Deep Neural Network Cashflow Forecasting with Linear Algebraic Constraints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional neural networks fail to accurately forecast cashflows across multiple accounts due to the lack of consideration for constraints such as equality and inequality constraints, which are vital for ensuring account balance consistency and reflecting real-world transaction dynamics.
Innovation Solution
A deep neural network (DNN) is used to construct a graph representing account transactions, with constraints like equality and inequality constraints applied to forecast time sequences, allowing the network to minimize a specified loss function and provide accurate, consistent cashflow forecasts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional neural networks are used for cashflow forecasting, then the forecasting process is simple, but the accuracy and consistency of account balance predictions deteriorate due to lack of constraint consideration
Solution Approach 1:
The patent introduces constraint layers as intermediary components between the neural network layers. These constraint layers act as mediators that enforce equality constraints (ensuring forecasted transactions sum to balance changes) and inequality constraints (ensuring non-negative balances) without fundamentally altering the neural network's learning capability. This intermediary approach resolves the contradiction by adding precision through constraints while maintaining manageable complexity through modular integration.
Solution Approach 2:
The patent transforms the forecasting problem by changing the parameters being optimized. Instead of directly optimizing balance predictions, the network optimizes transaction forecasts subject to constraint equations. The loss function is modified to include constraint violation penalties, changing the optimization landscape from unconstrained to constrained parameter space. This parameter transformation enables accurate forecasting while keeping the network architecture relatively simple.
2Reliability
If constraints are incorporated into the neural network to ensure account balance consistency, then forecasting accuracy improves, but the computational complexity increases
Solution Approach 1:
The patent applies preliminary action by pre-defining the constraint structure based on accounting principles before training begins. The equality constraints (transaction sums must equal balance changes) and inequality constraints (balances must be non-negative) are established a priori based on domain knowledge. This preliminary structuring enables the network to learn within predefined reliable boundaries without requiring complex real-time constraint solving during inference, thus improving reliability while controlling computational complexity.
Solution Approach 2:
The patent implements feedback mechanisms through the loss function that continuously monitors constraint satisfaction during training. The loss function includes penalty terms that provide feedback when equality constraints (transaction-balance consistency) or inequality constraints (non-negative balances) are violated. This feedback guides the network to adjust its parameters to satisfy constraints, improving balance consistency while using gradient-based optimization to manage computational complexity efficiently.
3Stability of the object's composition
If multiple constraints are applied to forecast time sequences, then the consistency of forecasts with historical transactions improves, but the training process becomes more complex
Solution Approach 1:
The patent segments the constraint system into distinct components: equality constraints (ensuring forecasted transactions sum to observed balance changes) and inequality constraints (ensuring non-negative balances). Each constraint type is handled separately in the loss function with dedicated penalty terms. This segmentation makes the training process more manageable by breaking down the complex multi-constraint optimization into structured, independently enforceable components, improving forecast consistency while maintaining training feasibility through modular constraint handling.
Data Source
AI summary
Systems and methods for forecasting cashflows across one or more accounts of a user disclosed. One example method may include retrieving a data set for each of a plurality of accounts from a database, constructing a graph including a plurality of nodes linked together by a multitude of edges, wherein each node identifies a time series value corresponding to one of the accounts, and each edge indicates a time series value of a corresponding set of transactions occurring between a corresponding pair of accounts, determining a plurality of constraints, determining a specified loss function based on the plurality of constraints, back-propagating a derivative of the specified loss function into a deep neural network (DNN) to determine a set of neural network parameters, forecasting, using the DNN, a time sequence for one or more of the nodes and one or more of the edges, and providing the forecasted time sequences to the user.