Determining resource requirements

Abstract

A method (200) for determining future resource requirements at a given location includes obtaining (110) transaction data, the transaction data indicating past resource usage at the given location. At least one model may be trained (120) based on the set of parameters by determining (121) a distribution associated with the transaction data and generating (122) a kernel based on the distribution and the set of parameters, the kernel being configured to output an estimated distribution. Based on a comparison between the distribution and the estimated distribution, the kernel is refined (123) and validated (124). A density map is then generated (125) based on the at least one trained model. The density map is then transmitted (130) to a control system to determine future resource requirements at the given location.

Description

[Technical Field]

[0001] METHODS, APPARATUS AND SYSTEMS FOR DETERMINING RESOURCE REQUIREMENTS FIELD OF THE INVENTION The present disclosure relates particularly, but not exclusively, to determining future resource requirements at a given location. [Background technology]

[0002] In transportation and logistics, such as delivery from a shipper to a consignee, locating resources is time-consuming and costly. Therefore, it is desirable to achieve cost-effective and time-efficient transit by strategically positioning resources, such as delivery vehicles, so that they are in the right place at the right time.

[0003] Being able to locate such resources is challenging and depends on several factors, including, but not limited to, weather conditions and seasonal events. It is therefore desirable to improve the efficiency and reduce the cost of determining resource locations while taking such factors into account. Summary of the Invention

[0004] According to a first aspect of the present invention, there is provided a method for determining future resource requirements at a given location, the method comprising: obtaining transaction data, the transaction data indicating past resource usage at the given location; determining a distribution associated with the transaction data; generating a kernel based on the distribution and a set of parameters, the kernel being configured to output an estimated distribution; refining and validating the kernel based on a comparison between the distribution and the estimated distribution; training at least one model based on the set of parameters; generating a density map based on the at least one trained model; and transmitting the density map to a control system for determining future resource requirements at the given location. This enables the control system to indicate more efficient locations for placing resources based on the density map.

[0005] Preferably, the step of determining the distribution includes the steps of extracting patterns from the transaction data and analysing the patterns from the transaction data to determine the distribution, which allows patterns in the transaction data to be used to determine the distribution and thereby assist in generating a more accurate density map.

[0006] The method may further include normalizing the transaction data, and the step of determining the distribution is associated with the normalized transaction data. This allows the data to be adjusted and a kernel to be generated that takes into account particularly busy locations so that they do not adversely affect less busy locations.

[0007] Preferably, the step of generating the density map includes determining a performance characteristic for each of a plurality of trained models and selecting at least one of the trained models based on the performance characteristic, which allows a plurality of different models to be trained and the most accurate one to be selected.

[0008] The step of generating the density map based on at least one trained model may include combining multiple trained models to generate a composite model, and generating the density map based on the composite model. This allows multiple models to be generated and the density map to be based on a combination of the models, so that, for example, different models can be used to generate different areas of the density map.

[0009] The set of parameters may include at least one of a date, a time, a public holiday, and weather conditions associated with the plurality of locations, which allows factors affecting a particular time period within a given time period to be taken into account when determining the density map.

[0010] Preferably, a given location represents a predefined area, which can be at least one of a building address, a zip / postal code, a street, a city district, a city, a county, a state, and a country. This allows density maps to be generated for multiple different locations at different levels to provide the most accurate density map for each of the different predefined areas.

[0011] Preferably, the method further comprises outputting the density map to at least one of a display of the control system and a resource of the control system, which enables the density map to be used by the system to position the resource at a desired location.

[0012] Optionally, the control system includes a vehicle, which allows the density map to be used to position a vehicle, such as an autonomous car.

[0013] According to a second aspect of the present invention, there is provided an apparatus for determining future resource requirements at a given location, the apparatus being adapted to carry out the method according to the first aspect.

[0014] According to a third aspect of the present invention there is provided a control system comprising the apparatus of the second aspect and an output device for receiving a density map output by the apparatus.

[0015] The system may further include a storage device for storing at least transaction data indicative of past resource usage, which allows the system to have efficient access to various transaction data and other data, such as sets of parameters, to quickly and efficiently generate density maps.

