Systems and methods for predicting equipment requirements using trained data models
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2026-01-29
- Publication Date
- 2026-08-13
Smart Images
Figure EP2026052277_13082026_PF_FP_ABST
Abstract
Description
[0001] SYSTEMS AND METHODS FOR PREDICTING EQUIPMENT REQUIREMENTS USING TRAINED DATA MODELS
[0002] Technical Field
[0003] The present disclosure relates to systems and methods for determining equipment requirements at locations. More particularly, the disclosure relates to predictive analysis using trained data models to determine equipment allocation requirements at specified locations and times.
[0004] Background
[0005] Equipment allocation across different locations involves processing large volumes of data containing multiple variables and factors that can affect equipment requirements at such locations. Organizations need to continuously process and analyze data to maintain appropriate equipment levels at different locations. The complexity of data processing increases as more locations and variables are considered. Manual analysis of extensive datasets can be time-consuming and may introduce errors during interpretation.
[0006] Equipment planning approaches using fixed rules may have limitations in adapting to changing requirements, and such approaches may not effectively account for variations between different locations or incorporate new data as it becomes available. The static nature of predetermined rules therefore, limits the ability to respond to evolving distribution patterns.
[0007] Summary
[0008] According to a first example, there is provided a method for determining requirements for equipment at specified locations and times based on active shipment information. The method receives requests indicating equipment requirements for given locations and times. Multiple trained data models are evaluated to determine performance characteristics using historical data that was not used in training the models. At least one trained data model is selected based on the performance characteristics. The selected model generates predictions of equipment requirements for the specified location and time based on the received request. These predictions are sent to a control system that allocates equipment to locations according to the predictions.
[0009] P25-002PCT1The requests may include various attributes such as commodity type, customer identifier, shipment origin, shipment destination, equipment type, and temporal data associated with active shipments.
[0010] According to a second example, there is provided a system for determining equipment requirements includes storage for maintaining historical data about past equipment needs at locations. The system includes a processor with multiple modules: an input module receives equipment requirement requests, an evaluation module assesses trained data models using historical data, a selection module chooses appropriate models based on performance, a generation module creates equipment requirement predictions, and an output module sends predictions to a control system for equipment allocation.
[0011] According to a third example, there is provided a computer-readable storage medium contains instructions that, when executed by a processor, enable the determination of equipment requirements at locations and times. The instructions direct the receiving of requirement requests, evaluation of trained data models, selection of models based on performance, generation of requirement predictions, and transmission of predictions to a control system for equipment allocation.
[0012] Brief Description of the Figures
[0013] Examples of the present disclosure will now be described with reference to the accompanying drawings:
[0014] Figure 1 is a flow diagram showing a method for determining equipment requirements at a location and time;
[0015] Figure 2 is a flow diagram showing a method for training and using data models to determine equipment requirements;
[0016] Figure 3 is a sequence diagram showing interactions between a data model training system and a data model usage system; and
[0017] Figure 4 is a schematic diagram showing a system for determining equipment requirements at a location and time.
[0018] P25-002PCT1Detailed Description
[0019] Equipment requirements, such as a need for a particular type of equipment (such as shipping containers, cranes, and ... ) at a given time and at a given location can change based on various factors. Different locations may experience varying patterns of equipment usage due to local conditions and demographic factors, such as holidays, strikes, and wars. These variations create challenges in maintaining appropriate equipment levels while avoiding excess or insufficient equipment across multiple locations.
[0020] Accordingly, managing global infrastructure, for example management of equipment availability in a traffic network, such as a maritime traffic network, requires accurate forecasts with reliable uncertainty estimates across all equipment types, sites, and operational windows. This can amount to hundreds of thousands of predictions which when combined with evolving market conditions and changes in global traffic networks make learning and / or training of traditional methods based on historical data more complicated due to the amount of data needing to be processed.. Such methods may not account for changes in recent patterns of demand while discarding outdated information. Further, the rate of discarding outdated information may also vary based on a number of factors, including but not limited to the geography and special events at different locations. For example, holiday effects may remain stable, while demand trends are more dynamic in areas experiencing changes in traffic network changes. Therefore, changes in process dynamics call for methods that can automatically adapt to these changes.
[0021] In light of this dynamic nature, traditional approaches to planning and forecasting face various limitations. Certain analytical methods may not fully capture the relationships between multiple variables in available data. The duration required for data collection and processing can affect the timing of equipment allocations, and some forecasting approaches may have reduced effectiveness in responding to changes in usage patterns or external conditions. This can lead to situations where equipment distribution does not align with actual requirements across different locations.
