Inference system, inference method, and inference program

Abstract

We provide an inference system that can predict the market price of rare metals with high accuracy. [Solution] The inference system of the present invention includes a processor and a memory device, wherein the processor infers the market price of rare metals, and the processor comprises a machine learning unit that generates a trained model by machine learning based on training data in which at least one feature data of the rare metals,

Description

Technical Field

[0001] The present invention relates to an inference system, an inference method, and an inference program, and particularly to an inference system, an inference method, and an inference program for inferring the market price of rare metals.

Background Art

[0002] There is a system for predicting energy prices and crude oil prices (Patent Document 1).

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, there are almost no systems for predicting the market price of rare metals. In particular, there is no system for predicting the market price of rare metals in consideration of the supply and demand of products using rare metals.

Means for Solving the Problems

[0005] The inference system of the present invention includes a processor and a memory device, wherein the processor infers the market price of rare metals, and the processor comprises a machine learning unit that generates a trained model by machine learning based on training data in which at least one feature data of the rare metals, [Effects of the Invention]

[0006] According to the present invention, it is possible to predict the market price of rare metals with high accuracy. [Brief explanation of the drawing]

[0007] [Figure 1] This is a block diagram showing an example of the system configuration of the inference system of this embodiment. [Figure 2] This is a block diagram showing an example of the system configuration of the artificial intelligence server in this embodiment. [Figure 3] This is a block diagram showing an example of the system configuration of the learning database server in this embodiment. [Figure 4] This is a block diagram showing an example of the computer system configuration of this embodiment. [Figure 5] This figure shows an example of machine learning data in which training feature data and training data on the actual market price of rare metals are associated. [Figure 6] This flowchart shows an example of the inference method of this embodiment. [Figure 7] This sequence diagram illustrates an example of the operations involved in acquiring and preprocessing various types of data, as well as generating and storing trained models. [Figure 8] This is a sequence diagram illustrating an example of the operation of inference using a trained model and the determination of the inference result. [Modes for carrying out the invention]

[0008] An inference system according to an embodiment of the present invention will be described with reference to the drawings. Figure 1 is a block diagram showing an example of the system configuration of the inference system according to this embodiment. The inference system 1 includes a processor and a memory device, and the processor infers the market price of rare metals.

[0009] As shown in Figure 1, the inference system 1 comprises an artificial intelligence server 3, a learning database server 4, and a computer (PC) 5, all electrically connected and able to communicate with each other via a network 2.

[0010] Figure 2 is a block diagram showing an example of the system configuration of the artificial intelligence server in this embodiment. As shown in Figure 2, the artificial intelligence server 3 includes a processor 21, a storage device (e.g., ROM, RAM, HDD, etc.) 22, an input device 23, an interface 24, and an output device 25.

[0011] The processor 21 is a control device such as a CPU, MPU, or GPU, and includes a data acquisition unit 26, a machine learning unit 27, a trained model storage unit 28, and an inference unit 29. The data acquisition unit 26, the machine learning unit 27, the trained model storage unit 28, and the inference unit 29 are electrically connected by a bus (not shown) and can communicate with each other.

[0012] Figure 3 is a block diagram showing an example of the system configuration of the learning database server in this embodiment. As shown in Figure 3, the learning database server 4 comprises a processor 30, a learning database 40, a storage device (e.g., ROM, RAM, HDD, etc.) 31, an input device 32, an interface 33, and an output device 34. The processor 30 is a control device such as a CPU, MPU, or GPU, and comprises a data storage unit 35, a data acquisition unit 36, and a pre-processing unit 37.

[0013] The training database 40 comprises feature data 41 and spot price data 42 for rare metals. The feature data 41 and spot price data 42 for rare metals are interconnected by being linked to first time data and second time data. Here, rare metals refer to metals that are scarce on Earth and have limited supply. Examples include cobalt, nickel, and lithium, and also include rare earth elements.

[0014] The feature data 41 includes rare metal supply chain data 43, non-ferrous metal supply chain data 44, rare metal resource development funding data 45, non-ferrous metal resource development funding data 48, rare metal futures price data 49, non-ferrous metal futures price data 50, non-ferrous metal spot price data 51, economic indicator data 52, and exchange rate data 53, and these data are associated with the first time period data.

