Algorithm Presentation Device

Abstract

To easily select an algorithm suitable for a required specification.SOLUTION: A method of training a model for presenting a suitable algorithm from a plurality of algorithms relating to a transport system which transports packages using a plurality of moving objects, comprises the steps of: acquiring output values obtained by inputting condition values for indicating at least one of a configuration and operation of the transportation system to the plurality of algorithms; and using the condition values as input data, and causing the model to learn the output values linked with identifiers for identifying the algorithms as correct values.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a model training method for presenting, by artificial intelligence, an algorithm for a conveying system that conveys a load using a plurality of mobile bodies, and to an algorithm presenting device.

Background Art

[0002] Conventionally, when constructing a conveying system that conveys a load using a plurality of mobile bodies, for example, as described in Patent Document 1, a layout created by inputting requirement specifications through a predetermined simulation or the like may be used.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, requirement specifications related to the configuration of the conveying system, such as the layout (size and shape of the space, etc.) where the conveying system is installed, the number of mobile bodies used, and the mechanical characteristics of the mobile bodies, are not constant. Furthermore, requirement specifications related to the operation of the conveying system, such as the generation conditions of conveying requests, the number of candidate conveying routes, the degree of freedom of reallocation destination, the degree of freedom of the moving destination of available mobile bodies, the charging pattern of mobile bodies, the presence or absence of directivity of the conveying route, the route cost evaluation scale, and the equipment state of the conveying destination, are not constant. Also, matters to be determined cover a wide range not only in the configuration and operation of the conveying system, but also in multiple aspects, and there are also a plurality of algorithms for deriving these matters.

[0005] Under such circumstances, designers and the like who design a conveyance system are unclear about which algorithm to adopt, and may perform inefficient operations that require a lot of trial and error, such as creating a conveyance system model on a computer for each requirement specification and executing simulations for each algorithm option.

[0006] The present invention has been made in view of the above problems, and an object thereof is to provide a model training method capable of presenting materials for selecting an algorithm suitable for requirement specifications and the like even to users who do not know much about the types and characteristics of algorithms, and an algorithm presentation device.

Means for Solving the Problems

[0007] In order to achieve the above object, a model training method according to one aspect of the present invention is a method for training a model that presents a suitable algorithm from a plurality of algorithms related to a conveyance system that conveys a load using a plurality of moving bodies, the method including obtaining output values obtained by inputting condition values indicating at least one of the configuration and operation of the conveyance system to each of the plurality of algorithms, using the condition values as input data, and causing the model to learn with the output values associated with identifiers for identifying the algorithms as correct values.

[0008] Also, in order to achieve the above object, another algorithm presenting device of the present invention is a model training method for presenting a suitable algorithm from a plurality of algorithms related to a transport system that transports luggage using a plurality of moving bodies, the method comprising: obtaining output values obtained by respectively inputting condition values indicating at least one of the configuration and operation of the transport system into the plurality of algorithms; using the condition values as input data; and training the model by causing the model to learn with the output values associated with identifiers for identifying the algorithms as correct values, the model training method comprising: a model unit including a model trained in this way; a condition value acquisition unit that acquires a condition value indicating at least one of the configuration and operation of the transport system; and a candidate presentation unit that inputs the acquired condition value into the model and causes the model to present candidates for algorithms suitable for the condition value.

Advantages of the Invention

[0009] According to the present invention, it becomes possible to design a transport system using an algorithm suitable for the required specifications.

Brief Description of the Drawings

[0010]

Figure 1

Figure 2

Figure 3

Figure 4

Embodiments for Carrying Out the Invention

[0011] Hereinafter, embodiments of the model training method and the algorithm presentation device according to the present invention will be described with reference to the drawings. Note that the following embodiments are examples for explaining the present invention and are not intended to limit the present invention. For example, the shapes, structures, materials, components, relative positional relationships, connection states, numerical values, mathematical formulas, the content of each step in the method, the order of each step, etc. shown in the following embodiments are examples, and may include content not described below. In addition, geometric expressions such as parallel and orthogonal may be used, but these expressions do not indicate mathematical precision and include substantially allowable errors, deviations, etc. Also, expressions such as simultaneous and identical include a substantially allowable range.