[0016] The system may further include an input interface configured to allow a user of the system to manually adjust the density map.

[0017] According to a fourth aspect of the present invention, there is provided a computer-readable storage medium having stored thereon instructions which, when executed by a processor, cause the processor to obtain transaction data, the transaction data indicative of past resource usage at a given location, determine a distribution associated with the transaction data, generate a kernel based on the distribution and a set of parameters, the kernel being configured to output an estimated distribution, refine and validate the kernel based on a comparison between the distribution and the estimated distribution, thereby causing the processor to train at least one model based on the set of parameters, generate a density map based on the at least one trained model, and transmit the density map to a control system for determining future resource requirements.

[0018] Further features and advantages of the present invention will become apparent from the following description of preferred embodiments of the invention, given by way of example with reference to the accompanying drawings, in which like reference symbols are used to represent like features, and in which: [Brief explanation of the drawings]

[0019] [Figure 1] 3 is a flowchart illustrating a method according to a first embodiment. [Figure 2] 10 is a flowchart illustrating a method according to a second embodiment. [Figure 3] 1 shows a schematic diagram of an apparatus according to a first embodiment; [Figure 4] 2 shows a schematic diagram of an apparatus according to a second embodiment; [Figure 5] 3 shows a schematic representation of a density map produced by the method of FIG. 1 or 2. [Figure 6] 1 illustrates a schematic diagram of a system according to an embodiment; DETAILED DESCRIPTION OF THE INVENTION

[0020] Details of methods, apparatus, and systems according to embodiments will become apparent from the following description, taken in conjunction with the drawings. For purposes of explanation, numerous specific details of particular embodiments are set forth in this description. References in the specification to "an embodiment" or similar language mean that a feature, structure, or characteristic described in connection with an embodiment is included in at least one embodiment, but not necessarily in other embodiments. Furthermore, it should be noted that particular embodiments are generally described, and that certain features have been omitted and / or necessarily simplified to facilitate explanation and understanding of the concepts of the embodiments.

[0021] Determining the location of areas that can result in cost-effective and time-efficient management of resources, such as when shipping merchandise, identifying medical supplies, or identifying vehicles for use in ride-sharing or shared resources such as autonomous vehicles, can be hindered by a lack of understanding of specific patterns evident in data representing the positioning of resources at prior times. Furthermore, even if one pattern of placement of a particular resource is valid during a particular time, external factors may affect the likelihood of the placement suggested by the pattern being inaccurate during another time, or even for another location.

[0022] It is therefore an object of the present invention to provide a method, apparatus and system for determining optimal locations for placement of resources in a given area.

[0023] FIG. 1 is a flowchart illustrating a method 100 according to an embodiment. At item 110, transaction data is obtained. The transaction data represents past resource usage at a given location and, in some embodiments, may be retrieved from memory or, in other embodiments, may be entered by a user of the system. The transaction data represents previous resource usage for a given location, such as data representing the locations of shipments sent and received within a particular time period for a particular location. A particular location may be a given country, a given county / state, a city, a street, a zip / postal code area, or a particular building address. Thus, for a particular location, the transaction data provides an overview of what resource requirements a particular location had within a time period. Data for a given shipment in the transaction data may include any of a shipment identifier, package dimensions, departure and arrival times, recipient address information, and return address information. It will be appreciated that individual transactions within the transaction data may include other fields or may include only some of the fields listed above.

[0024] Once the transaction data is obtained, method 100 proceeds to item 120, where the transaction data is used to train a model to determine a density map. The model is trained to generate a density map to illustrate future resource requirements for a given location. Training the model includes determining a distribution associated with the transaction data in item 121. The distribution indicates resource requirements for particular locations within the area covered by the transaction data. For example, if the transaction data includes data about resource requirements (such as sent and received shipments) in a given city, the distribution can provide an indication of the density of resources in particular streets, buildings, or towns within that city over a given period of time.

