The implementation of this expert system (MotorUGA) uses the artificial intelligence methods, namely the genetic algorithm (for the approximation of the price function) and the neural network (for the projection of the price function).
To be tested on real ticks only!
In the first phase the approximation of the price using the genetic algorithm is performed for the specified number of bars (ChromosomeCount). That is, the extremums (minimums/maximums) of the function are determined. The genetic algorithm is able to find the extremums and reflect them, but is unable to implement the body of the function that generates these extremums (i.e. the function of price changing over time). At this stage, the lines connecting the extremums displayed on the screen after the genetic algorithm operation do not have the rules of plotting depending on the price (the lines are displayed on the screenshot). Fields related to the genetic algorithm:
- ChromosomeCount - The number of chromosomes, in the context of the number of bars.
- Epoch - Maximum number of epochs.
- ProcMinStep - Limitation on the price noise.
- Spred - The maximum spread.
- Show - Line visibility.
The next part of the works is done by the neural network. Since simply the extremums have been found, but there is no function to express them, it needs to be found. The neural network is used for this. The body of the neural network can display the projection of the real price function in its structure. The projection of the function will be checked by the already found extremums. The projection of function in the nearest future (PeriodReOptimization) is expected to display the real function of price changes as close as possible. With the projection of the function it is possible to predict the price behavior using the trained neural network. Fields related to the neural network:
- Layer1 - The number of neurons in the first layer.
- Layer2 - The number of neurons in the second layer.
- Layer3 - The number of neurons in the third layer.
- EpochN - Maximum number of epochs.
- MSE - The required precision.
- Level - Signal triggering level.
The neural network must be trained on the extremum data built on the basis in the form of the available characteristics of the market. For this, the available market characteristics and extremums are passed to the neural network as input. Once the formation of the function projection body is complete, it must be checked using these extremums, and every extremum must match! Only then the operation will be enabled. In this case, it can be expected that in the short-term perspective (PeriodReOptimization) the real function will not deviate too much from the approximated one. And the approximated function will be as close to the real one as possible. Fields related to the market characteristics:
- Index1 - 1st vector of price characteristics [0..1..27].
- Index2 - 2nd vector of price characteristics [0..1..27].
- Index3 - 3rd vector of price characteristics [0..1..27].
- Index4 - 4th vector of price characteristics [0..1..27].
- Index5 - 5th vector of price characteristics [0..1..27].
- Index6 - 6th vector of price characteristics [0..1..27].
- Index7 - 7th vector of price characteristics [0..1..27].
- Index8 - 8th vector of price characteristics [0..1..27].
What the user needs to define is the set of input characteristics of the market, based on which the neural network will project the function (create the function body). In other words, the correct input data (characteristics) of the function are required. The expert system provides a collection of basic characteristics of the market (from 1 to 27, 0 - no characteristic). During the preparation for the system operation, it is necessary to select them using the optimization of the expert system in the strategy tester and their genetic algorithm. Since all the results depend on what market characteristics (input data) are used to project the function, some characteristics may have the information on the market laws, others may be insignificant or not have useful information at all.
The other setting for the expert system are not commented in this description, as most of them are intuitive, and the maximum limit for this description does not allow to provide more extensive information. More details can be provided when contacted via private messages. This project will be developed and improved, which will be reflected in the upcoming versions, which can be downloaded for free after purchasing this version.