[0012] Also, the drawings are schematic diagrams that are appropriately emphasized, omitted, or adjusted in ratio for explaining the present invention, and are different from the actual shapes, positional relationships, and ratios.

[0013] Also, in the following, a plurality of inventions may be comprehensively described as one embodiment. In addition, a part of the content described below is explained as an arbitrary component related to the present invention.

[0014] FIG. 1 is a block diagram showing the functional configuration of a model training system capable of executing a model training method. The model training system 110 is a system that realizes a model training method for presenting an algorithm selected from a plurality of algorithms used when constructing a transport system for transporting luggage using a plurality of mobile bodies. The model training system 110 includes a processor, and as a processing unit realized by causing the processor to execute a program, it includes a training condition value acquisition unit 111, an algorithm application unit 112, a model training unit 113, and a trained model unit 114.

[0015] A transportation system is a system in which a transportation process of transporting a load from one point to another according to a transportation request is repeatedly executed within a predetermined area, and includes a plurality of mobile bodies that can hold a load and move on their own. Note that the transportation system may also include a transfer device that transfers the load between the mobile bodies, a station that temporarily holds the load, and the like.

[0016] A mobile body is a device that can hold and move a load, and may be either a guided mobile body or an unguided mobile body. Examples of mobile bodies include those whose travel is controlled by a host computer and those that travel autonomously. In addition, mobile bodies include transport vehicles equipped with a transport device such as a stacker crane that can transport a load at least vertically, drones that can transport a load three-dimensionally, and the like.

[0017] The growth condition value acquisition unit 111 acquires a condition value indicating at least one of the configuration and operation of the transportation system. The condition value is a value included in the required specifications imposed on the transportation system at the stage of designing the transportation system. The types of condition values are not particularly limited, and examples include the mobile body density, a quantity related to the generation pattern of transportation requests, the number of candidate transportation routes, the mechanical characteristics of the mobile body, the degree of freedom of reallocation destination, the degree of freedom of the destination of an idle mobile body, the charging pattern, the presence or absence of directivity of the transportation route, the route cost evaluation scale, a quantity related to the facility state of the destination, and the like.

[0018] The mobile body density is a value indicating how many mobile bodies at most are desired to travel within a predetermined area where the mobile bodies travel. For example, when it is desired to operate 15 mobile bodies in an area of 400 square meters, the mobile body density is 15 / 400 [units / square meter]. Also, when 225 stop points are arranged within a predetermined area, the mobile body density may be set to 15 / 225 [units / stop P].

[0019] In addition, when the density of moving objects varies depending on the day, week, month, season, etc., the average value over the day, week, month, season, etc. may be adopted as the density of moving objects. Also, when it is known that there is a particularly congested area as a specification, the density of moving objects in the congested area may be adopted as a conditional value.

[0020] The quantities related to the generation pattern of conveyance requests include the number of in-out stations for delivering and receiving goods to / from moving objects, the frequency of occurrence of conveyance requests per unit time, the number of requests occurring within the time that can be regarded as simultaneous (the number of simultaneous occurrences of requests), etc.

[0021] The number of candidates for conveyance routes is the average value, maximum value, minimum value, etc. of the number of candidate routes selectable from the departure point to the arrival point of the goods included in the conveyance request.

[0022] The mechanical characteristics of the moving object are unique numbers corresponding to the mechanical characteristics of the moving object, such as 1 for a moving object that moves straight and curves while changing its posture, 2 for a moving object that only moves straight and can change direction at a fixed point, 3 for a moving object that can also move laterally without changing its posture, 4 for a moving object that can move laterally and diagonally without changing its posture, etc., or mechanical specification values such as acceleration / deceleration / maximum speed / idle time before and after operation.

[0023] The reallocation destination freedom degree is the number of buffers (the number of goods that the conveyance destination station can hold), the number of conveyance destination stations, the degree of freedom of conveyance order (indicating the ease of changing the conveyance order of goods), etc.

[0024] The moving destination freedom degree is the number of candidate temporary eviction destinations when an empty moving object not executing a conveyance task blocks the passage of a subsequent moving object, or the number of standby points per moving object. A standby point is a location that is close to the conveyance request source and does not interfere with the movement lines of other moving objects or people.