[0015] Furthermore, feature data 41 includes energy industry indicator data 54, automotive industry indicator data 55, aerospace industry indicator data 56, and defense industry indicator data 57, and these data are associated with the second time data.

[0016] A first time-based data set, based on at least one of the following: rare metal supply chain data 43, non-ferrous metal supply chain data 44, rare metal resource development financing data 45, non-ferrous metal resource development financing data 48, rare metal futures price data 49, non-ferrous metal futures price data 50, non-ferrous metal spot price data 51, economic indicator data 52, and exchange rate data 53, is associated with a second time-based data set, based on at least one of the following: energy industry indicator data 54, automotive industry indicator data 55, aerospace industry indicator data 56, and defense industry indicator data 57. This expands the feature data used for training.

[0017] The rare metal supply chain data 43 includes at least one of the producer data regarding at least one of the acceptance, consumption, shipment, inventory, and sales breakdown of rare metals by producers, the seller data regarding at least one of the acceptance, shipment, inventory, and sales breakdown of rare metals by sellers, the consumer data regarding at least one of the self-generation, self-production, acceptance, consumption, shipment, inventory, and consumption breakdown of rare metals by consumers, the air cargo data of rare metals, and at least one of the port cargo data of rare metals.

[0018] The supply chain data 44 of non-ferrous metals other than rare metals includes at least one of the producer data regarding at least one of the acceptance, consumption, shipment, inventory, and sales breakdown of non-ferrous metals other than rare metals by producers, the seller data regarding at least one of the acceptance, shipment, inventory, and sales breakdown of non-ferrous metals other than rare metals by sellers, the consumer data regarding at least one of the self-generation, self-production, acceptance, consumption, shipment, inventory, and consumption breakdown of non-ferrous metals other than rare metals by consumers, the air cargo data of non-ferrous metals other than rare metals, and at least one of the port cargo data of non-ferrous metals other than rare metals.

[0019] The rare metal resource development financing data 45 includes at least one of the financing data for at least one of the producers, sellers, and consumers of rare metals, and at least one of the financing data for the acquirers who acquire at least one of the producers, sellers, and consumers of rare metals.

[0020] The supply chain data 48 of non-ferrous metals other than rare metals includes at least one of the financing data for at least one of the producers, sellers, and consumers of non-ferrous metals other than rare metals, and at least one of the financing data for the acquirers who acquire at least one of the producers, sellers, and consumers of non-ferrous metals other than rare metals.

[0021] The futures market data for rare metals 49, the futures market data for non-ferrous metals other than rare metals 50, and the spot market data for non-ferrous metals other than rare metals 51 include not only the market price (execution price), but also the ask or offer price, ask or offer quantity, bid price, bid quantity, trading volume, rate of increase, rate of decrease, spread, and volatility.

[0022] Economic indicator data52 includes at least one of the following for each country: GDP, GDP per capita, economic growth rate, unemployment rate, inflation rate, producer price index, purchasing power parity, balance of payments, trade balance, interest rates, government bond yields, government debt, fiscal balance, labor productivity, retail sales, manufacturing production index, construction activity index, housing starts, industrial production index, automobile sales, consumer confidence index, business confidence index, inventory index, monetary reserves, tourism revenue, income inequality (Gini coefficient), population growth rate, labor force participation rate, average wage growth rate, import and export price index, and food price index.

[0023] The exchange rate data 53 includes not only the exchange rate price (executed price), but also the ask or offer price, ask or offer quantity, bid price, bid quantity, trading volume, percentage increase, percentage decrease, spread, and volatility of the spot market price of rare metals.

[0024] Energy industry indicator data54 includes at least one of the following: crude oil price, natural gas price, natural gas futures price, coal price, coal futures price, energy consumption, crude oil production, crude oil inventory, solar power capacity, wind power capacity, power generation, electricity supply, renewable energy share, energy prices, energy efficiency indicators, power plant CO2 emissions, and wholesale electricity price.