[0025] Determining the distribution (121) includes extracting patterns from the transaction data. For example, if the transaction data covers a particular period, such as a month, extracting the pattern may include, for example, determining that there is a spike in resource usage every Tuesday. This is just an example of such a pattern; it will be appreciated that any number of other patterns may be extracted and the transaction data may cover a longer or shorter period of time. To determine the distribution, the patterns extracted from the transaction data may be analyzed, and the distribution may be determined. The extracted patterns may then be analyzed to determine the distribution. In some embodiments, the transaction data may be normalized, which ensures that differences between locations in the transaction data are adjusted for each other. Normalization may be performed within the training model item 120, or alternatively, prior to training the model 120, such as when multiple models are to be trained as described below in connection with FIG. 2. For example, if location A has a large number of transactions, the data associated with location A may be normalized to a high transaction threshold, and if location B has a smaller number of transactions, the data associated with location B may be normalized to a lower transaction threshold. This allows the data to be adjusted and kernels to be generated that take into account particularly busy locations. In the shipment delivery example, if a particular logistics company has a hub, many transactions will be included in the data listing the hub's location, which can adversely affect the kernel generation process; therefore, normalizing the distribution generated from the transaction data can account for this unusual location.

[0026] In item 122, the distribution is used to generate a kernel that outputs an estimated distribution. The kernel is a machine learning kernel for use in a classifier, such as a support vector machine, that is used to determine the similarity between two sets of data, for example, transaction data, or patterns therein, and an estimated distribution to be generated that is used to generate a density map representing future resource requirements. The kernel is generated based not only on the distribution of the transaction data, but also on a set of one or more parameters.

[0027] One or more parameters represent information that can alter a distribution generated from transaction data, for example, if the transaction data represents a particular time period, such as over the Easter period of one year, then in another year, the Easter period may not be within the same period, and therefore, resource requirements over that period may be lower than would be predicted if future resource requirements were based on data representing the Easter period of the previous year. Thus, the set of parameters may include time of day information, holiday information, and even seasonal information, such as weather forecasts for particular dates. In some embodiments, the set of parameters may include a historical density map for a given location representing resource usage over different time periods. It will be appreciated that several other parameters may be used to ensure that the distribution more accurately reflects the time period for which future resource requirements will be determined. Furthermore, the parameters may vary based on a given location; for example, the set of holidays in the United States is different from the set of holidays in the United Kingdom; if resources are transported across borders, it may be necessary to consider, for example, the weather in both locations and the holidays in both locations.

[0028] Following generation of the kernel, the kernel is refined in item 123. That is, weighting factors associated with the kernel are adjusted so that the output of the kernel (the estimated distribution) more accurately represents the distribution determined in item 121. This is achieved by comparing the estimated distribution and the distribution determined in item 121. In some embodiments, refining the kernel may include selecting a different type of kernel, such as selecting one of a Gaussian kernel, a top-hat kernel, or an Epanechnikov kernel, if they more accurately reflect the actual slope of the distribution associated with the transaction data.

[0029] The output of the kernel is validated in item 124 to determine the accuracy of the kernel. Validating the kernel may include generating an output from the kernel using test data representing transaction data (e.g., a similar period within the previous year) to determine whether the output of the kernel is substantially similar to the distribution determined from the transaction data. It will be appreciated that other methods of validating the output of the kernel may be used. If the output of the kernel is substantially similar, the distribution is based on the transaction data generated in item 121, and the method continues with item 125, where a density map representing expected future resource usage is generated. If the output of the kernel is not substantially similar, the method returns to item 123, where the kernel is further refined, and the output of the refined kernel is again validated in item 124. Items 123 and 124 then loop until the kernel is validated and its output is substantially similar to the expected distribution.

[0030] As mentioned above, in item 125, a density map is generated. The density map represents areas where resources are needed at a given future time. The density map also provides an indication of the number of resources needed, for example in the form of a heat map, where greater resource requirements are indicated by higher density and areas with lower resource requirements are indicated by lower density, as explained below with reference to FIG. 5.