[0025] The quantities related to the charging pattern are the forced charging frequency when the moving object is driven by a battery, the number of charging points per moving object, etc.

[0026] The quantity related to the presence or absence of the directivity of the conveyance path is a unique number corresponding to the presence or absence of directivity, such as 1 when the moving direction of the main passage is set in advance and 2 when it is not set.

[0027] The route cost evaluation scale is a unique number corresponding to the method of weighting the route, such as 1 when the route cost is evaluated by distance, 2 when evaluated by time, and 3 when evaluated by energy.

[0028] The quantity related to the equipment state of the destination is a conditional value necessary for determining whether there is no luggage stay at the destination, such as how many conveyances the production equipment at the destination can process per unit time, the length and frequency of the downtime of the production equipment at the destination, and the number of buffers prepared at the destination.

[0029] The algorithm application unit 112 inputs the conditional value acquired by the growth condition value acquisition unit into one algorithm selected from a plurality of algorithms, performs calculations, and outputs an output value. It is desirable to select an algorithm corresponding to the input conditional value for one algorithm. Therefore, it is possible to verify a plurality of algorithms comprehensively, and it is desirable that a skilled person who is familiar with the differences in the characteristics of the plurality of algorithms makes the selection.

[0030] The types of algorithms are not particularly limited, and in addition to academically recognized algorithms such as Conflict-based search [Sharon et al. 2012], it also includes independently developed algorithms that exhibit routing functions, collision avoidance functions, and allocation functions respectively. Also included are algorithms for controlling the distribution movement system of moving bodies in the logistics conveyance system. Algorithms are identified based on the detailed functions of the algorithms, and if algorithms with the same name such as routing functions, collision avoidance functions, and allocation functions have subdivided functions, algorithms with different functions may be recognized as different algorithms.

[0031] Examples of the output values obtained by the algorithm selected in the algorithm application unit 112 include the production scheduling compliance rate, the placement performance degree of the moving body, and the logistics facility investment cost.

[0032] The production scheduling compliance rate is an index indicating how well the time-related results obtained by the algorithm comply with the required specifications. That is, the production scheduling compliance rate is used as an index representing performance. The production scheduling compliance rate is not an index focusing on the conveyance state of the moving body, but an important index for users who carry out production activities using the conveyance system. Production activities include not only part processing and assembly in the factory, but also shipping sorting by picking in the logistics center. The production scheduling compliance rate includes at least one of the throughput requirement fulfillment rate, the lead time constraint compliance rate, the waiting time constraint compliance rate, the order constraint compliance rate, etc.

[0033] The throughput requirement fulfillment rate indicates whether the number of packages that can be conveyed per unit time meets the required specifications.

[0034] The lead time constraint compliance rate indicates, for example, whether the time from the occurrence of the conveyance request to the completion of the conveyance of the package complies with the required specifications.

[0035] The waiting time constraint compliance rate indicates how well the results obtained by the algorithm comply with the constraint conditions of the conveyance time based on the replacement time of the production equipment.

[0036] The order constraint compliance rate indicates how well the conveyance order constraints of the required specifications are complied with.

[0037] The placement performance degree is an index indicating the ability related to placing a moving body. The placement performance degree of a moving body is used as an index representing performance. The placement performance degree refers to the performance focusing on the conveyance state of the moving body. The placement performance degree includes at least one of the congestion rate of the moving body, the margin rate of the moving body, the compound rate of the moving body, and the operation balance for each moving body.

[0038] The congestion rate of a moving body indicates the ratio of the time that the moving body temporarily stops waiting for the passage of other moving bodies during movement / conveyance to the time taken for the moving body to process a conveyance request.

[0039] The margin rate of a moving body indicates the ratio of the time during which no conveyance request is assigned within a certain period of time.

[0040] The compound rate of a moving body indicates the ratio of the number of times of conveying a load from a conveyance source in the same group as the conveyance destination to the total number of conveyances when grouping all stations serving as conveyance sources / destinations based on proximity.

[0041] The operation balance for each moving body is an index indicating the variation in the number of conveyance requests processed by each moving body within a certain period of time, for example, variance or standard deviation.

