo handling work jn. "When the cargo handling work jn can be distributed" is determined including the addition of the inter-work power consumption RCP.

The remuneration generation unit 54 further generates a remuneration for the result of the second neural network 53 additionally placing the cargo handling work jn. The power consumption of the arranged cargo handling work jn should be small so that the forklift 1a can perform more cargo handling work. Also, when the loading position of the cargo handling job jn to be arranged is closer to the loading position of the preceding cargo handling job jn, the travel time and power consumption between the cargo handling jobs of the forklift 1a are reduced. Also, the closer the pickup position or the placement position of the arranged cargo handling work jn is to the power transmission device 3, the easier the wireless power transmission. Therefore, the remuneration generating unit 54 determines whether the cargo handling work jn with less predicted cargo handling power consumption is arranged, or the pickup position of the arranged cargo handling work jn and the pickup position of the preceding cargo handling work (j18 in FIG. 8). If the distance to the position is close, or if either or both of the pickup position and the pickup position of the arranged cargo handling work jn are close to the power transmission device 3, a reward is generated according to the degree.

The second neural network 53 repeats the above work so as to maximize the reward. Note that, in the same manner as described above, the reward generating unit 54 sets the reward for shortening the travel distance between cargo handling operations and for the cargo pickup position or the cargo storage position being closer to the power transmission device 3 in the Q learning of reinforcement learning. Q value”. Through these reinforcement learning, the second neural network 53 can output the optimum additional cargo handling work of the forklift 1a when the remaining battery capacity of the forklift 1a and the predicted cargo handling power consumption of the forklift 1a in each cargo handling work are input. become.

Each time the forklift 1a completes each scheduled cargo handling task, the cargo handling task adding unit 57 adds the remaining battery level (BR2 in FIG. 8) of the forklift 1a and the predicted cargo handling power consumption of the forklift 1a in each cargo handling task. Input to the second neural network 53 and cause the second neural network 53 to output additional cargo handling work (scheduled cargo handling).

Based on the additional cargo handling schedule output by the cargo handling work addition unit 57, the cargo handling command unit 56 issues an additional cargo handling command to the forklift 1a.

The return command unit 58 commands the forklift 1a to return to the power supply location CP when the forklift 1a has completed all of the scheduled cargo handling operations and additional cargo handling operations cannot be performed.

A brief description of the operation of the management unit is as follows with reference to FIG.
(1) The cargo handling vehicle schedule output unit 55 causes the second neural network 53 to output the cargo handling schedule (S91 in FIG. 9).
(2) Each forklift 1a, 1b1, 1b2, 1b3 starts cargo handling work (S92 in FIG. 9).
(3) When each cargo handling schedule of the forklift 1a related to the cargo handling schedule is completed (S93 in FIG. 9),
(4) The cargo handling work adding unit 57 inputs the remaining battery capacity of the forklift 1a and the predicted cargo handling power consumption of the forklift 1a in each cargo handling work to the second neural network 53 (S94 in FIG. 9).
(5) The second neural network 53 outputs additional cargo handling work when cargo handling work can be added (Yes in S95 of FIG. 9). The cargo handling command unit 56 additionally commands the output cargo handling work to the forklift 1a (S96 in FIG. 9),
(6) The forklift 1a continues cargo handling work (S97 in FIG. 9).
(7) Even when cargo handling work cannot be added (No in S95 of FIG. 9) and all cargo handling work related to scheduled cargo handling has not been completed (No in S98 of FIG. 9), the forklift 1a , the cargo handling work is continued (S97 in FIG. 9).
(8) If the cargo handling work cannot be added (No in S95 of FIG. 9) and all the cargo handling work related to the cargo handling schedule is completed (Yes in S98 of FIG. 9), the return command unit 58 issues a return command. is applied to the forklift 1a, and the forklift 1a returns to the power supply point CP (S99 in FIG. 9).

As a result, the forklift 1 is not returned according to the remaining amount of the battery 20, but is returned as planned for cargo handling. Therefore, the management device 4 prevents the forklift 1 from being forced to return due to power shortage on the way to the next cargo handling position, and returns the forklift 1 at an appropriate timing. can be raised.

Moreover, with the above configuration, the management device 4 can increase the power efficiency of the entire cargo handling system S by allowing the forklift 1a with good power efficiency to perform more cargo handling work. In addition, the management device 4 preferentially wirelessly transmits power to the forklift 1a with good power efficiency, so that the forklift 1 with good power efficiency can perform more cargo handling work, so the power efficiency of the entire cargo handling system S is further improved. can be raised. Therefore, the power efficiency of the entire cargo handling system S is improved, and the number of power supply times at the power supply point CP in the entire forklift 1 is reduced, so the efficiency of the cargo handling work of the entire cargo handling system S is realized.

Although the cargo handling system S according to one embodiment of the present invention has been described above, the cargo handling system according to the present invention is not limited to the above embodiment. The cargo handling system according to the present invention may be implemented in combination with known techniques.

