Dynamic dormancy water purification equipment control method based on AI prediction

Through AI prediction technology, the LSTM network model is used to analyze historical water use data, accurately predict future water consumption and adjust the dormant state of water purification equipment, solving the problems of energy waste and shortening of equipment life in traditional methods, and achieving efficient and energy-saving and convenient purification Water equipment control.

CN118915457APending Publication Date: 2024-11-08CHENGDU QINGYI TECH CO LTD +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410978926.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-22
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

The dynamic dormant control method of traditional water purification equipment lacks flexibility and foresight, and cannot accurately reflect users' water needs, resulting in waste of energy and shortening of equipment service life.

Method used

Using deep learning technology based on AI, we predict future water consumption through LSTM network model, adjust the dormant time point of the water purification equipment, and realize intelligent control.

Benefits of technology

Effectively reduce energy consumption, extend the service life of the equipment, improve system operation efficiency and user experience, providing an innovative and practical energy-saving technology.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118915457A_ABST
    Figure CN118915457A_ABST
Patent Text Reader

Abstract

The invention discloses a dynamic dormancy water purification equipment control method based on AI prediction, and the method comprises the following steps: S1, collecting historical water consumption data, cleaning and preprocessing the collected data, and obtaining a preprocessed historical water consumption data sequence; s2, inputting the preprocessed historical water data sequence into a multi-layer LSTM network model, and performing model training by adopting a forward propagation algorithm, a back propagation algorithm and a dropout regularization technology to obtain an LSTM prediction model; and S3, predicting the water consumption of the water purification equipment in a future period of time according to the LSTM prediction model, and adjusting the sleep time point of the water purification equipment according to a prediction result. The water consumption of the water purification equipment can be predicted in real time, and the dormant state is intelligently adjusted according to the prediction result, so that the energy consumption is effectively reduced, the service life of the equipment is prolonged, and the operation efficiency of the whole system is improved on the premise of ensuring the user requirements.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of water dispensers, and in particular relates to a control method for a dynamic dormant water purification device based on AI prediction. Background Art

[0002] As people's requirements for drinking water quality continue to increase, water purification equipment has become an indispensable part of modern drinking water scenarios. However, water purification equipment consumes a lot of energy during operation, and long-term uninterrupted operation will accelerate the wear of the equipment and shorten its service life. Therefore, how to effectively control the dormancy and startup of water purification equipment has become the key to improving equipment energy efficiency and extending its service life.

[0003] In water purification equipment management, traditional dynamic dormancy systems mainly rely on fixed time intervals or simple thresholds to control the operation and dormancy of equipment. However, this method has obvious limitations in practical applications. Due to the diversity and dynamic changes in user water usage patterns, fixed time intervals or thresholds often cannot accurately reflect the actual needs of users. This may cause the equipment to continue running when the user does not need water, resulting in energy waste; while when the user urgently needs water, the equipment may be in a shut-down state due to reaching the dormancy condition, causing inconvenience to the user.

[0004] More importantly, traditional methods lack the ability to predict future water consumption and cannot intelligently adjust the dormancy strategy based on historical data and current water consumption trends. This makes the dormancy control of the equipment inflexible and inconvenient, making it difficult to meet the modern drinking water scenario's demand for efficient, energy-saving, and convenient water purification equipment.

[0005] In order to solve this problem, the present invention proposes a dynamic dormant water purification equipment control method based on AI prediction. By introducing deep learning technology, the water consumption in the future can be accurately predicted. According to the prediction results, the equipment can intelligently adjust the dormancy strategy to achieve maximum energy conservation and extend the service life of the equipment while ensuring user needs. This method not only improves the energy efficiency of water purification equipment, but also optimizes the user experience, bringing an innovative and practical energy-saving technology to the field of water purification equipment. Summary of the invention

[0006] In view of the above-mentioned deficiencies in the prior art, the present invention provides a dynamic dormant water purification equipment control method based on AI prediction to solve the problem of energy waste caused by the continuous operation of the existing water purification equipment during idle periods.