[0042] The logistics facility investment cost indicates an index for calculating the cost required to construct logistics facilities. That is, the logistics facility investment cost is used as an index representing cost. The logistics facility investment cost includes at least one of the number of input moving bodies, the physical space (area or volume) required for stable operation, and the equipment cost such as a distribution movement control device.

[0043] The number of input moving bodies indicates the number of moving bodies to be input when the production scheduling compliance rate and the placement performance degree are maximized for a given required specification.

[0044] The physical space required for stable operation is the minimum operating area of the mobile body for the production scheduling compliance rate to exceed the required level. However, in the case of a three-dimensional operating space where the mobile body also moves in the height direction between the shelves of the automated storage and retrieval system, it may be interpreted as the operating volume.

[0045] The equipment cost such as the distribution movement control CPU is, for example, the installation cost of the computer used for the calculation of the distribution movement control algorithm, and is determined by the minimum required memory capacity, clock number, number of GPUs, etc.

[0046] The model training unit 113 uses the condition values acquired by the training condition value acquisition unit 111 as input data, and causes the model included in the training model unit 114 to learn, as the correct value, information in which an identifier for identifying the selected algorithm is associated with the output value obtained by performing calculations using the algorithm selected by the algorithm application unit 112 with the condition values as inputs.

[0047] The model is a regression model used for machine learning of artificial intelligence, and is a model trained by so-called supervised learning. The type of the model is not particularly limited, and examples thereof include random forest, logistic regression, support vector machine (SVM), and the like.

[0048] Next, a specific model training method will be described. FIG. 2 is a diagram showing a specific relationship between input data and correct values. The operator inputs the condition values to the training condition value acquisition unit 111 for each case, selects the algorithm to be applied, and instructs the algorithm application unit 112 of the algorithm. In the case of Case 1 shown in FIG. 2, the condition values described in FIG. 2 are input, and the algorithm "A" is used as the algorithm. As the condition values, the average density of the mobile body and the local density of the mobile body are input as the mobile body density. Thus, similar condition values may be input. As the quantity related to the occurrence pattern of the conveyance request, the number of simultaneous conveyance requests (maximum value) is input, and as the quantity related to the charging pattern, the charging frequency is input. Further, other items are input.

[0049] Next, the algorithm application unit 112 inputs the condition values acquired by the growth condition value acquisition unit 111 into Algorithm A to perform calculations and derives the output values shown in FIG. 2. Specifically, as the output values derived, as the conveyance scheduling compliance rate, the maximum value of throughput, the compliance rate of LT (lead time) constraints, and the compliance rate of order constraints are output, and as cost metrics, the number of units capable of achieving maximum throughput, the area of the moving body passage region (physical space required for stable operation), and the equipment cost are output as calculation result values. In the case of Case 1, the placement performance degree is not output by Algorithm A.

[0050] Next, the model training unit 113 uses the condition values acquired from the growth condition value acquisition unit 111 as input data and the output values acquired from the algorithm application unit 112 as correct values to train the model.

[0051] The above processing is repeatedly executed up to Case 2, Case 3, ···, Case n (n is an integer).

[0052] Note that in Case n and Case m (m is an integer other than n), some or all of the items of the condition values may be different. Also, in Case n and Case m, the applied algorithms may be different. Also, in Case n and Case m, some or all of the items of the output values output may be different.

[0053] FIG. 3 is a block diagram showing the functional configuration of the algorithm presentation device. The algorithm presentation device 120 is a device that presents an algorithm suitable for the input condition values using a model trained by a model training method realized by a model training system 110 or the like, and includes a processor. The algorithm presentation device 120 includes a model unit 122, a condition value acquisition unit 121, and a candidate presentation unit 123 as processing units realized by causing the processor to execute a program. In the case of the present embodiment, the algorithm presentation device further includes an output value derivation unit 126 and a model re-learning unit 127.

[0054] When designing the transport system, the condition value acquisition unit 121 acquires condition values indicating at least one of the configuration and operation of the transport system as the required specifications. The items of the condition values acquired by the condition value acquisition unit 121 are the same as the items of the condition values acquired by the cultivation condition value acquisition unit 111.

[0055] The model unit 122 includes a model cultivated by the model cultivation method realized by the model cultivation system 110.

