To work, you need network training (optimization)!

A professional robot that implements a trading strategy using neural networks. Multi-layer, fully connected MLP (multilayer perceptron) direct propagation networks are used.

The ability to learn is a basic property of the brain. For artificial neural networks, learning refers to the process of configuring the network architecture (the structure of connections between neurons) and the weights of synaptic connections (the coefficients that affect the signals) to effectively solve the task. Usually, the training of a neural network is carried out on some sample (historical data). As the learning process, which takes place according to some algorithm (for training, optimization by a genetic algorithm is used), the network should respond better and better (more correctly) to the input signals.

It remains only to check how well the optimized parameters allow you to make a forecast for the future. What is the method of step-by-step verification of the results used for? An example, along with a setting for optimization (training), is provided in the discussion.

At the beginning of the current bar, the RSI, SSI, and WPR indicators are analyzed. The results from 10 bars of each indicator get to the input of the neural network. Weight coefficients are formed separately for purchases and sales. The network is trained on data from indicators and, depending on the signal levels, 4 commands are possible at the output of the neural network (if TypeDual = true): open / close a buy order, open / close a sell order. And depending on this, it will open a BUY or SELL and hold this deal until the closing signal from the network arrives. There is also a neural network mode with two outputs (when TypeDual = false): the first is a buy entry with an automatic exit from the sale, the second is a sell entry with an automatic exit from the purchase.

[spoiler title=”Read More…”]

The Expert Advisor correctly processes errors and works reliably with a capital of 100 USD or more. The Expert Advisor uses the basic concepts: breakeven, trailing stop, stop loss and take profit, as well as closing on the opposite signal, closing on the signal and correct risk calculation. 

Basic parameters:

• WorkOpenLong – Allows you to open a long position.
• WorkOpenShort – Allows you to open a short position.
• WorkCloseLong – Allows you to close a long position.
• WorkCloseShort – Allows you to close a short position.
• WorkCloseReversLong – Allows you to close a long position when opening a short one.
• WorkCloseReversShort – Allows you to close a short position when opening a long one.
• SignalBar – The main signal bar.
• PeriodSignal – The period on which the indicators work, and, accordingly, the neural network.
• TypeDual-Switch of the neural network operation mode with 2 or 4 outputs.
• LevelInLong – The neural network entry threshold for a long position.
• LevelOutLong – The neural network output threshold for a long position.
• LevelInShort  – The neural network entry threshold for a short position.
• LevelOutShort  – The neural network exit threshold for a short position.

• EnabledRSI – Activation of the indicator.
• k1_RSI_long – The coefficient of the indicator’s influence on the total signal for a long position.
• k1_RSI_short – The coefficient of the indicator’s influence on the total signal for a short position.
• Period1_RSI – The period of the indicator.
• Layer_1AL – Field 1 for configuring the neural network for a long position.
• Layer_1BL – Field 2 for configuring the neural network for a long position.
• Layer_1AS – Field 1 for setting up a neural network for a short position.
• Layer_1BS – Field 2 for configuring the neural network for a short position.

[/spoiler]