[0022] Methods and systems for determining equipment requirements at a given location and at a given time are described below. Such methods and systems include training multiple data models and selecting one of the trained data models based on their
[0023] P25-002PCT1performance. The selected data model can then be used to generate a prediction of the equipment requirements to be sent to a control system.
[0024] In addition to market conditions and network changes such as those described above, other characteristics such as the availability and need for equipment of a given type may be used by the data models for training and / or generating the prediction. When combined, these characteristics make it difficult to accurately and efficiently predict the requirement for such equipment at given locations. Solutions to this are described below, and involve the forecasting of equipment requirements using data models to predict expected future equipment pick up and returns while also quantifying uncertainty bands associated with that prediction. Such data models incorporate historical demand / supply patterns, relevant geographical information, external factors like holidays, seasonal information, special events (e.g., pandemics, strikes), specific local customer behavior, and booking trends. Various estimation techniques can be used to obtain, from historical information, such forecasts. These estimation techniques range from classical time series methods to advanced machine learning techniques, enabling the capture of complex dependencies and reliable uncertainty quantification. In the context of a traffic network, such probabilistic forecasting techniques can be used to decide on optimal equipment stock targets and on repositioning patterns at given locations, such as ports and depots.
[0025] Furthermore, the stability of the requirements for different equipment types (i.e., how often the requirement is likely to change) may vary depending on the equipment type. For example, certain equipment may be in demand at particular times of the year depending on global or local events, such as the increased need for a crane or truck in the lead up to Chinese New Year to account for increases in traffic which occurs at particular ports, and / or a decrease in the need for such requirements during the Chinese New Year period due to decrease in activity. As such, the data models need to learn to rapidly adapt and to tailor to this by tuning feature-specific forgetting factors to what is best suited for each of time series of interest. This may be achieved by combining global and local patterns to enhance forecasts. The effect of some global events like Chinese New Year, for example, may be best modelled using broader data, whereas countryspecific holidays or localized demand trends require finer granularity. The data models may also blend expert input with other techniques to integrate global, regional, and local learnings, enabling the identification of complex patterns that might otherwise be
[0026] P25-002PCT1overlooked, thereby boosting forecast accuracy. The blending of models means that unpredictable events such as pandemics, strikes, or wars which normally pose unique challenges, can be accounted for (after their occurrence) in the model structure to ensure the models recover quickly after disruptions, avoiding prolonged accuracy drops.
[0027] In order to ensure that accuracy is maintained, granular information may be contained in any historical data used for the training of and evaluation of the data models as will be described in further detail below with reference to Figures 1 and 2. Detailed data relating to previous transactions, such as shipments, which includes data such as equipment assignments, and previous customer behavior as it comes to pick up and returns may be used. However, to ensure efficiency, dimension reduction and staged inference techniques may be utilized to transform such granular insights into intermediate features that simplify model complexity while still extracting value from the detailed information.
[0028] Such granular insights can be used to improve the accuracy of the data models by integrating global / local effects, handling special events, enabling knowledge transfer for emerging markets, and leveraging information from data of high granularity to deliver accurate, scalable forecasts. Accordingly, key to the solutions described herein, are the ability to provide equipment recommendations in a scalable and computational efficient manner, capable of quick recover after an unpredictable / special event, which adapts quickly to changes in process dynamics by the incorporation of granular insights.
[0029] Receiving and Processing Equipment Requirement Requests
[0030] Figure 1 shows a flow diagram illustrating a method for determining equipment requirements at a location and time. A request indicative of equipment requirements at a given location and a given time is received 102. The request may specify data such as equipment type, including containers or other equipment as described above, for shipment via ocean or other means. The given location may include ports, customer depots, storage lots or any other suitable location.
[0031] The performance of multiple trained data models is evaluated 104 using historical data not previously used in training the models as will be described below with reference to method 200 of Figure 2. The evaluation considers various factors including historical
[0032] P25-002PCT1observations of pickup and return data, geography information such as country and region, event information including global events and country-specific holidays, and analysis of historical shipments. The data models can include simple linear models trained for a given particular purpose, regression models with knowledge transfer, time series models trained to incorporate exogenous features such as those listed above, benchmark models including moving averages and exponential smoothing for baseline comparison, global machine learning models trained on all data, and special event handling models trained using different policies for handing special events such as pandemics, strikes and wars. High accuracy may be achieved by combining global and local patterns, as some events like Chinese New Year are best modelled using broader data, while countryspecific holidays may require finer granularity.
