JP7637378B2 - Method for creating a solution derivation parameter vector, system for creating a solution derivation parameter vector, program for creating a solution derivation parameter vector, and program for creating a solution derivation parameter vector - Google Patents

Description

The present disclosure relates to a method for creating a solution derivation parameter vector, a system for creating a solution derivation parameter vector, a program for creating a solution derivation parameter vector, and a solution derivation program.

In production and logistics, etc., optimal operations or driving are desired to reduce costs such as labor, time, and expenses. In many cases, there are multiple indicators that must be considered for optimal operations or driving. Since it is difficult to clearly define the optimal values for these indicators, it is difficult to automate the determination of optimal operations or driving. In many cases, operations or driving that are considered optimal are planned based on the experience and practices of the person performing the operations or driving.

Optimal operation or operational efficiency is being sought by using artificial intelligence technology or mathematical optimization technology. For example, Patent Document 1 describes a device that derives a solution for an optimal loading plan when optimizing a loading plan for cargo. Patent Document 2 describes a system in which a computer derives a solution for an optimal operation plan when optimizing an operation plan for a power plant having multiple devices.

JP 2020-083486 A JP 2009-030476 A

N.Hansen: “The CMA Evolution Strategy: A Tutorial,”arXiv:1604.00772[cs, LG] (2016)

In various operation or operational plans, the plans are made to reflect the preferences or knowledge of the planner. For example, in cargo loading plans, factory production scheduling, or power plant operation plans, various plans are made to satisfy the planner's knowledge or preferences, such as experience or customs. Indicators to be considered for optimal operation or operation that satisfies the planner's preferences or knowledge include quantitative data such as the size or number of cargo, the number of workers, labor or operating hours, or the operating efficiency of equipment, as well as multiple types of parameters related to the planner's experience, sensibility, or customs. By setting these multiple types of parameters, an optimal solution that matches the planner's preferences or knowledge can be derived. Hereinafter, parameters related to the planner's experience, sensibility, or customs are referred to as "preference parameters." Multiple types of preference parameters used in one operation or operational plan are referred to as a "preference parameter vector."

For example, in a loading plan that reflects the various preferences or knowledge of the planner, in addition to the quantitative data, multiple types of preference parameters are required to reflect the preferences or knowledge of the planner in the solution. The multiple types of preference parameters include, for example, a parameter that determines the number of pallets to be used, a parameter that determines the loading rate of each pallet, a parameter that determines the degree to which the same type of cargo is bundled, a parameter that determines the stability of the cargo, and the like. The parameter that determines the number of pallets to be used can be set so that the number of pallets to be used is reduced in order to reduce transportation costs. The parameter that determines the loading rate of each pallet can be set so that the loading rate of each pallet is increased in order to reduce transportation costs. The parameter that determines the degree to which the same type of cargo is bundled can be set so that the degree to which the same type of cargo is bundled is increased in order to improve the ease of inspection. The parameter that determines the stability of the cargo can be set so that the stability of the cargo is increased in order to prevent cargo from collapsing.

In various plans, multiple preference parameters are often in a trade-off relationship with each other. In such cases, it is not possible to uniformly maximize or minimize all preference parameters. Furthermore, the preference parameter vector changes depending on various unpredictable situations and the preferences of each planner. For this reason, it is considered that planners set the preference parameter vector based on their knowledge or preferences, such as experience or custom, each time they make a plan. The preference parameter vector is directly related to the behavior of the algorithm that finds the solution to various plans, and it is difficult for those without knowledge of algorithms to adjust the preference parameter vector.

The device described in Patent Document 1 assumes that parameters equivalent to the preference parameter vector have already been set, and Patent Document 1 does not describe the process of determining an optimal preference parameter vector. It is unclear whether a solution that reliably reflects human preferences and knowledge can be derived. The system described in Patent Document 2 presents a solution for an optimal operation plan under parameters set by a planner. In this system, the planner decides whether or not he is satisfied with the presented solution, and if he is not satisfied, he resets the parameters. Because it is assumed that the preference parameter vector is directly input by a person, even if this system is used, it remains difficult to derive solutions for various plans that reliably reflect the preferences and knowledge of the planner, etc.

This disclosure describes a method for creating a solution derivation parameter vector that can easily and reliably determine a preference parameter vector that reflects human tastes and knowledge, a system for creating a solution derivation parameter vector, and a program for creating a solution derivation parameter vector, as well as a solution derivation program that can automatically output a solution that reliably reflects human tastes and knowledge.

The method of creating a solution derivation parameter vector disclosed herein includes acquiring provisional input data, acquiring a reference parameter vector, creating a plurality of formal parameter vectors by performing a predetermined operation on the reference parameter vector, creating a plurality of formal models in which each of the created plurality of formal parameter vectors is set, deriving a plurality of formal solutions by inputting provisional input data to each of the created plurality of formal models, displaying the derived plurality of formal solutions, acquiring selection information for the displayed plurality of formal solutions, updating the reference parameter vector based on the acquired selection information, and determining the updated reference parameter vector as a solution derivation parameter vector to be set in a solution derivation model that outputs a solution in response to input of input data.

In this method for creating a solution derivation parameter vector, multiple formal parameter vectors are created, and multiple provisional solutions are derived based on these formal parameter vectors. The derived multiple provisional solutions are displayed, and a reference parameter vector is updated based on selection information of the displayed multiple provisional solutions. The updated reference parameters are determined as the solution derivation parameter vector to be set in the solution derivation model. In this case, a solution derivation parameter vector that reflects human preferences and knowledge is determined according to the selection information of the provisional solutions, without directly inputting multiple types of preference parameters. Therefore, a solution derivation parameter vector that reflects human preferences and knowledge can be easily and reliably determined, and a solution derivation model that derives a solution that reliably reflects human preferences and knowledge can be created.

In the above method for creating a solution-derived parameter vector, the multiple formal parameter vectors may be derived by adding or subtracting at least one numerical value to the reference parameter vector. In this case, multiple formal parameter vectors based on the reference parameter vector can be easily created.

In the above method for creating a solution derivation parameter vector, at least one of the numerical values may include numerical values that are different from each other. In this case, more formal parameter vectors are derived, and therefore the solution derivation parameter vector can be determined more efficiently.

In the above method for creating a solution derivation parameter vector, at least one numerical value may be a numerical value derived by a calculation process that derives random numbers. In this case, a greater variety of provisional solutions is displayed, and the solution derivation parameter vector can be determined more efficiently.

In the above method for creating a solution derivation parameter vector, when at least one of the displayed multiple tentative solutions is selected, the reference parameter vector may be updated based on the formal parameter vector used in the calculation of the at least one tentative solution in updating the reference parameter vector. In this case, the formal parameter vector corresponding to the selected tentative solution is determined as the solution derivation parameter vector.