[0007] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a dynamic dormant water purification equipment control method based on AI prediction, comprising the following steps:

[0008] S1. Collect historical water use data, clean and preprocess the collected data, and obtain a preprocessed historical water use data sequence;

[0009] S2. Input the preprocessed historical water consumption data sequence into the multi-layer LSTM network model, use the forward propagation algorithm, the back propagation algorithm and the dropout regularization technology to train the model, and obtain the LSTM prediction model;

[0010] S3. Predict the water consumption of the water purification equipment in the future based on the LSTM prediction model, and adjust the sleep time point of the water purification equipment according to the prediction results.

[0011] Further: in S2, the multi-layer LSTM network model includes an input layer, an LSTM layer, a dropout layer, a fully connected layer and an output layer connected in sequence;

[0012] Wherein, the LSTM layer includes a plurality of LSTM units, each of which updates the cell state and the hidden state through a gating mechanism;

[0013] The input layer is used to input the preprocessed historical water use data sequence, the LSTM layer is used to capture the time dependency in the preprocessed historical water use data sequence, the dropout layer is used to randomly set the outputs of some neurons to zero during the training process to prevent the model from overfitting, the fully connected layer is used to integrate the outputs of the LSTM layer to generate prediction results, and the output layer is used to output the prediction results.

[0014] Further: the gating mechanism includes an input gate, a forget gate and an output gate, and the expression of the input gate is specifically:

[0015] I t =σ(W xi ·X t +W hi ·H t-1 +b i )

[0016] In the formula, I t is the output of the input gate at time t, σ(·) is the Sigmoid activation function, W xi Enter X for time t t The corresponding first weight matrix, W hi is the hidden state H at the previous moment t-1 The corresponding second weight matrix, b i is the bias term of the input gate;

[0017] The expression of the forget gate is specifically:

[0018] F t =σ(Wxf ·X t +W hf ·H t-1 +b f )

[0019] In the formula, F t is the output of the forget gate at time t, W xf Enter X for time t t The corresponding third weight matrix, W hf is the hidden state H at the previous moment t-1 The corresponding fourth weight matrix, b f is the bias term of the forget gate;

[0020] The expression of the output gate is specifically:

[0021] O t =σ(W xo ·X t +W ho ·H t-1 +b o )

[0022] In the formula, O t is the output of the output gate at time t, W xo Enter X for time t t The corresponding fifth weight matrix, W ho is the hidden state H at the previous moment t-1 The corresponding sixth weight matrix, b o is the bias term of the output gate;

[0023] The expression of the cell state is specifically:

[0024] C t =F t ⊙C t-1 +I t ⊙tanh(W xc X t +W hc ·H t-1 +b c )

[0025] In the formula, C t is the cell state at time t, ⊙ is element-by-element multiplication, tanh(·) is the hyperbolic tangent function, W xc Enter X for time t t The corresponding seventh weight matrix, W hc is the hidden state H at the previous moment t-1 The corresponding eighth weight matrix, b c is the bias term of the cell state;

[0026] The expression of the hidden state is specifically:

[0027] H t =O t ⊙tanh(C t )

[0028] In the formula, H t is the hidden state at time t.

[0029] Further: S2 comprises the following sub-steps:

[0030] S21, input the preprocessed historical water use data sequence into the multi-layer LSTM network model, and calculate the predicted value of the multi-layer LSTM network model output according to the current model parameters and input data through the forward propagation algorithm;

[0031] S22. Calculate the loss function according to the predicted value and the true value, calculate the gradient of the loss function with respect to the model parameters by the back propagation algorithm, and update the weights and bias items of the multi-layer LSTM network model according to the gradient of the model parameters by the gradient descent algorithm;

[0032] S23. Introducing the dropout regularization technology in the process of updating the weights and bias items of the multi-layer LSTM network model. In the process of calculating the predicted value of the output of the multi-layer LSTM network model by the forward propagation algorithm, randomly selecting neurons in the dropout layer and setting their outputs to zero by the dropout regularization technology, wherein the neurons whose outputs are not set to zero are used to participate in calculating the gradient of the loss function with respect to the model parameters.

[0033] S24. Train the multi-layer LSTM network model through iterative forward propagation algorithm, back propagation algorithm and dropout regularization technology to generate an LSTM prediction model.

[0034] The beneficial effect of the above further scheme is: the present invention trains a multi-layer LSTM network model using forward propagation, back propagation and dropout regularization technology to construct a robust and generalized LSTM prediction model for accurately predicting water consumption.