[0033] The pool of data models may comprise both direct and recursive multi-step data models, and both univariate and global forecasting models. The statistical univariate forecast models may be ARIMA models, recursive least squares regression models, ETS models and a moving average used as fallback. A direct global forecasting data model may be included by using gradient boosted regression trees. As mentioned above, the historical data used for training the data models may include historical demand patterns, seasonality, global holidays and country holidays of the geographical location that is being forecasted for. Furthermore, by considering requests currently placed by customers, the pick-up and return locations and time, may be used as a feature for the forecasting models that target the final total demand. In some examples, anomalous historical demand periods may be excluded so as not to pollute training data with artefacts caused by external special events that do not lie within usual operating conditions.
[0034] Based on the evaluation of performance characteristics 106, one or more data models are selected. The performance characteristics may include data associated with the accuracy of each data model based on the historical data not used to train the data model. In some examples, an entirely different data set of historical observations may be used to evaluate the performance of the data models. The selection may consider various conditions including parameter tuning and model performance on historical data. A plurality of selected models may be combined to generate a holistic data model for prediction, which can be updated periodically based on new data. This structured approach ensures that the most suitable model or models are selected and continuously
[0035] P25-002PCT1updated, providing accurate and reliable forecasts tailored to each request for equipment requirements.
[0036] Using the selected data model or models, predictions of equipment requirements are generated 108 for the given location and time based on the received request. The prediction process may incorporate multiple selected data models of different types working in combination. Using multiple models may enable more accurate predictions to be provided given that models trained for specific circumstances can be used. For example, one model may be trained and tuned on more local data, whereas another model may be trained and tuned on more global data. The prediction may be generated at multiple times using the selected data model or models, for example, a prediction may be generated weekly. The prediction may include information such as:
[0037] • Target Stock Levels: The target quantity of each equipment type required in each location as determined based on expected customer demand, incorporating forecast uncertainty and other risk estimates to add buffers. These buffers may help balance the cost of overstocking with the risk of missed opportunities and may be used for equipment management, including the repositioning of equipment based on the required quantity of each equipment type at a given location.
[0038] • Stock Projections: Forecasts of future equipment pick-ups and returns may help predict changes to the amount of equipment at a given location over time. These may also be used to determine the likely availability of a equipment of a required type when accepting a new request.
[0039] • Demand Forecasts split into contract / free-sale: Granular demand forecasts support more nuanced decision-making, such as request acceptance. If projected total equipment stock is lower than expected total demand, the system may restrict new equipment requests to prioritize equipment for contract commitments..
[0040] The generated predictions are sent 110 to a control system, such as an administration computer system, and / or management system operating either on a hardware device or remotely as a cloud-computing solution. The control system is responsible for managing equipment allocations, such as allocated containers (the equipment) to ports (the given location) at a given time. The control system then allocates
[0041] P25-002PCT1equipment to the given location based on the predictions, ensuring availability for use at the specified time.
[0042] The method leverages models which utilize machine learning techniques to integrate global, regional, and local learnings, identifying complex patterns that might otherwise be overlooked. This approach enhances forecast accuracy by blending inputs across different geographical scales and temporal patterns.
[0043] The evaluation of data models using previously unseen historical data enables the identification of models that are likely to perform well in unknown / new situations. This approach helps determine which models can reliably generalize beyond their training data. The selection process considers how each model handles novel scenarios, leading to more dependable predictions for equipment requirements.
[0044] The integration between predictive systems and control systems enables automated equipment allocation based on data-driven forecasts. Equipment can be distributed proactively according to predicted needs rather than reactive manual assignments. This automated approach helps optimize equipment distribution across different locations.
[0045] Using multiple trained models with performance-based selection provides robust predictions across varying conditions and scenarios. When individual models encounter difficulties with specific patterns or situations, other models in the ensemble can compensate. The performance-based selection mechanism automatically adjusts to use the most suitable models for given circumstances.
[0046] Historical Data Collection and Analysis
[0047] Figure 2 illustrates additional aspects of the method for determining equipment requirements shown in Figure 1. The method begins by obtaining historical data 202 indicative of past equipment requirements. As described above, this historical data may include data associated with historical demand and supply patterns, geographical information, external factors such as holidays and seasonal information, special events like strikes or pandemics, specific local customer behavior, and booking trends.
[0048] The historical data may be filtered or pruned based on various factors, including determining when information becomes outdated. The historical data may comprise
[0049] P25-002PCT1information about effects of events that increase or decrease equipment requirements at locations, along with previous customer behavior patterns. For example, certain data may be disregarded after a predetermined period in some models, such as seasonal parameters in time series models.