[0025] Automotive industry indicator data55 includes at least one of the following: vehicle production volume, vehicle sales volume, market share of vehicle sales by country, global market share of vehicle sales, vehicle export volume, vehicle import volume, electric vehicle sales volume, hybrid vehicle sales volume, autonomous vehicle adoption rate, average vehicle fuel consumption, vehicle CO2 emissions, number of employees in the automotive industry, contribution of the automotive industry to GDP, number of vehicle recalls, total amount of vehicle loans, used car market size, and vehicle registration volume.

[0026] Aerospace industry indicator data56 includes at least one of the following: aircraft production, aircraft orders, aircraft deliveries, aircraft utilization rates, number of flights, aircraft passenger kilometers, number of aircraft passengers, air cargo volume, airline revenue, airline operating costs, aircraft fuel consumption, average airfare price, aircraft CO2 emissions, aviation industry's contribution to GDP, number of new routes, number of rocket launches, number of satellite launches, number of satellites, and number of spacecraft development projects.

[0027] The military industry indicator data 57 includes at least one of the following: military budget size, defense spending, arms exports, arms imports, arms production, military enterprise revenue, number of military industry employees, contribution of the defense industry to GDP, number of missile defense systems deployed, and number of military satellite launches.

[0028] The learning data for rare metal spot market prices 42 includes not only the spot market price (executed price) of rare metals, but also the sell order (ask or offer) price, sell order (ask or offer) quantity, buy order (bid) price, buy order (bid) quantity, trading volume, rate of increase, rate of decrease, spread, and volatility.

[0029] The spot market price data 42 for rare metals used for training is associated with feature data and stored in the training database 40 as training data. The training data is the data used for machine learning by the machine learning unit 27 of the artificial intelligence server 3.

[0030] Figure 4 is a block diagram showing an example of the system configuration of the computer in this embodiment. As shown in Figure 4, the computer 5 comprises a processor 501, a storage device (e.g., ROM, RAM, HDD, etc.) 522, an input device 523, an interface 524, and an output device 525. The processor 501 is a control device such as a CPU, MPU, or GPU, and comprises a data acquisition unit 502, a selection unit 505, and a determination / modification unit 506.

[0031] The data acquisition unit 502 acquires training data from the training database 40, which is a combination of training feature data 41 and spot price data 42 for rare metals, and displays the training data on the display of the output device 525. The computer 5 is then operated by the operator, who, while viewing the training data displayed on the screen, selects the input location for the training data using the selection unit 505 and inputs or modifies the training data using the input device 523. The determination / modification unit 506 then determines the input or modified training data, and the determined training data is transmitted to the training database server 4 via the interface (transmit / receive unit) 524. The determined training data is then stored in the training database 40 via the interface (transmit / receive unit) 33.

[0032] Furthermore, the operator, while viewing the display, selects the input points for inference feature data and time data using the selection unit 505, and inputs or modifies the inference feature data and time data using the input device 523. Here, the inference feature data is the same as the training feature data. The time data is data that specifies a predetermined time for estimating the price of rare metals, but if time data (for example, one week later or one month later) is already set by default, inputting time data is optional.

[0033] Then, the determination / correction unit 506 determines the input feature data and time data for inference, and the determined feature data and time data for inference are transmitted to the artificial intelligence server 3 via the interface (transmitting / receiving unit) 524. The determined feature data for inference is input (stored) in the storage device 22 via the input device 23 and the interface (transmitting / receiving unit) 24.

[0034] Figure 5 shows an example of machine learning data (time series data) in which spot market price data and feature data of rare metals used for training are associated.

[0035] As shown in Figure 5, the first time data 60 for training associates the spot market data 64 and feature data 65 for training. The second time data 61 for training associates the spot market data 64 and feature data (for augmentation) 66 for training.

[0036] Spot market data 64 includes spot market prices 67 and spot market order book information 68 for rare metals used for learning. Feature data 65 includes supply chain data 70 for rare metals and non-ferrous metals other than rare metals, resource development funding data 71 for rare metals and non-ferrous metals other than rare metals, and futures market data 72 for rare metals and non-ferrous metals other than rare metals. Feature data (for expansion) 66 includes energy industry indicator data 73, automotive industry indicator data 74, and aerospace industry indicator data 75.