[0035] Further: the S21 is specifically:

[0036] The preprocessed historical water use data sequence is input into the input layer and LSTM layer in turn. The cell state and hidden state are updated by the LSTM layer according to the input at the current moment, the cell state and hidden state at the previous moment, and the calculation result of the gating mechanism to obtain the output of the LSTM layer. The output of the LSTM layer is transmitted to the dropout layer and the fully connected layer in turn to obtain the predicted value of the output of the multi-layer LSTM network model.

[0037] Further: In S22, the expression of the loss function L is specifically:

[0038]

[0039] Where N is the number of samples, y i is the true value of sample i, is the model’s predicted value for sample i;

[0040] In S22, the gradient descent algorithm updates the weights of the multi-layer LSTM network model through a gradient descent mathematical model, and the expression of the gradient descent mathematical model is specifically:

[0041]

[0042] In the formula, is the updated weight matrix, W is the weight matrix before the update, η is the learning rate, is the gradient of the loss function with respect to the new weight matrix.

[0043] The beneficial effect of the above further scheme is that the present invention uses a gradient descent algorithm to update the weights and bias terms of the model. By continuously iterating this process, the prediction error can be gradually reduced and the performance of the model can be optimized.

[0044] Further: In S23, the dropout regularization technique sets the output of the neuron through a dropout mathematical model, and the expression of the dropout mathematical model is specifically:

[0045] r j ~Bernoulli(p)

[0046] y j =r j *x j

[0047] In the formula, r j is a random variable that follows Bernoulli distribution (·), with probability p taking 1, y j is the neuron output after dropout is applied, x j is the original output of the neuron.

[0048] The beneficial effect of the above further scheme is that the dropout layer randomly selects a part of neurons and sets their output to zero, so that the model can learn different feature combinations in each iteration, thereby preventing the model from over-relying on specific features.

[0049] The beneficial effects of the present invention are as follows: the present invention provides a method for controlling a dynamic dormant water purification device based on AI prediction, which uses AI technology to analyze and predict the user's water consumption. By deeply learning the user's historical water consumption data, the LSTM prediction model can capture the complex pattern of water consumption changes over time, thereby accurately predicting the water consumption in the future, and intelligently adjusting the dormant state according to the prediction results, thereby effectively reducing energy consumption, extending the service life of the equipment, and improving the overall system operation efficiency and user experience while ensuring user needs, bringing an innovative and practical energy-saving technology to the field of water purification equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a flow chart of a method for controlling a dynamic dormant water purification device based on AI prediction according to the present invention. DETAILED DESCRIPTION

[0051] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0052] like Figure 1 As shown, in one embodiment of the present invention, a dynamic dormant water purification equipment control method based on AI prediction includes the following steps:

[0053] S1. Collect historical water use data, clean and preprocess the collected data, and obtain a preprocessed historical water use data sequence;

[0054] S2. Input the preprocessed historical water consumption data sequence into the multi-layer LSTM network model, use the forward propagation algorithm, the back propagation algorithm and the dropout regularization technology to train the model, and obtain the LSTM prediction model;

[0055] S3. Predict the water consumption of the water purification equipment in the future based on the LSTM prediction model, and adjust the sleep time point of the water purification equipment according to the prediction results.

[0056] The present invention introduces the LSTM model in AI to accurately predict future water consumption and automatically adjust the equipment to a dormant state during the predicted low water consumption period, so as to achieve the purpose of energy saving and extend the service life of the equipment.

[0057] In S1, historical water usage data including timestamps and water consumption information are collected, and the collected data are cleaned and preprocessed to correct abnormal values ​​and fill in missing data, thereby ensuring the accuracy and completeness of the data.

[0058] In this embodiment, the present invention plans the model architecture in detail when constructing and training the LSTM prediction model, and uses the standard gradient descent algorithm for weight optimization. At the same time, in order to improve the generalization ability of the model, the dropout regularization technology is also introduced. The present invention uses a deep learning framework to construct a multi-layer LSTM network model. In S2, the multi-layer LSTM network model includes an input layer, an LSTM layer, a dropout layer, a fully connected layer, and an output layer connected in sequence;

[0059] Wherein, the LSTM layer includes a plurality of LSTM units, each of which updates the cell state and the hidden state through a gating mechanism;

[0060] The input layer is used to input the preprocessed historical water use data sequence, the LSTM layer is used to capture the time dependency in the preprocessed historical water use data sequence, the dropout layer is used to randomly set the outputs of some neurons to zero during the training process to prevent the model from overfitting. This can prevent the model from being overly dependent on certain specific neurons during training, thereby improving the generalization ability of the model, the fully connected layer is used to integrate the output of the LSTM layer to generate a prediction result, and the output layer is used to output the prediction result.