[0050] Multiple data models are trained 204 using subsets of the historical data, such as the data described earlier, to determine patterns of equipment requirements at given locations and times. The training process includes allocating weights to events in the historical data subset. In some examples, the rate of discarding outdated information varies by feature and geography - for example, holiday effects may remain stable while demand trends may be more dynamic in areas experiencing network changes. Stability can differ across equipment types, with some types showing more dynamic patterns than others. The models adapt to these variations by tuning feature-specific factors appropriate for each time series.
[0051] When trained data models have been generated, a request for equipment 206 can be made specifying a given location and time. The request, as described above in relation to Figure 1, may include data such as equipment type, including containers for shipment via ocean or other means. Given locations may include ports and customer depots or other suitable locations.
[0052] As described above, the performance of the data models is evaluated 208 using historical data not used in training. High accuracy at individual locations and for specific equipment types is achieved by combining global and local patterns. The evaluation considers multiple factors including historical observations of pickup and return data, geographical information, event information, and analysis of historical shipments.
[0053] Data models are then selected 210 based on their performance characteristics, considering various conditions including parameter tuning and performance on historical data. The selected models may be combined to generate a comprehensive data model for prediction, which can be updated periodically based on new data. As described above, it will be appreciated that a number of factors may be considered when determining the performance of a given data model or models.
[0054] Using the selected data models, predictions of equipment requirements are generated 212 for the given location and time based on the received request. The
[0055] P25-002PCT1prediction process may incorporate multiple selected data models of different types working in combination.
[0056] The generated predictions are sent 214 to a control system responsible for managing equipment allocations. The control system then allocates equipment 216 to the given location based on the predictions, ensuring availability for use at the specified time.
[0057] The method employs probabilistic demand / supply forecasting using mathematical models to predict expected future equipment pickup and returns while quantifying uncertainty bands. Estimation techniques range from classical time series methods to advanced machine learning approaches, enabling capture of complex dependencies and reliable uncertainty quantification.
[0058] The weighting of events during model training enables differentiated handling of various event impacts on equipment requirements. Events can influence equipment needs to different degrees, and applying weights allows the models to account for these varying levels of influence. The weighted training approach helps capture the relative significance of different events in driving changes to equipment demand patterns.
[0059] Training and Deployment of Equipment Requirement Models
[0060] Figure 3 relates to the features shown in Figures 1 and 2, illustrating a sequence diagram showing the training and subsequent use of data models for determining equipment requirements.
[0061] A request system 302 may be a user device, such as a computer terminal or mobile telephone capable of being connected either wired or wirelessly to other aspects of the system. The request system 302 sends message 312 containing a request for equipment at a given location and time. The request may specify equipment types, including containers for shipment via ocean or other means, where the given location may include ports and customer depots.
[0062] A model evaluator 304 receives the request and sends message 314 to obtain historical data. The historical data includes data associated with historical demand and supply patterns alongside geographical information, external factors including holidays, seasons information and special events. This data can be combined with specific local customer behavior and booking trends. The historical information may be pruned based
[0063] P25-002PCT1on factors including a rate of determining when information becomes outdated. The model evaluator 304 may be a server, configured to perform evaluations using historical data not used to train the data models.
[0064] A training system 306 receives the historical data and performs train data models operation 316. The training system 306 is configured to train different types of data models, including simple linear models, regression models, time series models, benchmark models, global machine learning models, and special event handling models. The rate of discarding outdated information varies by feature and geography - holiday effects may remain stable, while demand trends are more dynamic in areas experiencing network changes.
[0065] The trained data models are sent via message 318 to the model evaluator 304. High accuracy is achieved by combining global and local patterns. Effects of global events like Chinese New Year are modeled using broader data, while country-specific holidays require finer granularity. The data models blend inputs with Al techniques to integrate global, regional, and local learnings.
[0066] The model evaluator 304 evaluates the models with test data via message 320, considering factors including historical observations, geography, event information, and analysis of historical shipments. Selected models are sent via message 322 to a prediction generator 308.
[0067] The prediction generator 308 generates equipment requirement predictions based on the request using selected data models via message 324. The prediction may incorporate multiple trained data models of different types working in combination. The generated prediction is sent via message 326 to a control system 310.
[0068] Based on the receipt of message 326, the control system 310 is responsible for allocating equipment to various locations based on the prediction of equipment requirements at the given location, ensuring availability at the specified time.