[0056] The candidate presentation unit 123 inputs the condition values acquired by the condition value acquisition unit 121 into the model included in the model unit 122, and causes the model to present candidates for algorithms suitable for the condition values. The method of presenting the algorithms is not particularly limited. For example, it is possible to present one algorithm that is optimal for the condition values. Also, it is possible to present the algorithm with the highest evaluation value indicating that it is an algorithm suitable for the condition values, and a plurality of algorithms with evaluation values lower than that, together with the corresponding evaluation values.

[0057] The method of presenting the algorithms is not particularly limited. For example, the algorithms may be presented to the operator by displaying the algorithms on a display device. Also, the algorithms may be presented by outputting the algorithms to another processing unit.

[0058] In this way, by presenting an algorithm suitable for the condition values, for example, a designer who designs a transport system can obtain output values using a suitable algorithm without being familiar with the characteristics of each of a plurality of types of algorithms, and it becomes possible to reflect the output values in the design of the transport system.

[0059] The output value derivation unit 126 inputs the condition values acquired by the condition value acquisition unit 121 into the algorithms presented by the candidate presentation unit 123 to derive output values. When there is one presented algorithm, the output value derivation unit 126 derives an output value based on the one algorithm. When there are multiple presented algorithms, calculations may be performed using each algorithm to derive multiple output values. The output value derivation unit 126 includes multiple types of algorithms and can perform calculations based on each algorithm.

[0060] By providing the output value derivation unit 126 in this way, the labor of inputting the condition values into the presented algorithms again can be saved, and designers of the conveyance system and the like can efficiently obtain output values.

[0061] The model re-learning unit 127 uses the condition values acquired by the condition value acquisition unit 121 as input data, and causes the model included in the model unit 122 to learn with the output values associated with the identifiers that identify the algorithms used by the output value derivation unit 126 as correct values. Thereby, the model can be continuously cultivated, and a model suitable for the environment of the owner of the algorithm presentation device 120 can be cultivated. Note that the model learning method executed by the model re-learning unit 127 may be the same as the learning method executed by the model cultivation unit 113, or may be different.

[0062] By cultivating a model using the model cultivation method in the above embodiment and presenting an algorithm using the cultivated model, designers of the conveyance system and the like can determine an algorithm suitable for the requirement specifications from among various algorithms without being familiar with the characteristics of the algorithms.

[0063] Also, according to the present embodiment, even without particularly being conscious of the types of algorithms, by inputting condition values, optimal output values can be obtained.

[0064] Note that the present invention is not limited to the above-described embodiments. For example, another embodiment realized by arbitrarily combining the components described in this specification and excluding some of the components may also be an embodiment of the present invention. Further, various modifications conceived by those skilled in the art without departing from the gist of the present invention, that is, the meaning indicated by the language described in the claims, are also included in the present invention.

[0065] For example, the candidate presenting unit 123 causes the model unit 122 to present a plurality of algorithms in order from the algorithm with the highest evaluation value, and when presenting each evaluation value, the algorithm presenting device 120 further, as shown in FIG. 4, for the plurality of algorithms presented by the candidate presenting unit 123, when there are a plurality of the evaluation values, a weight acquisition unit 124 that acquires a weight indicating which evaluation value to emphasize, and an analysis value presenting unit 125 that presents an analysis value for each algorithm by the analytic hierarchy process using the weight corresponding to the evaluation value presented by the model unit 122 may be provided.

[0066] Specifically, the model unit 122 presents an evaluation value related to the production scheduling compliance rate. For example, it presents Algorithm A: 56, Algorithm B: 84, Algorithm C: 140. At the same time, the model unit 122 presents an evaluation value related to the logistics facility investment cost. For example, it presents Algorithm A: 20, Algorithm B: 40, Algorithm C: 50. The evaluation value related to the logistics facility investment cost is such that the smaller the value, the less cost is required to construct the transportation system. The weight acquisition unit 124 acquires, as input from the user or designer, a ratio value indicating which of the evaluation values related to the production scheduling compliance rate, the vehicle allocation performance degree of the mobile body, and the logistics facility investment cost to emphasize. For example, it is assumed that the input is to emphasize at a ratio of production scheduling compliance rate: logistics facility investment cost = 9:7.