[0061] The gating mechanism includes an input gate, a forget gate and an output gate. The expression of the input gate is specifically:

[0062] I t =σ(W xi ·X t +W hi ·H t-1 +b i )

[0063] In the formula, I t is the output of the input gate at time t, σ(·) is the Sigmoid activation function, W xi Enter X for time t t The corresponding first weight matrix, W hi is the hidden state H at the previous moment t-1 The corresponding second weight matrix, b i is the bias term of the input gate;

[0064] The expression of the forget gate is specifically:

[0065] F t =σ(W xf ·Xt +W hf ·H t-1 +b f )

[0066] In the formula, F t is the output of the forget gate at time t, W xf Enter X for time t t The corresponding third weight matrix, W hf is the hidden state H at the previous moment t-1 The corresponding fourth weight matrix, b f is the bias term of the forget gate;

[0067] The expression of the output gate is specifically:

[0068] O t =σ(W xo ·X t +W ho ·H t-1 +b o )

[0069] In the formula, O t is the output of the output gate at time t, W xo Enter X for time t t The corresponding fifth weight matrix, W ho is the hidden state H at the previous moment t-1 The corresponding sixth weight matrix, b o is the bias term of the output gate;

[0070] The expression of the cell state is specifically:

[0071] C t =F t ⊙C t-1 +I t ⊙tanh(W xc X t +W hc ·H t-1 +b c )

[0072] In the formula, C t is the cell state at time t, ⊙ is element-by-element multiplication, tanh(·) is the hyperbolic tangent function, W xc Enter X for time t t The corresponding seventh weight matrix, W hc is the hidden state H at the previous moment t-1 The corresponding eighth weight matrix, b c is the bias term of the cell state;

[0073] The expression of the hidden state is specifically:

[0074] H t =O t ⊙tanh(C t )

[0075] In the formula, H t is the hidden state at time t.

[0076] The S2 comprises the following sub-steps:

[0077] S21, input the preprocessed historical water use data sequence into the multi-layer LSTM network model, and calculate the predicted value of the multi-layer LSTM network model output according to the current model parameters and input data through the forward propagation algorithm;

[0078] S22. Calculate the loss function according to the predicted value and the true value, calculate the gradient of the loss function with respect to the model parameters by the back propagation algorithm, and update the weights and bias items of the multi-layer LSTM network model according to the gradient of the model parameters by the gradient descent algorithm;

[0079] S23. Introducing the dropout regularization technology in the process of updating the weights and bias items of the multi-layer LSTM network model. In the process of calculating the predicted value of the output of the multi-layer LSTM network model by the forward propagation algorithm, randomly selecting neurons of the dropout layer and setting their outputs to zero by the dropout regularization technology, wherein the neurons whose outputs are not set to zero are used to participate in calculating the gradient of the loss function with respect to the model parameters, and updating the weights and bias items of the multi-layer LSTM network model in the back propagation algorithm.

[0080] S24. Train the multi-layer LSTM network model through iterative forward propagation algorithm, back propagation algorithm and dropout regularization technology to generate an LSTM prediction model.

[0081] In this embodiment, the present invention trains a multi-layer LSTM network model using forward propagation, back propagation and dropout regularization techniques to construct a robust and generalized LSTM prediction model for accurately predicting water consumption.

[0082] The S21 is specifically:

[0083] The preprocessed historical water use data sequence is input into the input layer and LSTM layer in turn. The cell state and hidden state are updated by the LSTM layer according to the input at the current moment, the cell state and hidden state at the previous moment, and the calculation result of the gating mechanism to obtain the output of the LSTM layer. The output of the LSTM layer is transmitted to the dropout layer and the fully connected layer in turn to obtain the predicted value of the output of the multi-layer LSTM network model.

[0084] In this embodiment, the forward propagation algorithm calculates the predicted value of the LSTM network model output based on the current model parameters and input data.