The above method for creating a solution derivation parameter vector further includes creating a plurality of formal parameter vectors by performing a predetermined calculation on the updated reference parameter vector, and the creation of the plurality of formal parameter vectors, the derivation of a plurality of tentative solutions, the display of the plurality of tentative solutions, and the update of the reference parameter vector are performed a plurality of times, and in determining the solution derivation parameter vector, the reference parameter vector updated a plurality of times may be determined as the solution derivation parameter vector. In this case, a solution derivation parameter vector that more reliably reflects human preferences and knowledge is determined.

In the above method for creating a solution derivation parameter vector, the provisional input data and the input data include information regarding the size and number of cargo, and the provisional solution and the solution may be a stowage pattern for multiple cargoes. In this case, a solution derivation parameter vector is determined according to the selection information for the stowage pattern for multiple cargoes. As a result, a solution derivation parameter vector that reflects human preferences and knowledge regarding stowage can be easily and reliably determined, and a solution derivation model that outputs a stowage pattern that reliably reflects human preferences and knowledge can be created.

The solution derivation model creation system of the present disclosure includes a first acquisition unit, a second acquisition unit, a formal parameter vector creation unit, a formal model creation unit, a derivation unit, a display unit, an update unit, and a determination unit. The first acquisition unit acquires formal input data. The second acquisition unit acquires a reference parameter vector. The formal parameter vector creation unit creates a plurality of formal parameter vectors by performing a predetermined calculation on the reference parameter vector. The formal model creation unit creates a plurality of formal models in which each of the created plurality of formal parameter vectors is set. The derivation unit derives a plurality of provisional solutions by inputting provisional input data to each of the created plurality of provisional models. The display unit displays the derived provisional solutions. The update unit updates the reference parameter vector based on the displayed plurality of provisional solutions. The determination unit determines the updated reference parameter vector as a solution derivation parameter vector to be set in a solution derivation model that outputs a solution in response to input of input data.

The program for creating a solution derivation parameter vector disclosed herein causes a computer to execute the following operations: acquiring provisional input data; acquiring a reference parameter vector; creating a plurality of formal parameter vectors by performing a predetermined operation on the reference parameter vector; creating a plurality of formal models in which each of the created plurality of formal parameter vectors is set; deriving a plurality of formal solutions by inputting provisional input data into each of the created plurality of formal models; displaying the derived plurality of formal solutions; acquiring selection information for the displayed plurality of formal solutions; updating the reference parameter vector based on the acquired selection information; and determining the updated reference parameter vector as a solution derivation parameter vector to be set in a solution derivation model that outputs a solution in response to input of input data.

The solution derivation program disclosed herein causes a computer to acquire input data and a solution derivation parameter vector created by the above-mentioned method for creating a solution derivation parameter vector, and derive a solution by inputting the input data into a solution derivation model in which the solution derivation parameter vector is set. In this case, a solution that reliably reflects human preferences and knowledge can be automatically output.

The solution derivation parameter vector creation method, solution derivation parameter vector creation system, and solution derivation parameter vector creation program disclosed herein can easily and reliably determine a preferred parameter vector that reflects human preferences and knowledge. The solution derivation program disclosed herein can automatically output a solution that reliably reflects human preferences and knowledge.

FIG. 1 is a block diagram of a solution deriving system according to the present embodiment. 2(a) and 2(b) are diagrams showing examples of cargo blocks. FIG. 3 is a conceptual diagram showing possible loading positions. FIG. 4 is a diagram showing a stowage pattern. FIG. 5 is a block diagram of a system for generating a solution-derived parameter vector. FIG. 6 is a diagram showing an example of a display screen of a provisional solution. FIG. 7 is a diagram showing an example of the hardware configuration of a system for creating a solution derivation parameter vector and a system for derivation of a solution. FIG. 8 is a flowchart showing a method for generating a solution derivation parameter vector. FIG. 9 is a flowchart showing the calculation process of the formal parameter vector. FIG. 10 is a flowchart showing the calculation process of the tentative solution. FIG. 11 is a flowchart showing a solution derivation method using the solution derivation model. FIG. 12 is a diagram showing an example of a solution derived using a preference parameter vector before learning. FIG. 13 is a diagram showing an example of a solution derived using the learned preference parameter vector. FIG. 14 is a diagram showing the transition of various preference parameters during the learning process.

The solution derivation system and solution derivation model creation system of the present disclosure will be described in detail below with reference to the attached drawings. In the description of the drawings, the same elements are given the same reference numerals, and duplicate explanations will be omitted.

First, referring to FIG. 1, a schematic configuration of the solution derivation system 1 in this embodiment will be described. FIG. 1 is a block diagram of the solution derivation system 1. The solution derivation system 1 derives a solution using a solution derivation model created in a solution derivation model creation system 20 described later. The solution derivation model creation system 20 also serves as a solution derivation parameter vector creation system. The solution derivation model is a model that has been trained to output a solution in response to input of input data. A solution derivation parameter vector is set in the solution derivation model. The solution derivation parameter vector is a preferred parameter vector used to derive a solution, and includes multiple types of preferred parameters. The solution derivation system 1 acquires input data and a solution derivation model. The solution derivation system 1 inputs the acquired input data into the acquired solution derivation model and outputs a solution.

In this embodiment, as an example, the derivation of a solution for a cargo stowage plan will be described, but the solution derivation system 1 and creation system 20 are not limited to this. For example, the solution derivation system 1 and creation system 20 may relate to factory production scheduling or power plant operation planning, etc. In this embodiment, the solution is a cargo stowage pattern.

The solution derivation system 1 includes a storage unit 2, an input data acquisition unit 3, a solution derivation model acquisition unit 4, a solution derivation unit 5, an image creation unit 6, and an output unit 7. The storage unit 2 stores pre-stored data, and stores various data acquired from other functional units. The storage unit 2 may also store data input from an external device other than the creation system 20 and the solution derivation system 1. The storage unit 2 stores the solution derivation parameter vector created in the creation system 20.

The input data acquisition unit 3 acquires input data. The input data may be stored in advance in the storage unit 2. The input data is problem data used for the problem to be solved. The problem data is quantitative data. In this embodiment, the problem data includes, for example, information on the type of cargo to be loaded, the size of each type of cargo, and the number of each type of cargo. Each cargo is, for example, a case such as a cardboard box in which contents are stored. Table 1 below shows an example of problem data as input data. In addition to the data shown in Table 1, the input data may include, for example, the weight of each cargo and information on the location where each cargo is placed. In addition, although the table shows four types of cargo, the number of types of cargo may be two or three, or may be five or more.

The solution derivation model acquisition unit 4 acquires a solution derivation model. In this embodiment, the solution derivation model acquisition unit 4 acquires a solution derivation parameter vector from the storage unit 2. The solution derivation model acquisition unit 4 acquires a solution derivation model by creating a solution derivation model in which the acquired solution derivation parameter vector is set. The solution derivation model acquisition unit 4 may acquire a solution derivation model in which a solution derivation parameter vector has already been set from the storage unit 2. In this embodiment, each of the multiple types of preference parameters included in the solution derivation parameter vector indicates a weight that prioritizes the corresponding attribute over attributes indicated by other preference parameters. The attribute relates to, for example, the user's preferences or knowledge. The user is, for example, a planner. For example, each preference parameter indicates a weight of an attribute related to the case volume, the case type, and the case position, etc.

