Stochastic demand model ensemble

Abstract

A system is provided that generates a capacity plan for a resource representing supply to meet demand based on minimizing a cost objective. The system generates demand scenarios by applying a stochastic process that factors in historical information, future goals, and uncertainty in demand. The system generates supply scenarios indicating supply over time for the resource by applying a stochastic process that factors in factors relating to quantity of supply units of the resource and uncertainty in supply. The system identifies a supply scenario that minimizes costs relating to delivery of supply at times other than the times at which supply is need to meet demand based on the demand scenarios. The supply scenario represents the capacity plan.

We claim: 
     
         1 . A method performed by one or more computing systems for generating a capacity plan ensuring availability of information technology resources of a cloud data center, the method comprising:
 generating a demand collection of demand series, each demand series providing a forecast of demand for the resource at periods within the demand series, the demand collection generated based on a stochastic process based on factors relating to historical information and future goals, the stochastic process including for each of a plurality of revenue milestones:
 sampling a probability distribution of possible periods when a revenue milestone will be reached; 
 for each sampled date, calculating a growth rate to reach the revenue milestone; and 
 generating a revenue collection of revenue series that is based on a process with an expected value and variance based on the calculated growth rate; 
   generating a supply collection of supply series, each supply series in a supply collection providing a forecast for delivery periods for the supply of the resource based on order information for the supply, the supply collection generated based on a stochastic process that factors in factors relating to availability of a supply unit;   identifying a supply series that minimizes a cost, the cost represented by a cost function that is based on costs associated with delivery of supply at periods other than the period at which supply is needed to meet demand as indicated by the demand collection; and   outputting an indication of an order period that the identified supply series is based on.   
     
     
         2 . The method of  claim 1  wherein the capacity plan is generated for a strategy that specifies supply units, each supply unit specifying a quantity of the resource, a lead time, and a cost for the quantity. 
     
     
         3 . The method of  claim 1  wherein the costs include carrying costs and loss opportunity costs. 
     
     
         4 . The method of  claim 1  wherein the identifying of the supply series includes applying a gradient descent technique to adjust the order period to identify an optimal order period based on minimizing the cost. 
     
     
         5 . The method of  claim 1  wherein the demand collection is based on milestones for revenue and productivity. 
     
     
         6 . The method of  claim 1  wherein the demand collection is based on adoption of technology. 
     
     
         7 . The method of  claim 1 , wherein the factors relate to uncertainty in order lead time, and wherein the uncertainty in order lead time is based on sentiment relating to whether a supplier of the resource will deliver the resource within a timeframe set by the supplier. 
     
     
         8 . The method of  claim 7  wherein the sentiment is based on data collected from publications relating to a suppler. 
     
     
         9 . The method of  claim 1  further comprising generating a modified capacity plan based on a modified factor and indicating a modified order period associated with the modified capacity plan wherein differences in costs represent sensitivity to the modified factor. 
     
     
         10 . The method of  claim 1  wherein multiple factors have modified values that are different from values used to generate the capacity plan and further comprising generating new capacity plans based on different combinations of the multiple factors having unmodified and modified values and indicating costs associated with each new capacity plan. 
     
     
         11 . A method performed by one or more computing systems for generating a capacity plan ensuring availability of information technology resources of a cloud data center, the method comprising:
 training a machine learning model with training data that includes usage data for a sequence of time periods labeled with usage data for a subsequent sequence of time periods;   forecasting one or more demand scenarios, each indicating demand over time for the resource by applying a process that factors in factors relating to historical information, future goals, and uncertainty in demand, wherein the forecasting of the one or more demand scenarios is based on sampling a distribution for possible usages for sequence of time periods based on an output of the machine learning model for that sequence of time periods;   forecasting a plurality of supply scenarios, each indicating supply over time for the resource by applying a process that factors in factors relating to quantity of supply units of the resource and uncertainty in supply;   identifying a supply scenario that minimizes costs relating to delivery of supply at times other than the times at which supply is needed to meet demand based on the one or more demand scenarios; and   outputting an indication of the identified supply scenario.   
     