[0069] System for Training and Predicting Equipment Requirements
[0070] Figure 4 shows a schematic diagram of a system 402 configured to perform the method described in relation to Figure 1 and Figure 2. The system 402 is configured to train a plurality of data models, and determine a requirement for equipment at a given location
[0071] P25-002PCT1in the form of a prediction 418 in response 416 to a request indicative of a requirement for equipment at a given location. The system 402 may form part of a remote server, remote from a sender device used to send the data package from the sender.
[0072] The system 402 includes storage 404 for storing at least historical data indicative of past equipment requirements at, at least the given location. The storage 404 may comprise a main memory or primary memory, which can be external to the system such as off-chip memory or remote cloud storage. The storage 404 may include magnetic or optical disks, solid-state drives (SSDs), or synchronous dynamic random-access memory (SDRAM).
[0073] A processor, such as a central processing unit, graphics processing unit, neural processing unit, or any other suitable hardware which may form part of a computing device, such as a desktop computer, laptop computer, mobile telephone, tablet computer, or wearable device, 406 processes inputs / requests 416 received by the system 402. The processor 406 includes an evaluation module 408, a selection module 410 and a generation module 412. In some examples, the processor 406 also comprises a training module 420 for training the data modules, although it will be understood that the training module may be separate from the system 402 and form part of a remote system, such as a cloud-based system.
[0074] The evaluation module 408 is configured to evaluate the performance of multiple trained data models using historical data not previously used in training the models as described above with reference to method 200 of Figure 2. The evaluation considers various factors including historical observations of pickup and return data, geography information such as country and region, event information including global events and country-specific holidays, and analysis of historical shipments. The data models can include simple linear models trained for a given particular purpose, regression models with knowledge transfer, time series models trained to incorporate exogenous features such as those listed above, benchmark models including moving averages and exponential smoothing for baseline comparison, global machine learning models trained on all data, and special event handling models trained using different policies for handing special events such as pandemics, strikes and wars. High accuracy may be achieved by combining
[0075] P25-002PCT1global and local patterns, as some events like Chinese New Year are best modelled using broader data, while country-specific holidays may require finer granularity.
[0076] The pool of data models may comprise both direct and recursive multi-step data models, and both univariate and global forecasting models. The statistical univariate forecast models may be ARIMA models, recursive least squares regression models, ETS models and a moving average used as fallback. A direct global forecasting data model may be included by using gradient boosted regression trees. As mentioned above, the historical data used for training the data models may include historical demand patterns, seasonality, global holidays and country holidays of the geographical location that is being forecasted for. Furthermore, by considering requests currently placed by customers, the pick-up and return locations and time, may be used as a feature for the forecasting models that target the final total demand. In some examples, anomalous historical demand periods may be excluded so as not to pollute training data with artefacts caused by external special events that do not lie within usual operating conditions.
[0077] The selection module 410 performance characteristics of data models in the pool of data models are evaluated and, based on the performance characteristics, one or more of the data models is selected. The performance characteristics may include data associated with the accuracy of each data model based on the historical data not used to train the data model. In some examples, an entirely different data set of historical observations may be used to evaluate the performance of the data models. The selection may consider various conditions including parameter tuning and model performance on historical data. A plurality of selected models may be combined to generate a holistic data model for prediction, which can be updated periodically based on new data. This structured approach ensures that the most suitable model or models are selected and continuously updated, providing accurate and reliable forecasts tailored to each request for equipment requirements.
[0078] Using the selected data model or models, predictions of equipment requirements are generated by the generation module 412 for the given location and time based on the received request. The prediction process may incorporate multiple selected data models of different types working in combination. Using multiple models may enable more accurate predictions to be provided given that models trained for specific circumstances
[0079] P25-002PCT1can be used. For example, one model may be trained and tuned on more local data, whereas another model may be trained and tuned on more global data. The prediction may be generated at multiple times using the selected data model or models, for example, a prediction may be generated weekly as described previously.
[0080] In some examples, the same system 402 may be used to train the data models. The training of the data models may be performed by a training module 420 using subsets of the historical data obtained from the storage 404, such as the data described earlier, to determine patterns of equipment requirements at given locations and times. The training process includes allocating weights to events in the historical data subset. In some examples, the rate of discarding outdated information varies by feature and geography -for example, holiday effects may remain stable while demand trends may be more dynamic in areas experiencing network changes. Stability can differ across equipment types, with some types showing more dynamic patterns than others. The models adapt to these variations by tuning feature-specific factors appropriate for each time series.