S Cargo handling system W Cargo CP Power supply location 1a, 1b Battery-powered unmanned forklift 10 Wheel 11 Car body 12 Laser scanner 13 Battery level detector 14 Mast 15 Lift bracket 16 Side rail 17 Shift carriage 18 Finger bar 19 Fork 20 Battery 21 Power receiving unit 3 Power transmission device 30 Pilot signal reception unit 31 Power transmission unit 4 Management device 40 Facility storage unit 41 Battery information storage unit 42 Running power consumption storage unit 43 Cargo handling storage unit 46 Cargo handling operation output unit 47 Power consumption calculation unit 48 First neural network 49 Power consumption Prediction unit 51 Power consumption calculation unit 53 Second neural network 54 Remuneration generation unit 55 Cargo handling vehicle schedule output unit 56 Cargo handling command unit 57 Cargo handling work addition unit 58 Return command unit 6, 61, 62, 63, 64 Shelf 7 Power supply device

Claims (8)

A cargo handling system comprising a plurality of unmanned forklifts having power receiving units, a power transmission device for wirelessly transmitting power to the unmanned forklifts for charging, and a management device,
The management device
a first neural network pre-learned the correlation between the identifier of the unmanned forklift, the cargo handling operation and the weight of the cargo in each cargo handling work, and the teacher data having the cargo handling power consumption as the output data;
a power consumption prediction unit;
a cargo handling vehicle schedule output unit;
an inter-loading power consumption calculation unit that calculates the running power consumption of the unmanned forklift during each load-handling task based on the coordinates of the load-placement position in the previous load-handling task and the coordinates of the load-taking position in the next load-handling task;
a second neural network;
a cargo handling addition section;
The power consumption prediction unit inputs the identifier of the unmanned forklift, the cargo handling operation, and the weight of the cargo to the first neural network, outputs predicted cargo handling power consumption of each unmanned forklift in each cargo handling work, The second neural network includes the battery capacity of the battery mounted on each of the unmanned forklifts, the predicted cargo handling power consumption of each of the unmanned forklifts in each cargo handling work, the cargo pickup position and the cargo placement position of each cargo handling work. and the traveling power consumption of the unmanned forklift during each cargo handling operation, the unmanned forklift with higher power efficiency rank performs more cargo handling operations than the unmanned forklift with lower power efficiency rank. It is learned to output the cargo handling schedule that can be done,
The cargo handling vehicle schedule output unit inputs the battery capacity of the battery mounted on each of the unmanned forklifts and the predicted cargo handling power consumption of each of the unmanned forklifts in each cargo handling work into the second neural network to input the cargo handling schedule. and
Furthermore, when the power-efficient unmanned forklift can be caused to perform further cargo handling work by charging through the wireless power transmission, the second neural network determines the remaining battery level of the power-efficient unmanned forklift and the The predicted cargo handling power consumption of the power efficient unmanned forklift, the coordinates of the cargo pickup position and the cargo placement position for each cargo handling work, and the traveling power consumption of the power efficient unmanned forklift between cargo handling tasks. is trained to output additional cargo handling schedules that allow the power efficient unmanned forklift to do more cargo handling work based on
The cargo handling work addition unit inputs the remaining battery capacity of the power efficient unmanned forklift and the predicted cargo handling power consumption of the unmanned forklift in each cargo handling work to the second neural network, and adds them to the second neural network. A cargo handling system characterized by outputting cargo handling work. 2. The cargo handling system according to claim 1, wherein the power transmission device preferentially transmits power to the unmanned forklift with higher power efficiency than to the unmanned forklift with lower power efficiency. 3. The plurality of unmanned forklifts of claim 1 or 2, wherein the plurality of unmanned forklifts includes a three-way forklift and a one-way forklift, the three-way forklift being more power efficient than the one-way forklift. cargo handling system. The cargo handling system according to any one of claims 1 to 3, wherein the second neural network is trained by reinforcement learning. Further comprising a reward generation unit that generates a reward to be given to the second neural network,
The remuneration generation unit is configured such that the more cargo handling work that the unmanned forklift with good power efficiency is assigned to the cargo handling schedule output by the second neural network, the more cargo handling work is assigned until the power supply at the power supply location. 5. The cargo handling system of claim 4, which generates more rewards. 3. The remuneration generation unit generates more remuneration for the cargo handling schedule output by the second neural network as the power consumption during cargo handling of each of the unmanned forklift trucks decreases. 5. The cargo handling system according to 5. The remuneration generating unit selects one or both of the loading position and the loading position of cargo handling work allocated to the unmanned forklift with good power efficiency for the cargo handling schedule output by the second neural network. 7. A cargo handling system according to claim 5 or 6, characterized in that the closer to the distance of the transmission unit, the more rewards are generated. The cargo handling system according to any one of claims 1 to 7, wherein the power transmission device is provided on a shelf having the cargo pickup position and the cargo storage position.

Images