     
         12 . The method of  claim 11  wherein the capacity plan is generated for a strategy that specifies supply units, each supply unit specifying a quantity of the resource, a lead time, and a cost for the quantity. 
     
     
         13 . The method of  claim 11  wherein the costs include carrying costs and loss opportunity costs. 
     
     
         14 . The method of  claim 11  wherein the identifying of the supply scenario includes applying a gradient descent technique to adjust an order date for the supply scenario to identify an optimal order date based on minimizing the costs. 
     
     
         15 . The method of  claim 11  wherein the forecasting of demand is based on milestones for revenue and productivity. 
     
     
         16 . The method of  claim 11  wherein the demand is based on adoption of technology. 
     
     
         17 . The method of  claim 11  wherein the uncertainty in supply is based on sentiment relating to whether a supplier of the resource will deliver the resource within a timeframe set by the supplier. 
     
     
         18 . The method of  claim 17  wherein the sentiment is based on data collected from publications relating to a suppler. 
     
     
         19 . The method of  claim 11  further comprising generating a modified capacity plan based on a modified factor and indicating a modified scenario associated with the modified capacity plan wherein differences in costs represent sensitivity to the modified factor. 
     
     
         20 . The method of  claim 11  further comprising generating a new capacity plan based on changed factors with changed values that are different from original values used in generating the capacity plan and generating other new capacity plans based on different combinations of the changed factors having original values and modified values and indicating costs associated with each new capacity plan. 
     
     
         21 . The method of  claim 11  wherein the resource is supplied via a supply chain with multiple suppliers in the supply chain. 
     
     
         22 . The method of  claim 11  further comprising sending orders to one or more suppliers of the resource. 
     
     
         23 . The method of  claim 11  wherein the forecasting of a demand scenario is based on historical usage of the resource. 
     
     
         24 . The method of  claim 23  wherein the forecasting based on historical usage applies a regression demand model. 
     
     
         25 . The method of  claim 23  wherein the forecasting based on historical usage applies a neural network demand model. 
     
     
         26 . The method of  claim 11  wherein the forecasting of a demand scenario is based on a project that uses the resource. 
     
     
         27 . One or more computing systems for generating a capacity plan ensuring availability of information technology resources of a cloud data center, the capacity plan providing a timing for supply of the information technology resource to meet demand for that resource, the one or more computing systems comprising:
 one or more computer-readable storage media storing computer-executable instructions that when executed by the one or more computing systems cause the one or more computing systems to:
 generate a demand collection of demand series, each demand series providing a forecast of demand for the resource at periods within the demand series, the demand collection generated based on a stochastic process based on factors relating to historical information and future goals, the stochastic process including for each of a plurality of revenue milestones: 
 sample a probability distribution of possible periods when a revenue milestone will be reached; 
 for each sampled date, calculate a growth rate to reach the revenue milestone; and 
 generate a revenue collection of revenue series that is based on a process with an expected value and variance based on the calculated growth rate; 
 generate a supply collection of supply series, each supply series in a supply collection providing a forecast for delivery periods for the supply of the resource based on order information for the supply, the supply collection generated based on a stochastic process that factors in factors relating to availability of a supply unit; 
 identify a supply series that minimizes a cost, the cost represented by a cost function that is based on costs associated with delivery of supply at periods other than the period at which supply is needed to meet demand as indicated by the demand collection; and 
 output an indication of an order period that the identified supply series is based on; and 
   one or more processors that control execution of the computer-executable instructions.   
     
     
         28 . The one or more computing systems of  claim 27  wherein the capacity plan is generated for a strategy that specifies supply units, each supply unit specifying a quantity of the resource, a lead time, and a cost for the quantity. 
     