[0037] The machine learning unit 27 generates a trained model by machine learning based on the trained feature data, which is augmented by associating it with first time data based on at least one of the following: rare metal supply chain data, non-ferrous metal supply chain data other than rare metals, rare metal resource development funding data, non-ferrous metal resource development funding data other than rare metals, rare metal futures price data, non-ferrous metal futures price data other than rare metals, non-ferrous metal spot price data other than rare metals, economic indicator data, and exchange rate data, and second time data based on at least one of the trained feature data (for augmentation) 66: energy industry indicator data, automobile industry indicator data, aerospace industry indicator data, and defense industry indicator data.

[0038] Trends among producers, sellers, and consumers of rare metals and other non-ferrous metals, as well as resource development funding and futures prices, correlate with spot price data for rare metals. Therefore, by generating a trained model using machine learning with this data, it is possible to predict the price of rare metals with high accuracy.

[0039] Furthermore, the supply and demand of rare metals in industries where they are used, such as the energy industry (e.g., wind turbines and solar panels), the automotive industry (e.g., batteries), the aerospace industry (e.g., high-performance alloys and heat-resistant materials), and the defense industry (e.g., high-performance alloys and heat-resistant materials), correlate with spot price data for rare metals. Therefore, by generating a trained model using machine learning with this augmented data, it is possible to infer the price of rare metals with high accuracy.

[0040] Furthermore, by associating the first time data with the second time data, it is possible to generate augmented data (augmented training feature data) that takes into account the time lag between feature data 65 and feature data (for augmentation) 66, enabling accurate inference of rare metal prices. Also, by associating the first time data with the second time data, it is possible to understand how feature data 65 has changed before and after the time based on the second time data of the feature data (for augmentation). Then, by linking this change with the spot price data of rare metals, and using the training data based on this change in machine learning, it is possible to accurately infer the price of rare metals.

[0041] For example, if supply chain data 70 changes before and after the second time-based data 61 of energy industry indicator data 73, it is possible to estimate the price of rare metals taking this change into account, thereby allowing for a highly accurate inference of the price of rare metals.

[0042] In this way, by associating feature data 65 and feature data (for augmentation) 66, which have a relationship with the spot market price of rare metals, and by using training data based on these relationships in machine learning, it is possible to infer the market price of rare metals from the feature data.

[0043] Furthermore, the feature data for training and inference may be at least one function of feature data 65 and feature data (for augmentation) 66.

[0044] Figure 6 is a flowchart illustrating an example of the inference method of this embodiment. Figures 7 and 8 are sequence diagrams showing examples of operations performed by the inference program of this embodiment. Figure 7 is a sequence diagram showing an example of operations for acquiring and preprocessing various data, and for generating and storing a trained model. Figure 8 is a sequence diagram showing an example of operations for inference using the trained model and determining the inference result. The inference program may be stored on a recording medium.

[0045] As shown in Figures 6 and 7, in step S1, the processor 501 of the computer 5 sends a command to store various data to the learning database server 4 (110), and the data storage unit 35 of the learning database server 4 executes the command to store various data (111), storing the various data input from the computer 5 or the input device 32 in the learning database 40. The various data to be stored include feature data 41 and spot price data 42 for rare metals.

[0046] The processor 501 of computer 5 sends various data acquisition and transmission commands to the learning database server 4 (112). In step S2, the data acquisition unit 36 ​​of the learning database server 4 executes various data acquisition and transmission commands (113), acquires various data from the learning database 40, and transmits the various data to computer 5 via interface 33.

[0047] In step S3, the computer 5 acquires various data using the data acquisition unit 502 and displays it on the display of the output device 525. In step S4, the selection unit 505 of the computer 5 selects a location for inputting training data, and the input device 523 inputs or modifies the training data (114). The determination / modification unit 506 then determines the input or modified training data, and the determined training data is transmitted to the training database server 4 via the interface 524.