[0085] In S22, the expression of the loss function L is specifically:

[0086]

[0087] Where N is the number of samples, y i is the true value of sample i, is the model’s predicted value for sample i;

[0088] In S22, the gradient descent algorithm updates the weights of the multi-layer LSTM network model through a gradient descent mathematical model, and the expression of the gradient descent mathematical model is specifically:

[0089]

[0090] In the formula, is the updated weight matrix, W is the weight matrix before the update, η is the learning rate, is the gradient of the loss function with respect to the new weight matrix.

[0091] In this embodiment, the present invention uses a gradient descent algorithm to update the weights and bias items of the model. By continuously iterating this process, the prediction error can be gradually reduced and the performance of the model can be optimized.

[0092] In S23, the dropout regularization technique sets the output of the neuron through a dropout mathematical model, and the expression of the dropout mathematical model is specifically:

[0093] r j ~Bernoulli(p)

[0094] y j =r j *x j

[0095] In the formula, r j is a random variable that follows Bernoulli distribution (·), with probability p taking 1, y j is the neuron output after dropout is applied, x j is the original output of the neuron.

[0096] In this embodiment, the dropout layer randomly selects a portion of neurons and sets their outputs to zero, so that the model can learn different feature combinations in each iteration, thereby preventing the model from being overly dependent on specific features.

[0097] In S3, the LSTM prediction model is used to make a real-time prediction of the water consumption of the water purification equipment in the future. According to the prediction results, the sleep time of the water purification equipment is adjusted dynamically and intelligently through the sleep strategy adjustment. Sleep strategy adjustment: when the predicted water consumption is low, the sleep time is extended to reduce energy consumption; when the predicted water consumption increases, the equipment is woken up in advance to meet the water demand.

[0098] The beneficial effects of the present invention are as follows: the present invention provides a method for controlling a dynamic dormant water purification device based on AI prediction, which uses AI technology to analyze and predict the user's water consumption. By deeply learning the user's historical water consumption data, the LSTM prediction model can capture the complex pattern of water consumption changes over time, thereby accurately predicting the water consumption in the future, and intelligently adjusting the dormant state according to the prediction results, thereby effectively reducing energy consumption, extending the service life of the equipment, and improving the overall system operation efficiency and user experience while ensuring user needs, bringing an innovative and practical energy-saving technology to the field of water purification equipment.

[0099] In the description of the present invention, it is necessary to understand that the orientation or positional relationship indicated by the terms "center", "thickness", "upper", "lower", "horizontal", "top", "bottom", "inner", "outer", "radial", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first", "second", and "third" are used only for descriptive purposes, and cannot be understood as indicating or implying the relative importance or the number of implicitly specified technical features. Therefore, the features defined by "first", "second", and "third" may explicitly or implicitly include one or more of the features.

Claims

1. A dynamic dormant water purification equipment control method based on AI prediction, characterized in that: The following steps are involved: S1. Collect historical water use data, clean and preprocess the collected data, and obtain a preprocessed historical water use data sequence; S2. Input the preprocessed historical water consumption data sequence into the multi-layer LSTM network model, use the forward propagation algorithm, the back propagation algorithm and the dropout regularization technology to train the model, and obtain the LSTM prediction model; S3. Predict the water consumption of the water purification equipment in the future based on the LSTM prediction model, and adjust the sleep time point of the water purification equipment according to the prediction results.

2. The method for controlling a dynamic dormant water purification device based on AI prediction according to claim 1 is characterized in that: In S2, the multi-layer LSTM network model includes an input layer, an LSTM layer, a dropout layer, a fully connected layer and an output layer connected in sequence; Wherein, the LSTM layer includes a plurality of LSTM units, each of which updates the cell state and the hidden state through a gating mechanism; The input layer is used to input the preprocessed historical water use data sequence, the LSTM layer is used to capture the time dependency in the preprocessed historical water use data sequence, the dropout layer is used to randomly set the outputs of some neurons to zero during the training process to prevent the model from overfitting, the fully connected layer is used to integrate the outputs of the LSTM layer to generate prediction results, and the output layer is used to output the prediction results.

