Membrane fouling early warning method and device based on machine learning

A membrane fouling prediction device based on machine learning, comprising:

• a model constructing unit, configured to construct a membrane fouling prediction model based on machine learning, which is capable of outputting electrochemical information values at different moments based on influent water quality parameters, for characterizing a fouling level of a membrane treatment system;
• a warning time setting unit, configured to set a time interval Δt for an early warning;
• a prediction unit, configured to output in advance the electrochemical information values of the membrane treatment system at the different moments, based on the influent water quality parameters, by means of the constructed membrane fouling prediction model, from which an electrochemical information value Z at a moment t and an electrochemical information value Zt+Δt at a moment t+Δt are obtained, respectively;
• a calculating unit, configured to calculate a determination value E of the early warning according to formula:
• E
• =
• |
• z
• t
• +
• Δ
• ⁢
• t
• -
• z
• t
• Δ
• ⁢
• t
• |
• ,
• Wherein, when 0<E<0.2, the membrane treatment system is determined to be normal;
• when 0.2≤E<0.4, the membrane treatment system is determined to be slightly fouled, corresponding to a slight fouling warning level;
• and when E≥0.4, the membrane treatment system is determined to be seriously fouled, corresponding to a serious fouling warning level;
• an early warning unit, configured to issue an early warning signal based on the determined warning level;
• a fourth acquiring unit, configured to acquire the influent water quality parameters of influent samples under different influent water quality conditions to form a sample set of the influent water quality parameters;
• a fifth acquiring unit, configured to acquire a data set of the electrochemical information values of the corresponding membrane treatment system changed with respect to time under the different influent water quality conditions to form a sample set of the electrochemical information values;
• a sample combining unit, configured to combine the sample set of the influent water quality parameters with the sample set of the electrochemical information values to create a combined sample set;
• a key water quality factor extraction unit, configured to characterize, based on the combined sample set, the influent water quality parameters therein to extract key water quality factors in the influent water quality parameters that cause significant changes in the electrochemical information values of the membrane treatment system with respect to time, a training unit, configured to train an original model by using the extracted key water quality factors and moments as independent variables and using the electrochemical information values, which correspond to each moment and are used to characterize the fouling level of the membrane treatment system, as dependent variables;
• and
• a model evaluation unit, configured to perform model evaluation for a model obtained by the training to obtain the membrane fouling prediction model,
• wherein the electrochemical information values are impedance mode values, or interface resistance values and capacitance values.