[0048] In step S5, the processor 501 of the computer 5 sends a preprocessing instruction to the learning database server 4 (115), and the preprocessing unit 37 of the learning database server 4 performs preprocessing of various data (116).

[0049] In step S6, the preprocessor 37 executes an instruction to expand the training data (117) and expands the training data. The preprocessor 37 expands the training feature data 41 by associating first time data based on at least one of the training rare metal supply chain data 43, non-ferrous metal supply chain data 44, rare metal resource development funding data 45, non-ferrous metal resource development funding data 48, rare metal futures price data 49, non-ferrous metal futures price data 50, non-ferrous metal spot price data 51, economic indicator data 52, and exchange rate data 53 with second time data based on at least one of the training energy industry indicator data 54, automobile industry indicator data 55, aerospace industry indicator data 56, and defense industry indicator data 57.

[0050] Furthermore, the preprocessing unit 37 may augment the feature data 41 by classifying at least one of the items in the feature data 41 into multiple groups (or scales). For example, a threshold may be set, and the preprocessing unit 37 may augment the training data by classifying the feature data 65 or feature data (for augmentation) 66 into multiple groups (or scales) based on the threshold.

[0051] In step S7, the preprocessing unit 37 executes a filtering command for the training data (118) and performs filtering. The preprocessing unit 37 performs filtering in order to exclude some of the items of the feature data 41 from the training data.

[0052] In step S8, a trained model is generated. The processor 501 of computer 5 sends a machine learning model generation command to the artificial intelligence server 3 (120). The data acquisition unit 26 of the artificial intelligence server 3 sends a command to acquire training data to the training database server 4 via interface 24 (121). The training database server 4 executes a command to send training data and sends the training data stored in the training database 40 to the artificial intelligence server 3 via interface 33 (122). The data acquisition unit 26 acquires training data from the training database 40, and the machine learning unit 27 executes a command to generate a trained model using machine learning based on the training data (including the expanded training data) (123).

[0053] The machine learning unit 27 generates a trained model using a neural network based on training data. The machine learning unit 27 generates a trained model using a neural network that inputs feature data 41 into the input layer and rare metal spot price data 42 as the ground truth value into the output layer. The neural network's hidden layers include either affine layers or convolutional layers. Downsampling may be performed as appropriate. The number of layers, the number of neurons, and the activation function of the hidden layers are selected to be optimal so that the inference results are highly accurate. Feed Forward Neural Networks (FFNNs) and Recurrent Neural Networks (RNNs) are used as the neural network. The machine learning unit 27 uses parameters (weights) initialized with predetermined values ​​such as random numbers to calculate a loss function that represents the discrepancy between the value output to the output layer when feature data 41 is input to the input layer and the correct value of the output layer (spot market price data 42 of rare metals). Using the derivative of the loss function as the gradient, the unit generates a trained model by changing the parameters (weights) so that the discrepancy between the value output to the output layer and the correct value of the output layer becomes smaller.

[0054] As described above, there is a certain relationship between the feature data 41 and the spot price data 42 for rare metals. Therefore, by generating a trained model using machine learning based on training data that associates the feature data 41 and the spot price data 42 for rare metals, and then inferring the price data for rare metals for inference from the feature data for inference input into the trained model, it is possible to infer the price of rare metals with high accuracy.

[0055] In step S9, the trained model storage unit 28 stores the generated trained model in the memory device 22 by executing a trained model storage instruction (123).

[0056] In step S10, if the process proceeds to the step of inferring the price of rare metals, it proceeds to step S9; otherwise, the process terminates.

[0057] As shown in Figures 6 and 8, in step S11, the processor 501 of computer 5 receives feature data and time data for inference via the input device 523 and transmits the feature data and time data for inference to the artificial intelligence server 3 (129). Then, the processor 501 of computer 5 transmits an inference command for the price of rare metals to the artificial intelligence server 3 (130). The feature data for inference is then input to the trained model via the interface (transmitting / receiving unit) 24 of the artificial intelligence server 3.