[0031] Following generation of the density map, the density map is transmitted to a control system. The control system may include a display or other resource for receiving the density map. For example, a display may be used to show the density map to a user of the system so that they can configure resources to be allocated accordingly, and / or the density map may be transmitted to another resource of the control system, such as a memory and / or the vehicle itself, such as a delivery truck. In some examples, if the vehicle is an autonomous vehicle, the density map may enable the vehicle to navigate to a location needed at a future time based on the resource requirements set in the density map.

[0032] FIG. 2 is a flowchart illustrating a method 200 according to a second embodiment. Method 200 includes several steps 110, 120, and 130 of the embodiment described above in connection with FIG. 1. Furthermore, train model item 120 also includes identical steps 121, 122, 123, and 124, except for step 125 of generating a density map. Following training of the first model in item 120, the model is stored and method 200 proceeds to item 210. In item 210, it is determined whether more than one trained model should be generated. For example, the second trained model may be based on a different type of kernel and / or a different set of parameters. If it is determined that a further model should be trained, method 200 returns to item 120 where the further model is trained.

[0033] Conversely, if it is determined that no further models are needed, method 200 proceeds to item 220 where models are combined. In item 210, multiple models with different parameters and / or kernel types are combined to produce a single ensemble / composite model with a high level of accuracy and minimize error. This can be achieved by using primal voting techniques across each of the models on their own or in any number of combinations.

[0034] After the ensemble model is generated by combining the models in item 220, the method 200 proceeds to item 230 where a density map is generated from the ensemble model. This density map is then transmitted to the control system in item 130.

[0035] In still further embodiments, multiple models may be trained as described above in connection with FIG. 2 , and each model may be trained on test data or via another suitable method to determine performance characteristics. The performance characteristics may indicate the accuracy and / or efficiency of each model. Based on the performance characteristics, one of the trained models having the best performance characteristics may be selected. Thus, the selected model provides the strongest results representing highly accurate forecasts of future resource requirements without the need to combine multiple models, thereby minimizing the processing required to store multiple models and reducing the requirements for storing multiple models.

[0036] Figure 3 illustrates a schematic diagram of an apparatus 300 according to an embodiment. The apparatus 300 is configured to perform the methods 100, 200 described above in relation to Figures 1 and 2. The apparatus 300 includes at least one processor 310 for performing the foregoing methods, where in some embodiments the method 200 requires training multiple models, and the apparatus may include multiple processors 310, as described below in relation to Figure 4.

[0037] The processor 310 may be a central processing unit (CPU), a neural network accelerator, or a neural processing unit (NPU), an image signal processor (ISP), or a graphics processing unit (GPU). It will be appreciated that the processor 310 may be any type of processor configured to perform the methods 100, 200 described above in connection with FIGS. 1 and 2.

[0038] The device 300 is configured to receive input data 320 in the form of transaction data. The transaction data represents past resource usage at a given location and, in some embodiments, may be retrieved from memory or, in other embodiments, may be entered by a user of the system. The transaction data represents previous resource usage for a given location, such as data representing the locations of shipments sent and received within a specific time period for a particular location. A specific location may be a given country, a given county / state, a city, a street, a zip / postal code area, or a specific building address. Thus, for a particular location, the transaction data provides an overview of what resource requirements for a particular location were within a time period. Data for a given shipment in the transaction data may include any of a shipment identifier, package dimensions, departure and arrival times, recipient address information, and return address information. It will be appreciated that individual transactions in the transaction data may include other fields or may include only some of the fields listed above.

[0039] Input data 320 is provided to an input module 330 of the processor 310, which may be configured to buffer or otherwise temporarily store the input data in processor internal memory, which may be volatile or "on-chip" memory, such as synchronous dynamic random access memory (SDRAM) or double data rate synchronous dynamic random access memory (DDR-SDRAM).

[0040] The processor 310 also includes a training module 340 that trains at least one model to generate a density map representing resource requirements for a given location. The training module 340 itself includes multiple modules 341, 342, 343, and 344. The input data 320 is first passed to a determination module 341 that determines a distribution associated with the input data 320. The determination module 341 may extract patterns from the transaction data, for example, patterns covering a particular time period. The extracted patterns may then be analyzed to determine the distribution.