The solution derivation unit 5 derives a solution based on the input data acquired by the input data acquisition unit 3 and the solution derivation model acquired by the solution derivation model acquisition unit 4. The solution derivation unit 5 inputs the input data to the solution derivation model in which the solution derivation parameter vector is set, and outputs a solution from the solution derivation model. In this embodiment, the solution is a loading pattern for multiple cargoes.

The image creation unit 6 creates an image to be displayed to the user based on the solution derived by the solution derivation unit 5. In this embodiment, the image creation unit 6 creates an image showing a loading pattern of cargo based on the solution derived by the solution derivation unit 5.

The output unit 7 outputs the solution derived by the solution derivation unit 5. In this embodiment, the output unit 7 outputs the image created by the image creation unit 6. For example, the output unit 7 displays the image created by the image creation unit 6 on a screen.

Next, an example of a solution derivation model in this embodiment will be described. In this embodiment, the solution derivation model is a trained model used for cargo stowage planning. The solution derivation model includes an equation with each preference parameter included in the solution derivation parameter vector as a coefficient. The stowage plan in this embodiment is based on the premise that blocks are created by combining cases of the same type, and these blocks are stacked on a pallet in order. Figures 2(a) and 2(b) show an example of a block. In this embodiment, the block is a rectangular parallelepiped. Figure 2(a) shows pinhole stacking, in which cases are alternately placed in odd and even rows. Pinhole stacking is excellent in stability. Figure 2(b) shows bar stacking, in which cases are densely arranged. Bar stacking is excellent in loading rate. In bar stacking, the number of cases arranged in the width direction, the depth direction, the height direction, etc. are arbitrary. The solution derivation system 1 and the creation system 20 can be applied to various other block configurations.

In the solution derivation model of this embodiment, a specific block is stacked at a specific position on a pallet, and then the blocks are stacked one by one using an equation in which a solution derivation parameter vector is set. Figures 3 and 4 are conceptual diagrams for explaining an example of the stacking process. In Figure 3, the two-dot chain line indicates the positions where blocks can be stacked. Hereinafter, the positions where blocks can be stacked are simply referred to as "stackable positions". For example, as shown in Figure 3, a specific block B1 is stacked at the left rear of the pallet, and thereafter, a stackable position at which a block is to be stacked is determined based on an equation in which a solution derivation parameter vector is set. The stackable positions are, for example, above, in front, and to the right of blocks already stacked. At this time, the stacking of new blocks also takes into account physical constraints, i.e., whether or not the block will fit into the space.

Figure 4 shows the stacking process as a tree structure. "T" indicates the top side. "F" indicates the front side. "R" indicates the right side. This shows a state in which a block B1 is stacked on the top side of the pallet bottom surface 11, and another block B2 is stacked on the right side of block B1 among the stackable positions "T", "F", and "R". When stacking more blocks from the state shown in Figure 4, the next block can be stacked on either the top or front side of block B1, or the top, front, or right side of block B2. In this way, blocks are stacked sequentially based on the stackable positions that can change depending on the stacking of blocks. The solution derivation unit 5 stacks blocks one by one based on the equation in which the solution derivation parameter vector is set, and ends the generation of the solution when there are no more blocks to be stacked. The solution derivation unit 5 derives a stacking pattern in which all the cases input as problem data are stacked as one solution.

Possible stacking positions are not limited to the three positions above, in front of, and to the right of blocks that have already been stacked. Positions other than these may be possible stacking positions. For example, as a variation of this embodiment, the possible stacking positions may be set at positions separated from blocks that have already been stacked.

In this embodiment, the solution derivation model is created on the premise of the above-mentioned stowage process. The solution derivation model includes, for example, the following formula (1) as the above-mentioned formula. The solution derivation model acquisition unit 4 creates the solution derivation model by setting corresponding preference parameters in one solution derivation parameter vector for each of w _{total_size} , w _unity , w _proximity , w _stability , w _area , w _zmax , and w _height in formula (1). Formula (1) shows an example of a formula in which a solution derivation parameter vector in a cargo stowage plan is set. The formula included in the solution derivation model in a cargo stowage plan is not limited to formula (1).

F is the score. In the above-mentioned stacking process, the solution derivation unit 5 determines which type of case should be stacked in which block configuration and at which possible stacking position based on the solution derivation model. The solution derivation unit 5 calculates the score F calculated by equation (1) for all possible stacking positions and various conditions, and selects the one with the maximum score. The solution derivation model is a multi-objective optimization problem.

Each of _{wtotal_size} , _wunity , _wproximity , _wstability , _warea , _wzmax , and _wheight is a preference parameter, and is acquired from the storage unit 2 by the solution derivation model acquisition unit 4. Equation (1) is created by the solution derivation model acquisition unit 4 from a solution derivation parameter vector including these preference parameters. _{wtotal_size} is a preference parameter related to the weight of an attribute indicating the priority of stacking cases of a type having a large volume. _wunity is a preference parameter related to the weight of an attribute indicating the priority of placing cases of the same type on the same pallet. wproximity is a preference parameter related to the weight of an attribute indicating the priority of placing cases of the same type next to each other. _wstability is a preference parameter related to the _weight of an attribute indicating the priority of stacking cases in a staggered manner in which even and odd tiers are stacked. w _area is a preference parameter related to the weight of an attribute indicating the priority of increasing the top area of a block. w _zmax is a preference parameter related to the weight of an attribute indicating the priority of reducing the top part of a block after stacking. w _height is a preference parameter related to the weight of an attribute indicating the priority of reducing the height of a block. It is extremely difficult for someone without knowledge of algorithms to directly adjust the values of these preference parameters.

Each of _{ftotal_size} , _funity , _fproximity , _fstability , _farea , _vzmaz , and _vheight is a subscore. Each subscore is normalized so that its value is approximately 0 to 1. Each subscore is determined by the solution derivation unit 5 based on the problem data and the above-mentioned premise conditions. Each subscore is determined by the progress of stacking, the type of case, the configuration of the block, and the possible stacking position of the block. The subscore can take multiple patterns depending on the possible stacking position of the block. _{ftotal_size} is a value obtained by dividing the total volume of the cases constituting the block to be stacked by the total volume of the remaining cases. _funity takes a value of 1 when the case is placed on the same pallet as the same type of case already stacked, and takes a value of zero otherwise. f _proximity takes a value of 1 when the block is placed adjacent to an already loaded case of the same type, and takes a value of zero otherwise. f _stability takes a value of 1 when the block is pinhole loaded, and takes a value of zero otherwise. f _area is the top area of the block to be loaded divided by the area of the pallet. v _zmaz is the height of the top of the load after loading divided by the maximum height that can be loaded on the pallet. v _height is the height of the block to be loaded divided by the maximum height that can be loaded on the pallet.