[0058] In step S12, the inference unit 29 executes a data check command to verify the data format of the feature data and time data for inference sent from the computer 5 and determines whether or not the data format is correct (131). If the data format is incorrect, the inference unit 29 sends data to the computer 5 indicating that fact.

[0059] If the data format is correct, in step S13, the inference unit 29 executes a data preprocessing command and encodes the feature data and time data for inference according to the encoding policy used during training (131). If the feature data and time data for inference contain values ​​that were not used during training, the inference unit 29 either excludes those values ​​or assigns predetermined values ​​according to predetermined conditions.

[0060] In step S14, the inference unit 29 executes an inference command, reads the trained model stored in the memory device 22, inputs feature data and time data for inference into the trained model, and infers the price of rare metals at a predetermined time from the feature data and time data input into the trained model (132).

[0061] The inference unit 29 executes a command to send the inference result via the interface 24 and sends the inference result to the computer 5 (133). The interface (transmitting / receiving unit) 524 of the computer 5 receives the inference result of the rare metal market price from the trained model.

[0062] In step S15, the processor 501 of the computer 5 executes a command to display the transmission result and causes the inference result to be displayed on the output device 525 (for example, a display) (134).

[0063] In step S16, the input device 523 of the computer 5 specifies the determination of the price of rare metals, and the determination / correction unit 506 of the computer 5 executes the command to determine the price of rare metals (135). Furthermore, the determined price of rare metals can be corrected by the determination / correction unit 506 to the actual price of rare metals.

[0064] In step S17, the computer 5 determines whether the rare metal market price corrected by the decision / correction unit 506 has been associated with the feature data for inference and stored in the training data of the training database 40. If it has not been stored in the training data of the training database 40, the processor 501 of the computer 5 sends the data, which is the rare metal market price data corrected by the decision / correction unit 506 and associated with the feature data, to the training database server 4 (136). The data storage unit 35 of the training database server 4 executes a training data storage command (137) and stores the data sent from the computer 5 in the training database 40. As a result, the training data in the training database 40 increases, and the rare metal market price can be inferred with even greater accuracy.

[0065] As described above, according to this embodiment, the market price of rare metals can be inferred with high accuracy from feature data using machine learning.

[0066] Although embodiments of the present invention have been described above, the present invention is not limited to these, and can be modified or altered within the scope described in the claims. [Industrial applicability]

[0067] This invention is useful as an inference system that can accurately predict the market price of rare metals from feature data. Furthermore, by extending the feature data with energy industry indicator data, etc., it is useful as an inference system that can accurately predict the market price of rare metals. [Explanation of symbols]

[0068] 1…Inference System 2…Network 4…Learning database server 5… Computer 21… Processor 22...Storage device 23…Input device 24… Interface 25…Output device 26...Data acquisition unit 27…Machine Learning Department 28…Model storage 29… Reasoning part 30… Processor 32…Input device 33… Interface 34…Output device 35...Data storage unit 36...Data acquisition unit 37…Pre-treatment section 40…Learning database 501… Processor 502...Data acquisition unit 505...Selection section 506…Revision section 523...Input device 524… Interface 525...Output device

Claims

1. An inference system comprising a processor and a memory device, wherein the processor infers the market price of rare metals, The aforementioned processor, A machine learning unit generates a trained model by machine learning based on training data in which at least one feature data of the rare metal supply chain data, the supply chain data of non-ferrous metals other than the rare metal, the resource development funding data of the rare metal, the resource development funding data of non-ferrous metals other than the rare metal, the futures market price data of the rare metal, the futures market price data of non-ferrous metals other than the rare metal, the spot market price data of non-ferrous metals other than the rare metal, economic indicator data, and exchange rate data is associated with the spot market price data of the rare metal. A receiving unit that inputs feature data for inference into the trained model in order to infer the spot market price data of the rare metals, An inference system characterized by comprising the following features.