[0041] The training module 340 further includes a kernel generation module 342 that generates kernels for outputting the estimated distribution. The kernels are machine learning kernels for use in a classifier, such as a support vector machine, that are used to determine the similarity between two sets of data, for example, between transaction data, or patterns therein, and a density map to be generated that represents future resource requirements. The kernels are generated based not only on the distribution of the transaction data, but also on a set of one or more parameters 350, which in some embodiments may be retrieved from external memory as described below with reference to FIG. 6.

[0042] One or more parameters 350 represent information that can alter the distribution generated from the transaction data, for example, if the transaction data represents a particular time period, such as through the Easter period of one year, then in another year, the Easter period may not be within the same time period, and therefore, resource requirements through that period may be lower than would be predicted if future resource requirements were based on data representing the Easter period of the previous year. Thus, the set of parameters 350 may include time of day information, holiday information, and even seasonal information, such as weather forecasts for particular dates. In some embodiments, the set of parameters may include a historical density map for a given location representing resource usage through different time periods. It will be appreciated that several other parameters may be used to ensure that the distribution more accurately reflects the time period for which future resource requirements will be determined. Furthermore, the parameters may vary based on a given location; for example, the set of holidays in the United States is different from the set of holidays in the United Kingdom; for example, if resources are transported across borders, it may be necessary to consider the weather in both locations and the holidays in both locations.

[0043] The training module 340 also includes a refinement module 343 for refining the kernels generated by the kernel generation module 342. The refinement module 343 may be configured to adjust weighting factors associated with the kernels so that the output of the kernels (the estimated distributions) more accurately represents the distributions determined by the determination module 341. This is accomplished by comparing the estimated distributions and the distributions determined by the determination module 341. In some embodiments, refining the kernels may include selecting different types of kernels, such as selecting one of a Gaussian kernel, a top-hat kernel, or an Epanechnikov kernel, if they more accurately reflect the actual slope of the distributions associated with the transaction data 320.

[0044] To determine the accuracy of any given kernel, the training module 340 further includes a validation module 344. The validation module 344 may include generating an output from the kernel using test data representing the input data 320 (e.g., a similar period within the previous year) to determine whether the kernel's output is substantially similar to the distribution determined by the distribution module 341. It will be appreciated that other methods of validating the kernel's output may be used. If the kernel's output is substantially similar, the distribution generated by the distribution module 341, the trained model, is used by the processor 310 to generate a density map using the generation module 360. If the kernel's output is not substantially similar, the refinement module 343 may be used to further refine the kernel, which is then validated again using the validation module 34. This process may be repeated until the kernel is validated and its output is substantially similar to the expected distribution.

[0045] The density map generated by generation module 360 ​​represents areas where resources are needed at a given future time. The density map also provides an indication of the number of resources needed, for example in the form of a heat map, where areas with greater resource requirements are indicated by a higher density and areas with lower resource requirements are indicated by a lower density, as explained below with reference to FIG.

[0046] Following generation of the density map by generation module 360, the density map is transmitted (370) to a control system (not shown). The control system may include a display or other resource for receiving the density map. For example, a display may be used to show the density map to users of the system so that they can be configured to allocate resources accordingly. Additionally and / or alternatively, the density map may be transmitted to another resource of the control system, such as memory and / or the vehicle itself, such as a delivery vehicle or ride-sharing vehicle. In some examples, if the vehicle is an autonomous vehicle, the density map may enable the vehicle to navigate to a location where it is needed at a future time based on the resource requirements set in the density map.

[0047] FIG. 4 illustrates a schematic diagram of an apparatus 400 according to a second embodiment. The apparatus 400 is configured to receive input data 320 in the form of transaction data. The transaction data represents past resource usage at a given location and, in some embodiments, may be retrieved from memory or, in other embodiments, may be entered by a user of the system. In this embodiment, the input data 320 may be normalized by a normalization module 410. This ensures that differences between locations in the transaction data are adjusted for each other. For example, if location A has a high number of transactions, the data associated with location A may be normalized relative to a high transaction threshold, and if location B has a lower number of transactions, the data associated with location B may be normalized relative to a lower transaction threshold. This allows the data to be adjusted, allowing kernels to be generated that take into account particularly busy locations. In a shipment delivery example, if a particular delivery company has a hub, a large number of transactions may be included in the data listing the hub's locations, which may adversely affect the kernel generation process. Therefore, normalizing the distribution generated from the transaction data may account for these anomalous locations.