Next, referring to FIG. 5, a schematic configuration of the creation system 20 for creating a solution derivation parameter vector in this embodiment will be described. FIG. 5 is a block diagram of the creation system 20. The creation system 20 outputs preference parameters learned based on various data as a solution derivation parameter vector.

The creation system 20 includes a storage unit 21, a provisional input data acquisition unit 22, a reference parameter acquisition unit 23, a provisional parameter vector creation unit 24, a provisional model creation unit 25, a provisional solution derivation unit 26, an image creation unit 27, and a display unit 28. The creation system 20 further includes a selection information acquisition unit 31, a reference parameter vector update unit 32, a parameter vector determination unit 33, and a solution derivation model determination unit 34.

The storage unit 21 stores pre-stored data and various data acquired from other functional units. The storage unit 21 may also store data input from a device external to the creation system 20. For example, the storage unit 21 stores data acquired from at least one of the provisional input data acquisition unit 22, the reference parameter acquisition unit 23, the provisional parameter vector creation unit 24, and the selection information acquisition unit 31.

The provisional input data acquisition unit 22 acquires provisional input data. The provisional input data may be stored in advance in the storage unit 21. The provisional input data is part of the teacher data used to learn various preference parameters. The provisional input data is provisional problem data consisting of the same type of data as the input data described above. In this embodiment, the provisional input data includes, for example, information on the type of cargo to be loaded, the size of each type of cargo, and the number of each type of cargo, similar to the input data described above.

In this embodiment, the provisional input data acquisition unit 22 includes a provisional input data creation unit 41. The provisional input data creation unit 41 creates provisional input data. For example, the provisional input data creation unit 41 creates virtual provisional problem data from a portion of the provisional problem data stored in the storage unit 21. For example, the provisional input data creation unit 41 randomly generates virtual provisional problem data based on past data. For example, when three types of cases α, β, and γ are included in past problem data related to stowage, the provisional input data creation unit 41 creates provisional problem data in which the number of cases α, β, and γ in the past problem data is randomly set within the range of the minimum number and the maximum number. For example, when the number of cases α is 10 to 100, the number of cases β is 20 to 100, and the number of cases γ is 50 to 400, the provisional input data creation unit 41 creates provisional problem data with 30 cases α, 50 cases β, and 200 cases γ.

The reference parameter acquisition unit 23 acquires the reference parameters. The reference parameters are, for example, parameters having the same attributes as any of the multiple types of preference parameters included in the solution derivation parameter vector. The reference parameters have the same configuration as any of the multiple types of preference parameters included in the solution derivation parameter vector. For example, each of the multiple types of reference parameters corresponds to any of _{wtotal_size} , _wunity , _wproximity , _wstability , _warea , _wzmax , and _wheight .

In this embodiment, the reference parameter acquisition unit 23 acquires reference parameters corresponding to each of the multiple types of preference parameters included in the solution derivation parameter vector. The reference parameters corresponding to each of the multiple types of preference parameters included in the solution derivation parameter vector constitute a reference parameter vector. The reference parameter acquisition unit 23 acquires a reference parameter vector corresponding to the solution derivation parameter vector. The reference parameter acquisition unit 23 acquires the reference parameters from the storage unit 21, for example. The reference parameters may be values set when constructing a database in the storage unit 21, or may be values obtained by the previous learning.

The formal parameter vector creation unit 24 creates a formal parameter vector based on the reference parameters acquired by the reference parameter acquisition unit 23. The formal parameter vector creation unit 24 creates a plurality of formal parameter vectors by performing a predetermined calculation on the reference parameter vector. Each formal parameter vector has the same configuration as the solution derivation parameter vector. Each formal parameter vector includes a plurality of types of formal parameters. Each of the plurality of types of formal parameters included in each formal parameter vector corresponds to each of the plurality of types of preference parameters included in the solution derivation parameter vector. For example, each of the plurality of types of formal parameter vectors corresponds to any one of _{wtotal_size} , _wunity , _wproximity , _wstability , _warea , _wzmax , and _wheight . The formal parameters are values obtained by perturbing the reference parameters.

The formal parameter vector creation unit 24 includes a numerical value derivation unit 42 and a formal parameter vector derivation unit 43. The formal parameter vector creation unit 24 creates multiple formal parameter vectors by adding or subtracting the numerical values derived in the numerical value derivation unit 42 to the reference parameter vector.

The numerical derivation unit 42, for example, derives at least one numerical value used to create a formal parameter vector. The numerical derivation unit 42, for example, derives the numerical value based on an arithmetic process that derives random numbers. In other words, the numerical value derived by the numerical derivation unit 42 is a numerical value related to a random number. A "numerical value related to a random number" is a numerical value that the user cannot predict as to which numerical value will be determined. The numerical derivation unit 42, for example, derives a numerical value that is independent and appropriately large for at least one of a plurality of types of reference parameters. The numerical derivation unit 42, for example, derives a numerical value that is independent and appropriately large for each reference parameter. In other words, the numerical derivation unit 42 derives multiple numerical values. In this embodiment, the multiple numerical values are numerical values related to multiple random numbers calculated by an arithmetic process. The multiple numerical values may include the same value.

The formal parameter vector derivation unit 43 creates multiple formal parameter vectors by adding or subtracting at least one numerical value derived by the numerical value derivation unit 42 to the reference parameters acquired by the reference parameter acquisition unit 23. For example, the formal parameter vector derivation unit 43 adds or subtracts the numerical value derived by the numerical value derivation unit 42 to each of all reference parameters included in the reference parameter vector. In this way, each formal parameter vector is created. In this case, multiple reference parameters are perturbed simultaneously, so a variety of formal parameters are derived. For this reason, this method is suitable for the initial stage of learning, when the user's preferences or knowledge are roughly reflected.

As a variation of this embodiment, the formal parameter vector derivation unit 43 may select a reference parameter included in the reference parameter vector, and add or subtract the numerical value derived by the numerical value derivation unit 42 to the selected reference parameter. This method perturbs only a specific reference parameter, so the resulting formal parameter vectors lack diversity, but is suitable for fine tuning in the final stage of learning.

In a modified example of this embodiment, the numerical value derivation unit 42 calculates the covariance matrix C of the reference parameters updated by the previous learning, and derives a numerical value based on a random number according to a multivariate normal distribution N(0,C). The formal parameter vector derivation unit 43 adds or subtracts the numerical value based on the random number derived by the numerical value derivation unit 42 to the reference parameters. For example, this value based on the random number is a constant multiple of the random number according to the multivariate normal distribution N(0,C). For example, the following formulas (2), (3), and (4) hold true.

Equation (3) indicates that the random vector r is generated according to a multivariate normal distribution N(0,C). The first term on the right side of equation (2) is the reference parameter vector acquired by the reference parameter acquisition unit 23. σ is an appropriate positive constant. The second term on the right side of equation (2) is the product of σ and the random vector r. The left side of equation (2) indicates one formal parameter vector corresponding to the value of the subscript of i. L' is the subscript set of the reference parameters updated by the previous learning. |L'| is the number of L'.