2. The aforementioned machine learning unit, As supply chain data for the rare metals, the following are included: producer data relating to at least one of the producers' receipts, consumption, shipments, inventory, and sales breakdowns of the rare metals; seller data relating to at least one of the sellers' receipts, shipments, inventory, and sales breakdowns of the rare metals; consumer data relating to at least one of the consumers' self-generation, self-production, receipt, consumption, shipments, inventory, and consumption breakdowns of the rare metals; air cargo data for the rare metals; and at least one port cargo data for the rare metals. As supply chain data for non-ferrous metals other than the rare metals, the following are included: producer data relating to at least one of the receipts, consumption, shipments, inventory, and sales breakdowns of producers of non-ferrous metals other than the rare metals; seller data relating to at least one of the receipts, shipments, inventory, and sales breakdowns of sellers of non-ferrous metals other than the rare metals; consumer data relating to at least one of the self-generation, self-production, receipt, consumption, shipments, inventory, and consumption breakdowns of consumers of non-ferrous metals other than the rare metals; air cargo data for non-ferrous metals other than the rare metals; and at least one port cargo data for non-ferrous metals other than the rare metals. As data on funding for the resource development of the rare metals, the data on funding provided to at least one of the producers, sellers, and consumers of the rare metals, and the data on funding provided to at least one of the buyers who acquire the producers, sellers, and consumers of the rare metals, As data on funding for resource development of non-ferrous metals other than the rare metals, the data on funding provided to at least one of the producers, distributors, and consumers of the non-ferrous metals other than the rare metals, and the data on funding provided to at least one of the acquirers who acquire at least one of the producers, distributors, and consumers of the non-ferrous metals other than the rare metals, As the aforementioned economic indicator data, at least one of the following is used: GDP, GDP per capita, economic growth rate, unemployment rate, inflation rate, producer price index, purchasing power parity, balance of payments, trade balance, interest rates, government bond yields, government debt, fiscal balance, labor productivity, retail sales, manufacturing production index, construction activity index, housing starts, industrial production index, automobile sales, consumer confidence index, business confidence index, inventory index, monetary reserves, tourism revenue, income inequality (Gini coefficient), population growth rate, labor force participation rate, average wage growth rate, import / export price index, and food price index. The inference system according to claim 1, characterized in that it generates the trained model by machine learning based on training data in which the feature data and the spot market price data of the rare metals are associated.

3. The inference system according to claim 1, characterized in that the machine learning unit generates the trained model by machine learning based on training data in which at least one feature data of energy industry indicator data, automobile industry indicator data, aerospace industry indicator data, and defense industry indicator data is associated with the spot market price data of the rare metals.

4. The aforementioned machine learning unit, As energy industry indicator data, at least one of the following will be used: crude oil price, natural gas price, natural gas futures price, coal price, coal futures price, energy consumption, crude oil production, crude oil inventory, solar power generation capacity, wind power generation capacity, power generation, electricity supply, renewable energy share, energy price, energy efficiency index, power plant CO2 emissions, and wholesale electricity price. As the aforementioned automotive industry indicator data, at least one of the following is used: number of automobiles produced, number of automobiles sold, market share of automobile sales by country, global market share of automobile sales, number of automobiles exported, number of automobiles imported, number of electric vehicles sold, number of hybrid vehicles sold, adoption rate of autonomous vehicles, average fuel consumption of automobiles, automobile CO2 emissions, number of employees in the automotive industry, contribution rate of the automotive industry to GDP, number of automobile recalls, total amount of automobile loans, used car market size, and number of registered automobiles. As aerospace industry indicator data, at least one of the following is used: aircraft production volume, aircraft orders, aircraft deliveries, aircraft utilization rate, number of flights, aircraft passenger kilometers, number of aircraft passengers, air cargo volume, airline revenue, airline operating costs, aircraft fuel consumption, average airfare price, aircraft CO2 emissions, GDP contribution of the aviation industry, number of new routes, number of rocket launches, number of satellite launches, number of satellites, and number of spacecraft development projects. As the aforementioned military industry indicator data, at least one of the following is included: military budget size, defense expenditure, arms exports, arms imports, number of weapons produced, military enterprise revenue, number of military industry employees, contribution of the defense industry to GDP, number of missile defense systems deployed, and number of military satellite launches. The inference system according to claim 3, characterized in that it generates the trained model by machine learning based on training data in which the feature data and the spot market price data of the rare metals are associated.