[0048] The apparatus 400 may also include multiple processors 310a, 310b, 310c, such as the processor 310 described above in connection with FIG. 4. Each of the multiple processors 310a, 310b, 310c may be configured to train a different model, e.g., a model based on a different kernel, or alternatively, they may be configured to train a model using a different set of parameters. Each of the processors 310a, 310b, 310c outputs a different trained model, which is then combined by an optimization module 420 configured to generate a density map based on the multiple trained models output by the processors 310a, 310b, 310c.

[0049] The optimization module 420 may be configured to generate performance characteristics that indicate the accuracy and / or efficiency of each model. Based on the performance characteristics, the trained model with the best performance characteristics may be selected. The selected model therefore provides the strongest results in outputting highly accurate forecasts of future resource requirements without the need to combine multiple models, thereby minimizing the processing required. The selected model is then used to generate a density map that is output 320 to a control system (not shown).

[0050] Alternatively, an optimization module 420 may be used to combine multiple trained models to produce a single ensemble / composite model with a high level of accuracy and minimize error. This can be achieved by using primacy voting techniques across each of the models, either on their own or in any number of combinations. The ensemble model may then be used to generate a density map that is output 320 to a control system (not shown).

[0051] Figure 5 shows schematically a density map 500 generated by the methods 100, 200 described above in relation to Figures 1 or 2. The density map 500 represents, in the form of a heat map, an area 510 and resource requirements 520, 530 for locations within the area 510. It will be appreciated that the density map 500 may represent the resource requirements for a given area 510 in any other manner suitable for showing relative resource requirements.

[0052] In Figure 5, density map 500 shows the United Kingdom 510 and a trained model or models that indicate that specific resources are needed. In some embodiments, each of the locations may be based on different transaction data and / or trained models, such that multiple individual density maps for each location are generated using the methods 100, 200 described above in Figures 1 and 2. Those individual density maps may be combined to produce a composite density map 500 that covers multiple locations.

[0053] Each location 520, 530 on the density map 500 is shown based on the density of the resource required. The density of the resource is determined by the model trained in methods 100, 200 and provides an indication of areas of high demand for resources and areas of low demand for resources. This may be indicated by separated areas on the density map as shown in Figure 5, or alternatively, may be indicated using different colors, with the colors blending to indicate changes in resource requirements.

[0054] For example, areas 520a, 530a with high resource requirements may be depicted on the map with solid lines or lines of a particular color, as shown in density map 500 of FIG. 5. Alternatively, the areas may be overlaid with a color indicating high resource requirements. It will be appreciated that a combination of these different ways of indicating density may be used, as may other ways of indicating resource requirements. Similarly, areas 520b, 530b with low resource requirements may be indicated with thinner lines, lines of a different color, or overlaid with a color indicating lower resource requirements. The different indicators used may be detailed in a legend or key for user understanding.

[0055] In some examples, it may be desirable for density map 500 to be manipulated or altered by a user to obtain a more detailed view of a particular location. In such examples, density map 500 is provided as part of a user interface to a computer program that includes zoom functionality 540. The user interface may provide other controls, such as panning controls, that allow a user to navigate around the density map to view a particular area of ​​interest.

[0056] As a user zooms in and out of density map 500, density map 500 may be updated to provide a more detailed view of the area of ​​interest, such as at the county / state level, city level, or even street level. It will be appreciated that in some embodiments, if density map 500 includes such information, it may be provided directly to the user, while in other embodiments, a further density map may need to be generated based on transaction data for new locations, new sets of parameters, and / or different models or kernels.