The method using equations (2) to (4) relates to the method described in Patent Document 3. In this case, perturbation is performed taking into account the correlation between the reference parameters included in the reference parameter vector. This method is suitable for cases where multiple types of reference parameters are strongly correlated with each other.

The temporary model creation unit 25 creates a temporary model based on the formal parameter vector created in the formal parameter vector creation unit 24. The temporary model is a temporary solution derivation model that outputs a temporary solution in response to input of temporary input data. The temporary model has the same configuration as the solution derivation model. For example, the temporary model includes an equation similar to equation (1). One formal parameter vector is set in one temporary model. The temporary model creation unit 25 creates multiple temporary models in which each of the multiple formal parameter vectors is set. In the temporary model, the formal parameter vector is set to _{wtotal_size} , _wunity , _wproximity , _wstability , _warea , _wzmax , and _wheight . The temporary model creation unit 25 creates a temporary model by setting corresponding formal parameters in one formal parameter vector for each of _{wtotal_size} , _wunity , _wproximity , _wstability , _warea , _wzmax , and _wheight in formula (1). In this embodiment, the temporary model creation unit 25 creates a temporary model corresponding to each of the multiple formal parameter vectors created by the formal parameter vector creation unit 24. That is, the temporary model creation unit 25 creates multiple temporary models.

The tentative solution derivation unit 26 derives a tentative solution based on the tentative input data acquired by the tentative input data acquisition unit 22 and the tentative model created by the tentative model creation unit 25. The tentative solution derivation unit 26 inputs the tentative input data to the tentative model created by the tentative model creation unit 25, and outputs a tentative solution from the tentative model. In this embodiment, the tentative solution derivation unit 26 derives a tentative solution by a method similar to the method by which the solution derivation unit 5 derives a solution. The tentative solution derivation unit 26 derives multiple tentative solutions based on multiple tentative models. The tentative solution derivation unit 26 derives a tentative solution corresponding to each tentative model by inputting tentative input data to each of the multiple tentative models. In this embodiment, the tentative solutions are loading patterns for multiple cargoes, similar to the solutions.

The image creation unit 27 creates an image to be displayed to the user based on the multiple tentative solutions derived in the tentative solution derivation unit 26. In this embodiment, the image creation unit 27 creates multiple images showing cargo stowage patterns based on the tentative solutions derived in the tentative solution derivation unit 26. The multiple images correspond to the multiple tentative solutions derived in the tentative solution derivation unit 26. The multiple images and the multiple tentative solutions correspond one-to-one.

The display unit 28 displays the multiple images created by the image creation unit 27. In other words, the display unit 28 displays the multiple tentative solutions that have been derived. For example, as shown in FIG. 6, the display unit 28 displays multiple images showing different tentative solutions side by side. For example, the user clicks on a button labeled "Learning" in FIG. 6 to select an image that corresponds to the button.

The selection information acquisition unit 31 acquires selection information corresponding to a tentative solution selected by the user. The selection information acquisition unit 31 acquires information identifying a formal parameter vector corresponding to the tentative solution selected by the user. For example, the user selects at least one of the multiple images displayed by the display unit 28. The selection information acquisition unit 31 acquires information identifying a formal parameter vector corresponding to the image selected by the user. In other words, the selection information acquisition unit 31 acquires information identifying a formal parameter vector corresponding to the tentative solution selected by the user. If a tentative solution is not selected, the selection information acquisition unit 31, for example, causes the formal parameter vector creation unit 24 to create a formal parameter vector again and causes the tentative solution derivation unit 26 to derive a new tentative solution. As a modified example of this embodiment, the selection information acquisition unit 31 may acquire the priority order of multiple formal parameter vectors corresponding to multiple tentative solutions by the user's selection. The selection information may be information indicating that a tentative solution is not selected.

The reference parameter vector update unit 32 updates the reference parameter vector based on the displayed multiple tentative solutions. The reference parameter vector update unit 32 updates the reference parameter vector based on the selection information acquired by the selection information acquisition unit 31. In other words, the reference parameter vector update unit 32 updates the reference parameter vector based on the formal parameter vector corresponding to the tentative solution selected by the user. In this embodiment, the reference parameter vector update unit 32 updates the formal parameter vector corresponding to one of the selected tentative solutions as the reference parameter vector.

In a modified example of this embodiment, the reference parameter vector update unit 32 may calculate the average value of multiple formal parameter vectors corresponding to multiple selected provisional solutions, and update the calculated average value as the reference parameter vector. In this case, the reference parameter vector update unit 32 may calculate the average value by arithmetic averaging, or may calculate the average value by weighted averaging. For example, when the selection information acquisition unit 31 acquires priorities of multiple formal parameter vectors, weighting may be performed according to the acquired priorities.

In yet another modification of this embodiment, the reference parameter vector update unit 32 may update the reference parameter vector by setting a learning rate less than 1. For example, the reference parameter vector update unit 32 may update the reference parameter vector to a value calculated by equation (5).

η is a learning rate, which is greater than 0 and less than 1. The first term of equation (5) represents the current reference parameter vector. The second term of equation (5) is a value obtained by subtracting the current reference parameter vector from the formal parameter vector corresponding to the selected tentative solution and multiplying it by the learning rate. The formal parameter vector corresponding to the selected tentative solution in the second term of equation (5) may be a formal parameter vector corresponding to one of the selected tentative solutions, or may be the average value of multiple formal parameter vectors corresponding to the selected multiple tentative solutions.

The parameter vector determination unit 33 determines the reference parameter vector updated by the reference parameter vector update unit 32 as the solution derivation parameter vector to be set in the solution derivation model. When terminating the learning of the reference parameter vector, the parameter vector determination unit 33 determines the reference parameter vector at that time as the solution derivation parameter vector. In this embodiment, the parameter vector determination unit 33 judges whether or not to terminate the learning of the reference parameter vector.

The solution derivation model determination unit 34 determines information about the solution derivation model based on the solution derivation parameter vector determined by the parameter vector determination unit 33, and stores the determined information in the storage unit 21. The solution derivation model determination unit 34 may store the solution derivation model itself, in which the solution derivation parameter vector determined by the parameter vector determination unit 33 is set, in the storage unit 21, or may store the solution derivation parameter vector in the storage unit 21.

Next, the hardware configuration of the solution derivation system 1 and the creation system 20 will be described with reference to FIG. 7. FIG. 7 is a diagram showing an example of the hardware configuration of the solution derivation system 1 and the creation system 20. In the example shown in FIG. 7, the solution derivation system 1 and the creation system 20 are configured as an integrated unit. The solution derivation system 1 and the creation system 20 may also be configured separately from each other.

The solution derivation system 1 and the creation system 20 include a system 100. The system 100 includes a processor 101, a main memory device 102, an auxiliary memory device 103, a communication device 104, an input device 105, and an output device 106. The system 100 includes one or more computers configured with these pieces of hardware and software such as programs. The solution derivation system 1 and the creation system 20 are realized in cooperation with the hardware.

When the system 100 is composed of multiple computers, these computers may be connected locally or via a communication network such as the Internet or an intranet. This connection logically constructs a single solution derivation system 1 and creation system 20.