3. The dynamic dormant water purification equipment control method based on AI prediction according to claim 2 is characterized in that: The gating mechanism includes an input gate, a forget gate and an output gate. The expression of the input gate is specifically: I t =σ(W xi ·X t +W hi ·H t-1 +b i ) In the formula, I t is the output of the input gate at time t, σ(·) is the Sigmoid activation function, W xi Enter X for time t t The corresponding first weight matrix, W hi is the hidden state H at the previous moment t-1 The corresponding second weight matrix, b i is the bias term of the input gate; The expression of the forget gate is specifically: F t =σ(W xf ·X t +W hf ·H t-1 +b f ) In the formula, F t is the output of the forget gate at time t, W xf Enter X for time t t The corresponding third weight matrix, W hf is the hidden state H at the previous moment t-1 The corresponding fourth weight matrix, b f is the bias term of the forget gate; The expression of the output gate is specifically: O t =σ(W xo ·X t +W ho ·H t-1 +b o ) In the formula, O t is the output of the output gate at time t, W xo Enter X for time t t The corresponding fifth weight matrix, W ho is the hidden state H at the previous moment t-1 The corresponding sixth weight matrix, b o is the bias term of the output gate; The expression of the cell state is specifically: C t =F t ☉C t-1 +I t ☉tanh(W xc X t +W hc ·H t-1 +b c ) In the formula, C t is the cell state at time t, ⊙ is element-by-element multiplication, tanh(·) is the hyperbolic tangent function, W xc Enter X for time t t The corresponding seventh weight matrix, W hc is the hidden state H at the previous moment t-1 The corresponding eighth weight matrix, b c is the bias term of the cell state; The expression of the hidden state is specifically: H t =O t ☉tanh(C t ) In the formula, H t is the hidden state at time t.

4. The dynamic dormant water purification equipment control method based on AI prediction according to claim 3 is characterized in that: The S2 comprises the following sub-steps: S21, input the preprocessed historical water use data sequence into the multi-layer LSTM network model, and calculate the predicted value of the multi-layer LSTM network model output according to the current model parameters and input data through the forward propagation algorithm; S22. Calculate the loss function according to the predicted value and the true value, calculate the gradient of the loss function with respect to the model parameters by the back propagation algorithm, and update the weights and bias items of the multi-layer LSTM network model according to the gradient of the model parameters by the gradient descent algorithm; S23. Introducing the dropout regularization technology in the process of updating the weights and bias items of the multi-layer LSTM network model. In the process of calculating the predicted value of the output of the multi-layer LSTM network model by the forward propagation algorithm, randomly selecting neurons in the dropout layer and setting their outputs to zero by the dropout regularization technology, wherein the neurons whose outputs are not set to zero are used to participate in calculating the gradient of the loss function with respect to the model parameters. S24. Train the multi-layer LSTM network model through iterative forward propagation algorithm, back propagation algorithm and dropout regularization technology to generate an LSTM prediction model.

5. The method for controlling a dynamic dormant water purification device based on AI prediction according to claim 4 is characterized in that: The S21 is specifically: The preprocessed historical water use data sequence is input into the input layer and LSTM layer in turn. The cell state and hidden state are updated by the LSTM layer according to the input at the current moment, the cell state and hidden state at the previous moment, and the calculation result of the gating mechanism to obtain the output of the LSTM layer. The output of the LSTM layer is transmitted to the dropout layer and the fully connected layer in turn to obtain the predicted value of the output of the multi-layer LSTM network model.

6. The method for controlling a dynamic dormant water purification device based on AI prediction according to claim 4 is characterized in that: In S22, the expression of the loss function L is specifically: Where N is the number of samples, y i is the true value of sample i, is the predicted value of the model for sample i; In S22, the gradient descent algorithm updates the weights of the multi-layer LSTM network model through a gradient descent mathematical model, and the expression of the gradient descent mathematical model is specifically: In the formula, is the updated weight matrix, W is the weight matrix before the update, η is the learning rate, is the gradient of the loss function with respect to the new weight matrix.

7. The method for controlling a dynamic dormant water purification device based on AI prediction according to claim 4 is characterized in that: In S23, the dropout regularization technique sets the output of the neuron through a dropout mathematical model, and the expression of the dropout mathematical model is specifically: r j ~Bernoulli(p) y j =r j *x j In the formula, r j is a random variable that follows Bernoulli distribution (·), with probability p taking 1, y j is the neuron output after dropout is applied, x j is the original output of the neuron.