     
         29 . The one or more computing systems of  claim 27  wherein the costs include carrying costs and loss opportunity costs. 
     
     
         30 . The one or more computing systems of  claim 27  wherein the identifying of the supply series includes applying a gradient descent technique to adjust an order date for the supply series to identify an optimal order date based on minimizing the costs. 
     
     
         31 . The one or more computing systems of  claim 27  wherein the forecasting of demand is based on milestones for revenue and productivity. 
     
     
         32 . The one or more computing systems of  claim 27  wherein the demand is based on adoption of technology. 
     
     
         33 . The one or more computing systems of  claim 27  wherein an uncertainty in supply is based on sentiment relating to whether a supplier of the resource will deliver the resource within a timeframe set by the supplier. 
     
     
         34 . The one or more computing systems of  claim 33  wherein the sentiment is based on data collected from publications relating to a suppler. 
     
     
         35 . The one or more computing systems of  claim 27  further comprising generating a modified capacity plan based on a modified factor and indicating a modified scenario associated with the modified capacity plan wherein differences in costs represent sensitivity to the modified factor. 
     
     
         36 . The one or more computing systems of  claim 27  further comprising generating a new capacity plan based on changed factors with changed values that are different from original values used in generating the capacity plan and generating other new capacity plans based on different combinations of the changed factors having original values and modified values and indicating costs associated with each new capacity plan. 
     
     
         37 . The one or more computing systems of  claim 27  wherein the resource is supplied via a supply chain with multiple suppliers in the supply chain. 
     
     
         38 . The one or more computing systems of  claim 27  further comprising sending orders to one or more suppliers of the resource. 
     
     
         39 . The one or more computing systems of  claim 27  wherein the forecasting of a demand scenario is based on historical usage of the resource. 
     
     
         40 . The one or more computing systems of  claim 39  wherein the forecasting based on historical usage applies a regression demand model. 
     
     
         41 . The one or more computing systems of  claim 39  wherein the forecasting based on historical usage applies a neural network demand model. 
     
     
         42 . The one or more computing systems of  claim 27  wherein the forecasting of a demand scenario is based on a project that uses the resource. 
     
     
         43 . One or more computing systems for ensuring availability of information technology resources in a cloud data center to meet demand for servers, the one or more computing systems comprising:
 one or more computer-readable storage media storing computer-executable instructions that when executed by the one or more computing systems cause the one or more computing systems to:
 generate resource demand scenarios for resources that are based on historical usage of resources in the cloud data center, each demand scenario providing a forecast of demand for resources, the resource demand scenarios generated by applying a process that factors in historical information on demand of resources and uncertainty in future demand of resources, the process including for each of a plurality of milestones:
 sampling a distribution of possible periods when a milestone will be reached; 
 for each sampled period, calculating a growth rate metric to reach the milestone; and 
 generating a revenue collection of revenue series that is based on a randomization process with an expected value and variance based on the growth rate metric; 
 
 generate resource availability scenarios for resources that are based on forecasted availability of resources by applying a process that factors in availability of resources; 
 identify a quantity of resources to meet the demand for resources based on the demand scenarios and server availability scenarios; and 
 directing that the identified quantity of resources is available to meet the demand for resources; and 
   one or more processors that control execution of the computer-executable instructions.   
     
     
         44 . The one or more computing systems of  claim 43  wherein the demand factors in a productivity increase in a computer program. 
     
     
         45 . The one or more computing systems of  claim 44  wherein the productivity is based on processing of the computer program that is performed by a graphics processing unit (GPU). 
     
     
         46 . The one or more computing systems of  claim 43  wherein the instructions further comprise instructions that cause the one or more computing systems to:
 access data based on historical usage of resources; 
 generate training data that includes information relating to usage over a first time period and usage over a second time period that is later than the first time period; 
 train a machine learning model using the training data; and 
 employ the trained machine learning model to sample usages when generating the demand scenarios.