The processor 101 executes an operating system, application programs, etc. The main memory device 102 is composed of a ROM (Read Only Memory) and a RAM (Random Access Memory). For example, at least some of the various functional units of the solution derivation system 1 and the solution derivation model creation system 20 can be realized by the processor 101 and the main memory device 102.

The auxiliary storage device 103 is a storage medium composed of a hard disk, a flash memory, etc. The auxiliary storage device 103 generally stores a larger amount of data than the main storage device 102. For example, at least a portion of the storage units 2 and 21 may be realized by the auxiliary storage device 103.

The communication device 104 is configured with a network card or a wireless communication module. For example, at least a part of the input data acquisition unit 3, the solution derivation model acquisition unit 4, the output unit 7, the provisional input data acquisition unit 22, and the reference parameter acquisition unit 23 may be realized by the communication device 104. The input device 105 is configured with a keyboard, a mouse, a touch panel, and the like. For example, at least a part of the input data acquisition unit 3, the provisional input data acquisition unit 22, and the selection information acquisition unit 31 may be realized by the input device 105. The output device 106 is configured with a display, a printer, and the like. For example, at least a part of the output unit 7 and the display unit 28 may be realized by the output device 106. For example, the output device 106 displays an image created in at least one of the image creation unit 6 and the image creation unit 27 on a display.

The auxiliary storage device 103 stores in advance the programs and data necessary for processing. The solution derivation program causes the computer to execute each functional element of the solution derivation system 1. The solution derivation model creation program causes the computer to execute each functional element of the solution derivation model creation system 20. The solution derivation model creation program also serves as a solution derivation parameter vector creation program. The program causes, for example, processes S11 to S22, S31 to S32, S41 to S45, and S51 to S56, which will be described later, to be executed in the computer. This program may be provided after being recorded on a tangible recording medium such as a CD-ROM, DVD-ROM, or semiconductor memory. This program may be provided as a data signal via a communication network.

Next, a method for creating a solution derivation parameter vector will be described with reference to Figures 8, 9, and 10. Figure 8 is a flowchart showing the method for creating a solution derivation parameter vector. First, the provisional input data acquisition unit 22 acquires provisional input data (process S11).

Next, the reference parameter acquisition unit 23 acquires reference parameters (process S12). In this embodiment, the reference parameter acquisition unit 23 acquires a reference parameter vector including multiple types of reference parameters from the storage unit 21.

Next, the formal parameter vector creation unit 24 executes a calculation process of the formal parameter vector (process S13). In this embodiment, in this calculation process, the formal parameter vector creation unit 24 creates multiple formal parameter vectors based on the reference parameter vector acquired in process S12. The formal parameter vector creation unit 24 creates multiple formal parameter vectors by performing a predetermined calculation on the reference parameter vector.

Next, the temporary model creation unit 25 creates a temporary model (process S14). The temporary model creation unit 25 creates a temporary model based on the formal parameter vector created in process S13. The temporary model creation unit 25 creates a temporary model in which the formal parameter vector is set. In this embodiment, the temporary model creation unit 25 selects one from the multiple formal parameter vectors derived in process S13, and creates a temporary model in which the selected formal parameter vector is set. The temporary model creation unit 25 sets the selected formal parameter vector to, for example, w _{total_size} , w _unity , w _proximity , w _stability , w _area , w _zmax , and w _height in formula (1).

Next, the tentative solution derivation unit 26 executes a calculation process of the tentative solution (process S15). The tentative solution derivation unit 26 derives a tentative solution based on the tentative input data acquired in process S11 and the tentative model created in process S14. The tentative solution derivation unit 26 derives the tentative solution by inputting the tentative input data into the tentative model created in process S14.

Next, the provisional solution derivation unit 26 determines whether provisional solutions have been derived for all the formal parameter vectors created in process S13 (process S16). If it is determined that provisional solutions have not been derived for all the formal parameter vectors, the process returns to process S14. As a result, by performing processes S14 and S15 multiple times, the provisional model creation unit 25 creates multiple provisional models in which each of the multiple formal parameter vectors is set, and the provisional solution derivation unit 26 derives multiple provisional solutions. The provisional solution derivation unit 26 derives multiple provisional solutions by inputting provisional input data into each of the multiple provisional models.

When it is determined that tentative solutions have been derived for all tentative parameter vectors, the image creation unit 27 creates an image (process S17). The image creation unit 27 creates a plurality of images corresponding to each of the tentative solutions derived by processes S14 to S16. In this embodiment, each of the plurality of images shows a cargo stowage pattern corresponding to each tentative solution.

When process S17 is completed, the display unit 28 displays an image (process S18). The display unit 28 displays the multiple images created in process S17. In other words, the display unit 28 displays multiple tentative solutions.

When process S18 ends, the selection information acquisition unit 31 determines whether or not selection information has been acquired (process S19). The selection information acquisition unit 31 acquires selection information for the tentative solution displayed in process S18. The selection information acquisition unit 31 acquires selection information corresponding to the image selected by the user. In other words, the selection information acquisition unit 31 acquires information that identifies the formal parameter vector corresponding to the tentative solution selected by the user. If it is determined that selection information has not been acquired, the selection information acquisition unit 31 repeats process S19.

When it is determined that the selection information has been acquired, the reference parameter vector update unit 32 updates the reference parameter vector (process S20). The reference parameter vector update unit 32 updates the reference parameter vector based on the selection information acquired in process S19. In other words, the reference parameter vector update unit 32 updates the reference parameter vector based on the formal parameter vector corresponding to the tentative solution selected in process S19. In further other words, when at least one of the displayed multiple tentative solutions is selected, the reference parameter vector update unit 32 updates the reference parameter vector based on the formal parameter vector used in the calculation of the selected at least one tentative solution.

When process S20 ends, the parameter vector determination unit 33 determines whether or not to end learning of the reference parameter vector (process S21). If it is determined that learning should not end, the process returns to process S13. In this case, processes S13 to S21 are performed multiple times. In this case, in process S13, a predetermined calculation is performed on the updated reference parameter vector to derive multiple formal parameter vectors. The parameter vector determination unit 33 determines that learning should end when learning has been performed a predetermined number of times. The parameter vector determination unit 33 may determine that learning should end in response to a user operation.

When it is determined that learning should be terminated, the parameter vector determination unit 33 determines a solution derivation parameter vector (process S22). The parameter vector determination unit 33 determines the reference parameter vector updated in the immediately preceding process S20 as the solution derivation parameter vector. The parameter vector determination unit 33 determines the reference parameter vector updated multiple times as the solution derivation parameter vector by performing processes S13 to S21 multiple times.

When process S22 is completed, the solution derivation model determination unit 34 determines information about the solution derivation model (process S23). The solution derivation model determination unit 34 determines information about the solution derivation model based on the solution derivation parameter vector determined in process S21. When process S23 is completed, the series of processes ends.