[0067] The analysis value presentation unit 125 outputs analysis values using the Analytic Hierarchy Process (AHP). Specifically, the evaluation values related to the production scheduling compliance rate are converted to satisfy the following equations (1) and (2).

[0068] x:y:z = 56:84:140 ··· Equation (1) x + y + z = 1 ··· Equation (2)

[0069] From the above, Algorithm A: 0.2, Algorithm B: 0.3, Algorithm C: 0.5. On the other hand, the evaluation value related to the logistics investment equipment cost is converted to a reciprocal because the larger the value, the better the evaluation value. Algorithm A: 1 / 20, Algorithm B: 1 / 40, Algorithm C: 1 / 50. Without changing the ratio of these reciprocals, they are converted so that the total becomes 1. As a result, Algorithm A: 0.5, Algorithm B: 0.3, Algorithm C: 0.2.

[0070] Apply Equation (3) according to the Analytic Hierarchy Process to the above-converted evaluation values and cost evaluation values

[0071]

Number

[0072] The upper part of the above equation relates to Algorithm A, the middle part relates to Algorithm B, and the lower part relates to Algorithm C. The one with the higher calculation result value, that is, Algorithm A, becomes the optimal option.

[0073] According to the above alternative example, by applying the Analytic Hierarchy Process to the results obtained by artificial intelligence, it becomes possible to present an algorithm that can consider the suppression of equipment investment costs (manufacturing cost suppression) of the conveying system.

Industrial Applicability

[0074] The present invention can be used when designing or changing a conveying system.

Explanation of Signs

[0075] 110 Model Training System 111 Training Condition Value Acquisition Unit 112 Algorithm Application Unit 113 Model Training Unit 114 Trained Model Unit 120 Algorithm Presentation Device 121 Condition Value Acquisition Unit 122 Model Unit 123 Candidate Presentation Unit 124 Acquisition Unit 125 Analysis Value Presentation Unit 126 Output Value Derivation Unit 127 Model Relearning Unit 140 Evaluation Value

Claims

1. An algorithm presentation device having a model that presents algorithm candidates from a plurality of algorithms for designing a transport system that transports luggage using a plurality of moving bodies, A logistics facility investment cost, which is an output value obtained by inputting, to each of the plurality of algorithms, the number of inbound / outbound stations that receive luggage from the moving bodies as a conditional value and the buffer number, which is the number of pieces of luggage that the destination station can hold, is acquired, and the model is trained with the conditional value as input data and the output value associated with the identifier that identifies the algorithm as the correct value, a model unit trained by a model training method; A conditional value acquisition unit that acquires a conditional value indicating at least one of the configuration and operation of the transport system; A candidate presentation unit that presents to the model a combination of an evaluation value, which is the output value obtained by inputting the acquired conditional value into the model, and an algorithm; An algorithm presentation device comprising the above.

2. The conditional value further includes at least one of the moving body density, a quantity related to the generation pattern of transport requests, the number of transport route candidates, the mechanical characteristics of the moving body, the reallocation destination freedom degree, the moving destination freedom degree of the idle moving body, the charging pattern, the presence or absence of directivity of the transport route, the route cost evaluation scale, and a quantity related to the facility state of the destination. The algorithm presentation device according to Claim 1.

3. The output value further includes at least one of the transport scheduling compliance rate and the moving body arrangement performance degree. The algorithm presentation device according to Claim 1 or 2.

4. The algorithm presentation device comprises a weight acquisition unit that acquires a weight indicating which evaluation value to prioritize among the plurality of evaluation values associated with the plurality of algorithms presented by the candidate presentation unit; and an analysis value presentation unit that presents an analysis value for each algorithm by the analytic hierarchy process using the evaluation value and the weight. The algorithm presentation device according to Claim 1.

5. An output value derivation unit that inputs the conditional value acquired by the conditional value acquisition unit into the algorithm presented by the candidate presentation unit to derive an output value The algorithm presentation device according to Claim 1 or 4 comprising the above.

6. A relearning unit is provided that uses the condition value acquired by the condition value acquisition unit as input data and causes the model included in the model unit to learn using the output value associated with an identifier that identifies the algorithm used in the output value derivation unit as the correct value. The algorithm presentation device according to claim 5.

Images