[0081] An input-output device 414 which may be a combined I / O device or separate input and output modules facilitates communication between the system and users, either locally or via a wide area network, like the Internet. The input-output device receives requests 416 from senders and outputs the generated predictions 418. The request 416 may be an input or other digital communications received from a sending device (not shown), while the predictions 418 are generated based on the selected data model(s) as described above in relation to Figures 1 and 2. The outputted prediction 418 may be sent to a control system responsible for allocating the equipment to the given location.
[0082] Components of the system may be connected via an interconnection such as a system bus, or other wired or wireless connections enabling communication across networks such as the Internet. The system continuously processes incoming communications, analyzing context and sentiment to generate appropriate responses that maintain consistency while adapting to specific sender needs.
[0083] The modular architecture separates key processing functions into dedicated components for evaluation, selection, and generation. Each component performs specialized tasks while maintaining seamless integration through standardized
[0084] P25-002PCT1data flows. This structured approach supports efficient processing of communications while allowing independent optimization of each function.
[0085] The database structure enables systematic organization of historical data. This organization facilitates quick retrieval of relevant previous interactions and reply patterns. The structured storage supports efficient analysis of the historical data subsequent generation of predictions.
[0086] The integrated input-output functionality enables continuous processing of requests. Requests may be received, analyzed, and predictions generated without interruption or delay. This streamlined handling maintains communication flow while ensuring appropriate processing of each interaction.
[0087] The implementation using a computer-readable storage medium enables deployment across various computing platforms and environments. The storage medium can be integrated with different processor architectures and operating systems. This flexibility allows the functionality to be utilized on diverse hardware configurations while maintaining consistent operation.
[0088] The system incorporates ongoing analysis of requests to refine future predictions. As new requests are processed, the analysis results inform adjustments to the historical data, and in some examples the training of the data models. This continuous refinement helps maintain and enhance the relevance of the generated predictions.
[0089] Additional Variations
[0090] The trained data models may be tuned through various approaches. For example, parameters of the data models can be adjusted based on time series data. Additionally, a complexity parameter associated with the data models may be adjusted to optimize performance.
[0091] Multiple trained data models may be combined to generate predictions. For instance, a first trained data model and a second trained data model may be selected and combined to create a combined data model. The combined data model can then be used to generate the predictions of equipment requirements.
[0092] The request for determining equipment requirements may include various attributes. These attributes can comprise a type of commodity, a customer identifier, an
[0093] P25-002PCT1origin of a shipment, a destination of the active shipment, a type associated with the equipment, and temporal data associated with the active shipment.
[0094] The historical data used for training and evaluation may encompass different types of information. This can include equipment usage data collected from multiple locations. Customer data indicating previous customer behavior patterns may also be incorporated into the historical data.
[0095] The control system 310 can allocate equipment based on various factors and constraints. The allocation process may take into account the availability of different types of equipment, transportation routes, and scheduling requirements. The system may also consider priorities and urgency levels associated with different shipments when making allocation decisions.
[0096] Events affecting equipment requirements may be categorized in different ways. Fixed events may be associated with specific locations and times. Global events may be tied to particular times without being location-specific. Special events, which may be one-off occurrences or events without fixed timing, can also impact equipment requirements. When evaluating model performance, historical data associated with special events may be excluded from the evaluation dataset to ensure more accurate assessment of model capabilities under typical conditions.
[0097] The categorization of events into fixed, global and special types enables processing of diverse influences on equipment demand. Fixed events tied to specific locations and times, global events linked to particular times, and special one-off events each contribute distinct patterns to the analysis. This structured classification approach allows the data models to account for regular patterns while also incorporating exceptional circumstances that may affect equipment requirements.
[0098] Excluding special events from the historical data used for evaluation allows the data models to be assessed based on typical operational patterns. The evaluation focuses on regular, recurring conditions rather than exceptional circumstances. This approach helps establish a more stable baseline for measuring the predictive capabilities of the models under standard operating scenarios.
[0099] P25-002PCT1The incorporation of equipment usage data from multiple locations enables broader pattern recognition across different operational contexts. Analysis of usage trends across various sites can reveal correlations and commonalities in equipment requirements. This multi -location perspective supports more comprehensive and robust predictions by accounting for diverse usage scenarios and operational environments.
[0100] The incorporation of customer behavior data into the historical dataset enables the system to recognize and account for recurring patterns in how customers utilize equipment. Analysis of previous customer actions and preferences allows the data models to capture individual usage trends and requirements. This behavioral information contributes to generating more precise predictions that reflect actual customer needs and equipment utilization patterns.
[0101] The tuning of model parameters enables the data models to adapt and evolve over time. By adjusting parameters based on time series data and modifying complexity parameters, the models can respond to changing patterns and trends. This dynamic refinement process helps maintain and improve the accuracy of equipment requirement predictions as conditions change.