[0057] 6 illustrates a schematic diagram of a system 600 according to an embodiment. The system may form part of a computer terminal used to allocate resources to a given location, and in some embodiments may form part of a vehicle capable of reviewing information in a density map and navigating to a required location. System 600 includes an input device 610, such as devices 300, 400 described above in connection with FIGS. 3 and 4, and an output device 620. The output device may be a display for presenting the density map to a user, or, as mentioned above, a vehicle navigation system that analyzes the density map and determines a location to which the vehicle should navigate, either automatically or with the assistance of driver input.

[0058] The input device 610 may be a user interaction device, such as a mouse, keyboard, or touch screen element, associated with a display or other output device, that can be used by a user to input transaction data, a set of parameters, or even adjust the density map generated by the apparatus 300, 400.

[0059] In some embodiments, system 600 may include external storage 630 for storing a set of parameters, such as a density map for a given location and time period, previous density maps, and information about holidays. Storage 630 is accessed via memory controller 640. Storage 630 may also be configured to store other information for use by system 600, such as computer programs for determining the density map and / or providing a user interface for viewing or editing the density map.

[0060] The memory controller 640 may include a dynamic memory controller (DMC). The memory controller 640 is coupled to the storage device 630. The memory controller 640 is configured to manage the flow of data to and from the storage device 630. The storage device 630 may include main memory, otherwise referred to as “primary memory.” The storage device 630 may be external storage, in that the storage device 630 is external to the system 600. For example, the storage device 630 may include “off-chip” memory. The storage device 630 may have a larger storage capacity than the memory cache(s) of the device 300, 400. In some embodiments, the storage device 630 is included in the system 600. For example, the storage device 630 may include “on-chip” memory. The storage device 630 may include, for example, a magnetic or optical disk and a disk drive or a solid-state drive (SSD). In some embodiments, the storage device 630 includes synchronous dynamic random access memory (SDRAM). For example, the storage device 630 may include double data rate synchronous dynamic random access memory (DDR-SDRAM).

[0061] One or more of the input device(s) 610, the devices 300, 400, the output device(s) 620, and the memory controller 640 may be interconnected using, for example, a system bus 650, which allows data to be transferred between the various components. The system bus 650 may include or include any suitable interface or bus.

[0062] The above-described embodiments are to be understood as illustrative examples of the present invention. Further embodiments of the present invention are contemplated. It will be understood that any feature described in connection with any one embodiment may be used alone, in combination with other features described, in combination with one or more other features of any of the embodiments, or in any other combination of any of the embodiments. Furthermore, equivalents or modifications not described above may also be employed without departing from the scope of the present invention, as defined by the appended claims. [Aspect 1] 1. A method for determining future resource requirements at a given location, comprising: obtaining transaction data, the transaction data indicating past resource usage at a given location; determining a distribution associated with the transaction data; generating a kernel based on the set of parameters, the kernel configured to output an estimated distribution; refining and validating the kernel based on a comparison between the distribution and the estimated distribution; training at least one model based on said set of parameters by generating a density map based on the at least one trained model; and transmitting the density map to a control system for determining future resource requirements at the given location; A method comprising the steps of: [Aspect 2] The step of determining the distribution comprises: extracting patterns from the transaction data; analyzing the patterns from the transaction data to determine the distribution; 2. The method of embodiment 1, comprising the steps of: [Aspect 3] 3. The method of claim 1 or 2, further comprising: normalizing the transaction data; and determining a distribution is associated with the normalized transaction data. [Aspect 4] The step of generating the density map comprises: determining performance characteristics for each of the plurality of trained models; selecting at least one of the trained models based on the performance characteristics; and 4. The method of any one of aspects 1 to 3, comprising: [Aspect 5] generating the density map based on the at least one trained model, combining multiple trained models to generate a composite model; and generating the density map based on the composite model; 5. The method of any one of aspects 1 to 4, comprising: [Aspect 6] The set of parameters is: Date and Time, Holidays and weather conditions associated with the plurality of locations; 3. The method of embodiment 1 or 2, comprising at least one of: [Aspect 7] a display of the control system; a resource of the control system; 7. The method of any one of aspects 1-6, further comprising outputting the density map to at least one of: [Aspect 8] 1. An apparatus for determining future resource requirements at a given location, comprising: at least one processor; An apparatus configured to perform the method of any one of aspects 1 to 7. [Aspect 9] an apparatus according to embodiment 8; and an output device for receiving the density map output by the apparatus; A control system comprising: [Aspect 10] A computer-readable storage medium having stored thereon instructions that, when executed by a processor, cause the processor to: obtaining transaction data, the transaction data indicating past resource usage at a given location; determining a distribution associated with the transaction data; generating a kernel based on the set of parameters, the kernel configured to output an estimated distribution; refining and validating the kernel based on a comparison between the distribution and the estimated distribution; training at least one model based on said set of parameters by generating a density map based on the at least one trained model; transmitting said density map to a control system for determining future resource requirements; A computer-readable storage medium.