5. The aforementioned machine learning unit, A first time data based on at least one of the following: the supply chain data for the rare metals used for learning, the supply chain data for non-ferrous metals other than the rare metals, the resource development funding data for the rare metals, the resource development funding data for non-ferrous metals other than the rare metals, the futures market data for the rare metals, the futures market data for non-ferrous metals other than the rare metals, the spot market data for non-ferrous metals other than the rare metals, the economic indicator data, and the exchange rate data, A second time data based on at least one of the aforementioned learning energy industry indicator data, automotive industry indicator data, aerospace industry indicator data, and defense industry indicator data, The inference system according to claim 3, characterized in that it generates the trained model by machine learning based on the training feature data which has been expanded by associating the features.

6. An inference system comprising a processor and a memory device, wherein the processor infers the market price of rare metals, The aforementioned processor, A transmission unit transmits inference feature data to a trained model generated by machine learning, based on training data in which at least one feature data of the rare metal supply chain data, non-ferrous metal supply chain data other than the rare metal, resource development funding data for the rare metal, resource development funding data other than the rare metal, futures price data for the rare metal, futures price data for non-ferrous metals other than the rare metal, spot price data for non-ferrous metals other than the rare metal, economic indicator data, and exchange rate data is associated with the spot price data for the rare metal. A receiving unit that inputs the inference result of the price of the rare metal from the trained model, An inference system characterized by comprising the following features.

7. A method for a processor to infer the market price of rare metals, A step of generating a trained model by machine learning based on training data in which at least one feature data of the rare metal supply chain data, the supply chain data of non-ferrous metals other than the rare metal, the resource development funding data of the rare metal, the resource development funding data of non-ferrous metals other than the rare metal, the futures market price data of the rare metal, the futures market price data of non-ferrous metals other than the rare metal, the spot market price data of non-ferrous metals other than the rare metal, economic indicator data, and exchange rate data is associated with the spot market price data of the rare metal, The steps include inputting feature data for inference into the trained model in order to infer the spot market price data of the rare metals, An inference method characterized by comprising:

8. A method for a processor to infer the market price of rare metals, The steps include sending inference feature data to a trained model generated by machine learning, based on training data in which at least one feature data of the rare metal supply chain data, non-ferrous metal supply chain data other than the rare metal, resource development funding data for the rare metal, resource development funding data for non-ferrous metals other than the rare metal, futures price data for the rare metal, futures price data for non-ferrous metals other than the rare metal, spot price data for non-ferrous metals other than the rare metal, economic indicator data, and exchange rate data is associated with the spot price data for the rare metal, The steps include inputting the inference result of the price of the rare metal from the trained model, An inference method characterized by comprising:

9. A processor is an inference program that runs on a computer to infer the market price of rare metals, The aforementioned computer, A machine learning function that generates a trained model by machine learning based on training data in which at least one feature data of the rare metal supply chain data, the supply chain data of non-ferrous metals other than the rare metals, the resource development funding data of the rare metals, the resource development funding data of non-ferrous metals other than the rare metals, the futures market price data of the rare metals, the futures market price data of non-ferrous metals other than the rare metals, the spot market price data of non-ferrous metals other than the rare metals, economic indicator data, and exchange rate data is associated with the spot market price data of the rare metals. An input function for inputting feature data for inference into the pre-trained model in order to infer the spot market price data of the aforementioned rare metals, An inference program characterized by achieving this.

10. A processor is an inference program that runs on a computer to infer the market price of rare metals, The aforementioned computer, A transmission function that transmits inference feature data to a trained model generated by machine learning, based on training data in which at least one feature data of the rare metal supply chain data, non-ferrous metal supply chain data other than the rare metal, resource development funding data for the rare metal, resource development funding data other than the rare metal, futures price data for the rare metal, futures price data for non-ferrous metals other than the rare metal, spot price data for non-ferrous metals other than the rare metal, economic indicator data, and exchange rate data is associated with the spot price data for the rare metal; An input function for inputting the inference result of the price of the rare metal from the aforementioned trained model, An inference program characterized by achieving this.

Images