[0102] The combination of multiple trained data models enables leveraging distinct predictive capabilities. Different models may excel at capturing various aspects of equipment requirement patterns. By integrating complementary approaches through model combination, the resulting predictions can benefit from the strengths of each individual model.
[0103] The combined data model enables more accurate predictions by leveraging multiple analytical approaches. The integration of different modeling techniques allows for capturing diverse patterns and relationships within the data. This multi-faceted analytical approach produces predictions that reflect a broader range of underlying factors and behaviors.
[0104] The processing of multiple attributes related to shipments enables enhanced analysis of equipment requirements. By considering comprehensive contextual information such as commodity types, customer details, shipment origins, destinations, equipment types, and temporal aspects, the system can develop more precise predictions.
[0105] P25-002PCT1The combination of these different data points allows for a more complete understanding of the operational context when determining equipment needs.
[0106] The evaluation of trained data models using previously unseen historical data enables identification of models that perform well on new scenarios. The performance characteristics determined through this evaluation provide an objective basis for model selection. This approach helps ensure the selected models can effectively handle novel situations while maintaining reliable prediction capabilities.
[0107] The direct connection between prediction outputs and the control system streamlines the equipment allocation process. The control system can automatically translate predictions into concrete allocation decisions without manual intervention. This integration allows for rapid response to changing equipment needs across different locations.
[0108] The incorporation of historical data patterns specific to each location enhances the accuracy of equipment requirement predictions. Location-specific patterns may reflect regular fluctuations in demand, seasonal variations, or recurring local events. The system can leverage these historical patterns to anticipate future equipment needs more precisely at each distinct location.
[0109] The analysis of historical data incorporating both event impacts and customer behavior patterns enables more accurate forecasting of equipment requirement variations. The combination of these data types allows the system to identify correlations between specific events and changes in equipment demand. This comprehensive approach helps account for how different events and customer responses can lead to increases or decreases in equipment needs at particular locations.
[0110] The allocation of weights to different events during the training of data models enables enhanced handling of varying event impacts. Events can have different levels of influence on equipment demand patterns and requirements. The weighting approach allows the models to appropriately account for these differences in impact when generating predictions.
[0111] The implementation using a computer-readable storage medium allows the equipment requirement determination functionality to operate across different computing
[0112] P25-002PCT1platforms and hardware configurations. The storage medium can contain instructions executable by various processor types. This flexibility enables deployment of the predictive capabilities across diverse technical environments and computing systems.
[0113] The evaluation of trained data models using previously unused historical data provides an objective measure of model performance. Testing with separate validation data helps assess how well each model generalizes to new situations. The performance characteristics obtained through this evaluation inform the selection of appropriate models for generating predictions.
[0114] The direct integration with control systems enables automated equipment allocation based on the generated predictions. The predictions can be transmitted to the control system which can then dynamically assign equipment. This automated flow from prediction to allocation helps optimize equipment distribution across locations in response to changing requirements.
[0115] The training framework incorporates multiple complementary data sources to build predictive capabilities. Historical data provides insights into past equipment needs, while event weighting enables calibration based on situational factors. The integration of customer behavior patterns allows the system to account for usage trends and preferences when determining equipment requirements.
[0116] The combination of data sources and modeling techniques enables enhanced prediction accuracy. Multiple trained models working together can capture different aspects of equipment demand patterns. The weighted event analysis and behavioral data integration support more nuanced forecasting compared to simpler approaches.
[0117] P25-002PCT1
Claims
Claims1. A method for determining a requirement for equipment at a given location, and at a given time, based on active shipment information, the method comprising the steps of:receiving a request indicative of the requirement for equipment at the given location, and the given time;evaluating each of a plurality of trained data models to determine a performance characteristic based on given historical data, where the given historical data has not been used to train the plurality of data models;selecting at least one of the trained data models based on the performance characteristic;generating at least one prediction of the requirement for equipment at the given location at the given time based on the request and using the selected trained data models; andsending the at least one prediction of the requirement for equipment at the given location at the given time, to a control system to allocate at least some equipment to the given location based on the at least one prediction.
2. The method for determining a requirement for equipment at the given location, and at the given time, according to claim 1, further comprising:obtaining historical data indicative of past equipment requirements at, at least, the given location, the historical data comprising at least:data indicative of an effect of one or more events on the past equipment requirements; andprevious customer behaviour,where the effect increases or decreases equipment requirements at the given location;training the plurality of data models based on a subset of the historical data, the training of each of the plurality of data models comprising:allocating a weight to each of the one or more events in the subset of historical data;P25-002PCT1training each of the plurality of data models based on at least some of the subset of historical data and the allocated weights.