Next, the calculation process of the formal parameter vector in step S13 will be described in detail with reference to FIG. 9. FIG. 9 is a flowchart showing the calculation process of the formal parameter vector.

First, the numerical value derivation unit 42 derives a numerical value related to a random number (process S31). At least one numerical value is a numerical value derived by a calculation process for deriving a random number. In this embodiment, the numerical value derivation unit 42 derives multiple random numbers calculated by the calculation process as multiple numerical values. The multiple numerical values may be adjusted to include numerical values different from each other.

Next, the formal parameter vector derivation unit 43 derives a formal parameter vector based on the reference parameter vector acquired in process S12 and the numerical value derived in process S31 (process S32). The formal parameter vector derivation unit 43 derives a plurality of formal parameter vectors by adding or subtracting at least one numerical value derived in process S31 to the reference parameter vector. Each formal parameter vector includes, for example, the provisional preference parameters w _{total_size} , w _unity , w _proximity , w _stability , w _area , w _zmax , and w _height in formula (1). When process S32 ends, the calculation process of the formal parameter vector ends, and the derived plurality of formal parameter vectors are returned.

Next, the calculation process of the provisional solution in step S14 will be described in detail with reference to FIG. 10. FIG. 10 is a flowchart showing the calculation process of the provisional solution.

First, the tentative solution derivation unit 26 determines sub-scores based on the tentative input data acquired in step S11 (step S41). The sub-scores are, for example, _{ftotal_size} , _funity , _fproximity , _fstability , _farea , _vzmaz , and _vheight in formula (1). The tentative solution derivation unit 26 determines one pattern from a plurality of patterns of sub-scores that can be set.

Next, the tentative solution derivation unit 26 calculates the score (process S42). The tentative solution derivation unit 26 calculates the score from the sub-score determined in process S41 and equation (1) in which the tentative parameter vector derived in process S32 is set.

Next, the tentative solution derivation unit 26 determines whether or not scores have been calculated for all patterns of the subscore (process S43). If it is determined that scores have not been calculated for all patterns, the process returns to process S41.

When it is determined that scores have been calculated for all patterns, the tentative solution derivation unit 26 determines a pattern based on the calculated scores (process S44). The tentative solution derivation unit 26 determines the pattern corresponding to the largest score among the multiple scores calculated by repeating processes S41 to S43. In other words, the tentative solution derivation unit 26 determines the type and arrangement of one block in process S44.

When process S44 ends, the tentative solution derivation unit 26 judges whether or not a tentative solution has been created (process S45). If it is judged that a tentative solution has not been created, the process returns to process S41. The tentative solution derivation unit 26 stacks blocks one by one by repeating processes S41 to S44. For example, if the tentative solution derivation unit 26 judges that a tentative solution has been created when it judges that there are no blocks to stack. If it is judged that a tentative solution has been created, the calculation process of the tentative solution is terminated and the tentative solution is returned.

Although an example of a method for creating a solution derivation parameter vector has been described above, the order of each process is not limited to this. For example, process S11 may be executed before process S15.

Next, a solution derivation method using the solution derivation system 1 will be described with reference to FIG. 11. FIG. 11 is a flowchart showing a method for creating a solution derivation parameter vector. Prior to this solution derivation method, the solution derivation parameter vector is stored in the storage unit 2. For example, the storage unit 2 stores the solution derivation parameter vector determined in step S22 in the method for creating a solution derivation parameter vector. First, the input data acquisition unit 3 acquires input data (step S51).

Next, the solution derivation model acquisition unit 4 acquires a solution derivation parameter vector (process S52). In this embodiment, the solution derivation model acquisition unit 4 acquires a solution derivation parameter vector including multiple types of preference parameters from the storage unit 2.

Next, the solution derivation model acquisition unit 4 acquires a solution derivation model (process S53). The solution derivation model acquisition unit 4 acquires a solution derivation model in which the solution derivation parameter vector acquired in process S52 is set. For example, the solution derivation model acquisition unit 4 acquires a solution derivation model in which the solution derivation parameter vector is set to _{wtotal_size} , _wunity , _wproximity , _wstability , _warea , _wzmax , and _wheight in formula (1).

Next, the solution derivation unit 5 executes a solution calculation process (process S54). The solution derivation unit 5 derives a solution based on the input data acquired in process S51 and the solution derivation model acquired in process S53. The solution derivation unit 5 may derive multiple solutions by creating parameters that perturb the solution derivation parameter vector, similar to the tentative solution derivation unit 26.

The solution calculation process executed by the solution derivation unit 5 is similar to the provisional solution calculation process executed by the provisional solution derivation unit 26. The solution calculation process executed by the solution derivation unit 5 executes processes from step S41 to step S45, with the provisional solution derivation unit 26 replaced by the solution derivation unit 5, the provisional input data replaced by the input data, the provisional parameter vector replaced by the solution derivation parameter vector, and the provisional solution replaced by the solution.

Next, the image creation unit 6 creates an image (process S55). The image creation unit 6 creates an image corresponding to the solution derived in process S54. In this embodiment, the created image shows the cargo stowage pattern corresponding to the solution. In the case where multiple solutions are derived in process S54, the image creation unit 6 creates multiple images corresponding to each of the multiple solutions.

Next, the output unit 7 outputs the solution (process S56). For example, the output unit 7 displays the image created by the image creation unit 6 on a screen. The output unit 7 may output the image created by the image creation unit 6 to another device.

Although an example of a solution deriving method has been described above, the order of each process is not limited to this. For example, process S51 may be executed before process S54.

Next, we will explain the disclosed method for creating a solution derivation parameter vector, the system for creating a solution derivation parameter vector, the program for creating a solution derivation parameter vector, and the effects of the solution derivation program.

In the method, system, and program for creating a solution derivation parameter vector disclosed herein, a plurality of formal parameter vectors are created, and a plurality of provisional solutions are derived based on these formal parameter vectors. The derived provisional solutions are displayed, and a reference parameter vector is updated based on the selection information of the displayed provisional solutions. The updated reference parameters are determined as the solution derivation parameter vector to be set in the solution derivation model. In this case, the solution derivation parameter vector reflecting human preferences and knowledge is determined according to the selection information of the provisional solutions, without the user directly inputting a plurality of types of preference parameters. For this reason, the solution derivation parameter vector reflecting human preferences and knowledge can be easily and reliably determined, and a solution derivation model that derives a solution that reliably reflects human preferences and knowledge can be created. For example, the solution derivation parameter vector reflecting the preferences and knowledge of an experienced planner can be easily and reliably determined, even if the user does not directly set preference parameters.