Claims

1. 1. A method for determining future resource requirements at a given location, the method being executed by a processor and comprising: obtaining transaction data, the transaction data indicating past resource usage at a plurality of locations, including the given location; determining a distribution of resources associated with the transaction data; generating a machine learning kernel based on a set of parameters including a historical density map for the given location representing the resource usage over different time periods, the machine learning kernel configured to output an estimated distribution of the resource across the plurality of locations; refining and validating the machine learning kernel based on a comparison between the distribution of the resources and the estimated distribution; training at least one machine learning model based on the set of parameters by if the estimated distribution is similar to the distribution of the resource, generating a density map of the resource at the plurality of locations using at least one trained machine learning model; transmitting the density map to a control system for determining future resource requirements at the given location; A method comprising the steps of: The set of parameters is: A specific date and A specific period of time, and Holidays and weather conditions associated with the plurality of locations; further comprising at least one of method.

2. The step of determining a distribution of resources comprises: extracting patterns from the transaction data; analyzing the patterns from the transaction data to determine a distribution of the resources; The method includes the steps of: Patterns in the transaction data are used to determine the distribution to assist in generating a more accurate density map. The method of claim 1.

3. The method of claim 1 or 2, further comprising the step of normalizing the transaction data, and wherein the step of determining a distribution of resources is associated with the normalized transaction data.

4. The step of generating the density map of the resource comprises: determining performance characteristics for each of a plurality of trained models by testing each model against test data, the performance characteristics indicative of the accuracy and / or effectiveness of each model; selecting at least one of the trained models based on the performance characteristics; 4. The method of claim 1, comprising:

5. The step of generating a density map of the resource, comprising: combining multiple trained machine learning models to generate a composite model; and generating the density map of the resource using the composite model; Including, an ensemble model is generated by combining the machine learning models, and the density map is generated from the ensemble model.

5. The method according to any one of claims 1 to 4.

6. a display of the control system; a resource of the control system; The method of claim 1 , further comprising outputting the density map of the resource to at least one of:

7. 1. An apparatus for determining future resource requirements at a given location, comprising: at least one processor; An apparatus configured to carry out the method according to any one of claims 1 to 6.

8. An apparatus according to claim 7; an output device for receiving the density map output by the apparatus; A control system comprising:

9. A computer-readable storage medium having stored thereon instructions that, when executed by a processor, cause the processor to: acquiring transaction data, the transaction data indicating past resource usage at a plurality of locations, including the given location; determining a distribution of resources associated with the transaction data; generating a machine learning kernel based on a set of parameters including a historical density map for the given location representing the resource usage over different time periods, the machine learning kernel configured to output an estimated distribution of the resource across the plurality of locations; refining and validating the machine learning kernel based on a comparison between the distribution of the resources and the estimated distribution; training at least one machine learning model based on the set of parameters by if the estimated distribution is similar to the distribution of the resource, generating a density map of the resource at the plurality of locations using at least one trained machine learning model; transmitting said density map to a control system for determining future resource requirements at said given location; 1. A computer-readable storage medium, comprising: The set of parameters is: A specific date and A specific period of time, and Holidays and weather conditions associated with the plurality of locations; further comprising at least one of A computer-readable storage medium.

Images