3. The method for determining a requirement for equipment at the given location, and at the given time according to claim 2, wherein the one or more events comprise at least data associated with:fixed events associated with the given location and at the given time; global events associated with the given time; andspecial events, wherein the special events do not occur at a fixed time, and / or are one-off events.
4. The method for determining a requirement for equipment at the given location, and at the given time according to claim 3, wherein the given historical data that has not been used to train the plurality of data models excludes historical data associated with special events.
5. The method for determining a requirement for equipment at the given location, and at the given time according to any previous claim, wherein the historical data further comprises equipment usage at a plurality of locations.
6. The method for determining a requirement for equipment at the given location, and at the given time according to any previous claim, wherein the historical data further comprises customer data indicative of previous customer behaviour.
7. The method for determining a requirement for equipment at the given location, and at the given time according to any previous claim, further comprising tuning each of the plurality of trained data models, wherein tuning comprises at least one of:adjusting parameters of the data models based on a time series; and adjusting a complexity parameter associated with the data models.
8. The method for determining a requirement for equipment at the given location, and at the given time according to any previous claim, wherein selecting at least one ofP25-002PCT1the trained data models further comprises selecting a first trained data model and a second trained data model and, based on the selection generating a combined data model.
9. The method for determining a requirement for equipment at the given location, and at the given time according to claim 8, wherein generating at least one prediction comprises using the combined data model.
10. The method for determining a requirement for equipment at the given location, and at the given time, wherein the request comprises a plurality of attributes, the plurality of attributes comprising:a type of commodity;a customer identifier;an origin of a shipment;a destination of the active shipment;a type associated with the equipment; andtemporal data associated with the active shipment.
11. A system for determining a requirement for equipment at a given location, and at a given time, based on active shipment information, the system comprising:storage for storing historical data indicative of past equipment requirements at, at least the given location;a processor comprising:an input module for receiving a request indicative of a requirement for equipment at a given location and the given time;an evaluation module for evaluating each of a plurality of trained data models to determine a performance characteristic based on given historical data, where the given historical data has not been used to train the plurality of data models;a selection module for selecting at least one of the trained data models based on the performance characteristic;a generation module generating at least one prediction of the requirement for equipment at the given location at the given time based on the request and using the selected trained data models; andP25-002PCT1an output module for sending the at least one prediction of the requirement for equipment at the given location at the given time, to a control system to allocate at least some equipment to the given location based on the at least one prediction.
12. The system for determining the requirement for equipment at the given location, and at the given time according to claim 11, wherein the historical data comprising at least:data indicative of an effect of one or more events on the past equipment requirements; andprevious customer behaviourwhere the effect increases or decreases equipment requirements at the given location.
13. The system for determining the requirement for equipment at the given location, and at the given time according to claim 12, wherein the processor further comprises a training module for:training the plurality of data models based on a subset of the historical data, the training of each of the plurality of data models comprising:allocating a weight to each of the one or more events in the subset of historical data;training each of the plurality of data models based on at least some of the subset of historical data and the allocated weights14. A computer-readable storage medium, storing instructions that, when executed by a processor cause the processor to determine a requirement for equipment at a given location, and at a given time, based on active shipment information, the instruction comprising:receiving a request indicative of the requirement for equipment at the given location, and the given time;evaluating each of a plurality of trained data models to determine a performance characteristic based on given historical data, where the given historical data has not been used to train the plurality of data models;P25-002PCT1selecting at least one of the trained data models based on the performance characteristic;generating at least one prediction of the requirement for equipment at the given location at the given time based on the request and using the selected trained data models; andsending the at least one prediction of the requirement for equipment at the given location at the given time, to a control system to allocate at least some equipment to the given location based on the at least one prediction.
15. The computer-readable storage medium, storing instructions that, when executed by a processor cause the processor to determine the requirement for equipment at the given location, and at the given time, based on active shipment information, according to claim 14, where the instructions further comprise:obtaining historical data indicative of past equipment requirements at, at least, the given location, the historical data comprising at least:data indicative of an effect of one or more events on the past equipment requirements; andprevious customer behaviour,where the effect increases or decreases equipment requirements at the given location;training the plurality of data models based on a subset of the historical data, the training of each of the plurality of data models comprising:allocating a weight to each of the one or more events in the subset of historical data;training each of the plurality of data models based on at least some of the subset of historical data and the allocated weights.P25-002PCT1