FIG. 12 shows an example of an image showing a stacking pattern as a solution derived using a reference parameter vector stored in advance in the storage unit 21. In other words, FIG. 12 shows an example of a solution derived using a preference parameter vector before learning of the present disclosure. FIG. 13 shows an example of an image showing a stacking pattern as a solution derived using a solution derivation parameter vector of the present disclosure. In other words, FIG. 13 shows an example of a solution derived using a preference parameter vector after learning of the present disclosure. In creating the solution derivation parameter vector, an image of a provisional solution in which "cases of the same type are placed together" was selected. FIG. 14 shows the transition of various reference parameters in the process of repeating the selection of a provisional solution and the update of the reference parameter vector in the method of creating the solution derivation parameter vector. In other words, FIG. 14 shows the transition of various preference parameters in the learning process. In FIG. 14, data D1 indicates w _{total_size} , data D2 indicates w _unity , data D3 indicates w _proximity , data D4 indicates w _stability , data D5 indicates w _area , data D6 indicates w _zmax , and data D7 indicates w _height .

As shown in FIG. 14, the values of w _area and w _unity are large in the initial reference parameter vector. Therefore, as shown in FIG. 12, blocks with large upper areas are preferentially stacked downward. As the selection of the provisional solution and the update of the reference parameter vector are repeated, the value of w _{total_size} increases. Then, as shown in FIG. 13, in the solution derived using the solution derivation parameter vector, blocks that combine types of cases with large total volumes can be stacked downward. In the process of creating the solution derivation parameter vector, the user does not refer to the values of the individual preference parameters, and the user does not directly input the preference parameters.

In creating the multiple formal parameter vectors, the multiple formal parameter vectors are derived by adding or subtracting at least one numerical value to the reference parameter vector. In this case, multiple formal parameter vectors based on the reference parameter vector can be easily created.

At least one of the numerical values may include numerical values that are different from each other. In this case, more formal parameter vectors are derived, and therefore the solution derivation parameter vector can be determined more efficiently.

At least one of the numerical values may be a numerical value derived by a calculation process that derives random numbers. In this case, a greater variety of provisional solutions is displayed, and the solution derivation parameter vector can be determined more efficiently.

When at least one of the displayed multiple tentative solutions is selected, the reference parameter vector is updated based on the formal parameter vector used in the calculation of the at least one tentative solution in updating the reference parameter vector. In this case, the formal parameter vector corresponding to the selected tentative solution is determined as the solution derivation parameter vector.

A plurality of formal parameter vectors are created by performing a predetermined calculation on the updated reference parameter vector. The creation of the plurality of formal parameter vectors, the derivation of the plurality of tentative solutions, the display of the plurality of tentative solutions, and the updating of the reference parameter vector are performed a plurality of times. In determining the solution derivation parameter vector, the reference parameter vector that has been updated a plurality of times is determined as the solution derivation parameter vector. In this case, a solution derivation parameter vector that more reliably reflects human preferences and knowledge is determined.

The provisional input data and the input data include information regarding the size and number of cargo, and the provisional solution and the solution are patterns for loading multiple cargoes. In this case, a solution derivation parameter vector is determined according to the selection information for the loading pattern for multiple cargoes. As a result, a solution derivation parameter vector that reflects human preferences and knowledge regarding loading is easily and reliably determined, and a solution derivation model that outputs a loading pattern that reliably reflects human preferences and knowledge can be created.

The solution derivation program disclosed herein causes a computer to acquire input data and a solution derivation parameter vector created by the above-mentioned method for creating a solution derivation parameter vector, and derive a solution by inputting the input data into a solution derivation model in which the solution derivation parameter vector has been set. In this case, a solution that reliably reflects human preferences and knowledge can be automatically output. Since a stowage plan that reflects the knowledge of an experienced planner is automatically obtained, the entire stowage work can be performed appropriately and quickly, even when stowage is performed by a robot or the like.

20 Creation system 22 Provisional input data acquisition unit 23 Reference parameter acquisition unit 24 Provisional parameter vector creation unit 25 Provisional model creation unit 26 Provisional solution derivation unit 28 Display unit 32 Reference parameter vector update unit 33 Parameter vector determination unit

Claims (9)

Obtaining provisional input data;
Obtaining a reference parameter vector;
generating a plurality of formal parameter vectors by performing a predetermined operation on the reference parameter vector;
creating a plurality of formal models in which the created plurality of formal parameter vectors are set;
deriving a plurality of provisional solutions by inputting the provisional input data into each of the plurality of provisional models created;
displaying the derived tentative solutions; and
obtaining selection information for the displayed provisional solutions; and
updating the reference parameter vector based on the obtained selection information;
and determining the updated reference parameter vector as a solution derivation parameter vector to be set in a solution derivation model that outputs a solution in response to input of input data. The method for creating a solution derived parameter vector according to claim 1, wherein the plurality of formal parameter vectors are derived by adding or subtracting at least one numerical value to the reference parameter vector in creating the plurality of formal parameter vectors. The method for creating a solution derivation parameter vector according to claim 2, wherein the at least one numerical value includes numerical values that are different from each other. The method for creating a solution derivation parameter vector according to claim 2 or 3, wherein the at least one numerical value is a numerical value derived by a calculation process that derives a random number. The method for creating a solution derivation parameter vector according to any one of claims 1 to 4, wherein, when at least one of the displayed provisional solutions is selected, the reference parameter vector is updated based on the provisional parameter vector used in the calculation of the at least one provisional solution in updating the reference parameter vector. A method for creating a solution-deriving parameter vector, further comprising causing the one or more computers to perform the predetermined calculation on the updated reference parameter vector to create the plurality of formal parameter vectors,
the generation of the plurality of formal parameter vectors, the derivation of the plurality of tentative solutions, the display of the plurality of tentative solutions, and the updating of the reference parameter vector are performed a plurality of times;
The method for creating a solution derivation parameter vector according to claim 1 , wherein in determining the solution derivation parameter vector, the reference parameter vector that has been updated a plurality of times is determined as the solution derivation parameter vector. the preliminary input data and the input data include information regarding size and number of shipments;
The method for creating a solution deriving parameter vector according to claim 1 , wherein the provisional solution and the solution are loading patterns of a plurality of cargoes. A first acquisition unit that acquires provisional input data;
A second acquisition unit that acquires a reference parameter vector;
a formal parameter vector creation unit that creates a plurality of formal parameter vectors by performing a predetermined calculation on the reference parameter vector;
a temporary model creation unit that creates a plurality of temporary models in which each of the created plurality of formal parameter vectors is set;
a derivation unit that derives a plurality of provisional solutions by inputting the provisional input data into each of the plurality of provisional models that have been created;
a display unit that displays the derived provisional solutions;
an update unit that updates the reference parameter vector based on the displayed provisional solutions;
A system for creating a solution derivation parameter vector comprising: a determination unit that determines the updated reference parameter vector as a solution derivation parameter vector to be set in a solution derivation model that outputs a solution in response to input of input data. Obtaining provisional input data and a reference parameter vector;
generating a plurality of formal parameter vectors by performing a predetermined operation on the reference parameter vector;
creating a plurality of formal models in which the created plurality of formal parameter vectors are set;
deriving a plurality of provisional solutions by inputting the provisional input data into each of the plurality of provisional models created;
displaying the derived tentative solutions; and
updating the reference parameter vector based on the displayed provisional solutions; and
and determining the updated reference parameter vector as a solution derivation parameter vector to be set in a solution derivation model that outputs a solution in response to input of input data.

Images