A home energy control method based on artificial emotion time series Q learning

Abstract

The application provides a household energy control method based on artificial emotion time sequence Q learning, which is used for realizing self-management of household electricity use by users and comprises two parts, namely, an artificial emotion deep neural network and a long short-term memory network. Firstly, the result of artificial emotion quantization is delivered to the long short-term memory network after being processed by the emotion deep network structure, the artificial emotion deep neural network is continuously trained in practice by using Q learning, the result of the Q learning training is transmitted to the long short-term memory network, the trained long short-term memory network and the predicted target or actual value of the previous time step are used as inputs, and the predicted value of the current time step is obtained, so that the intelligent household energy management can be predicted. The method can realize real-time monitoring and management of household electricity use in a humanized manner.

Description

Technical Field

[0001] This invention belongs to the field of home energy management in new energy, power systems, new power systems, and smart grids, and relates to home energy control methods applicable to home energy control. Background Technology

[0002] Many regions have to implement regional power rationing during peak electricity consumption periods to alleviate the pressure on urban power supply due to overloaded urban electricity demand. In this situation, user-driven management and energy conservation become particularly important. To enable household users to manage their electricity usage independently, achieve two-way interaction between the power grid and users, and ensure the implementation of demand-side response, household energy control methods need to be proposed.

[0003] Traditional home energy control suffers from several problems. First, traditional home energy control systems are typically fixed, lacking intelligence and personalization, and cannot adapt to the behavior and preferences of family members, leading to energy waste and inefficient use. Second, traditional home energy control systems usually require the installation of sensors and other devices to acquire energy data, but this method is costly and involves a complex installation process. Furthermore, traditional home energy control systems often rely on rule-based or experience-based methods for energy prediction, but these methods are prone to errors and uncertainties, making it difficult to accurately predict household energy usage. Additionally, traditional home energy control systems typically rely on manual intervention for energy control, but this method has limitations and shortcomings, failing to achieve intelligent and automated energy control. Finally, the adoption rate of traditional home energy control systems is low, often only used by a small number of households, and therefore cannot have a significant impact on energy utilization. These problems make it difficult for traditional home energy control systems to effectively control and utilize energy, to adapt well to living needs and energy usage, and to have a relatively low significant impact on energy utilization, making it difficult to achieve the goals of high efficiency, intelligence, and energy conservation.

[0004] The Artificial Emotional Long Short-Term Memory (ALSTM) network algorithm is a deep learning-based artificial neural network model that effectively extracts features when processing time-series data and builds a LSSTM model to predict future data trends. This algorithm shows promising application prospects in smart home energy management, enabling it to predict household energy consumption, optimize energy usage plans, and achieve intelligent, efficient, and energy-saving energy management goals. Therefore, researching smart home energy control based on the ALSTM network algorithm has both theoretical significance and practical application value. It can not only improve the energy efficiency and energy-saving effect of smart homes but also contribute to the intelligent construction and sustainable development of smart homes. Summary of the Invention

[0005] A family energy control method based on artificial emotion time series Q-learning combines artificial emotion, long short-term memory networks, deep learning, and Q-learning for family energy control. It can effectively extract features when processing time series data and can build a long short-term memory model to predict future data trends. The steps in its application are as follows:

[0006] Step (1) Preprocess the data by collecting the user's historical electricity consumption as the input sequence x. t ;

[0007] Input sequence x t The input sequence is fed into a multilayer perceptron to extract features, resulting in a feature vector v.

[0008] Input sequence x t The output v of the multilayer perceptron t for:

[0009] v t =f(W1*x t +b1), 1≤t≤T

[0010] In the formula, W1 and b1 are the weights and bias parameters of the multilayer perceptron, f represents the activation function; * represents matrix multiplication; T is the length of the sequence;

[0011] The outputs v of a total of T multilayer perceptrons t The feature vector v is formed;

[0012] Step (2) Set up the sentiment classification, and feed the feature vector v into the sentiment classifier to classify the sentiment and obtain the sentiment representation vector e;

[0013] A sentiment classifier is a classifier or a combination of classifiers, and the sentiment representation vector e is:

[0014] e = g(W² * v + b²)

[0015] In the formula, W2 and b2 are the weights and bias parameters of the sentiment classifier, and g represents the activation function;

[0016] Step (3) combines the emotion representation vector e and the input sequence x t Input the data into the prediction model to predict electricity consumption;

[0017] The input to the prediction model is the input sequence x. t Given the sentiment representation vector e, the prediction model outputs the predicted electricity consumption value y for the current time step. t ;

[0018] Use h t c represents the hidden state vector of an artificial emotion long short-term memory network. tRepresents the cell state vector of the artificial emotional long short-term memory network; the input gate i of the artificial emotional long short-term memory network. t for:

[0019] i t =σ(W i *[h t-1 ,x t ,e]+b i )

[0020] In the formula, W i It is the weight matrix of the input gate, b i It is the bias vector of the input gate, σ represents the sigmoid function, [h t-1 ,x t [e] indicates that h t-1 ,x t Concatenate e to form a vector;

[0021] Forgetting gates in artificial emotional long short-term memory networks t for:

[0022] f t =σ(W f *[h t-1 ,x t ,e]+b f )

[0023] In the formula, W f This is the weight matrix of the forget gate, b f It is the bias vector of the forget gate;

[0024] Cellular states of artificial emotion long short-term memory networks from cell state c t-1 Update to cell state c t The formula is:

[0025] c t =f t *c t-1 +i t *tanh(W c *[h t-1 ,x t ,e]+b c )

[0026] In the formula, W c It is the weight matrix for cell state updates, b c It is the bias vector for cell state updates, and tanh represents the hyperbolic tangent function;

[0027] Output gate of artificial emotional long short-term memory network t for:

[0028] o t=σ(W O *[h t-1 ,x t ,e]+b o )

[0029] In the formula, W O It is the weight matrix of the output gate, b o It is the bias vector of the output gate;

[0030] The hidden state of the artificial emotional long short-term memory network is updated as follows:

[0031] h t =o t *tanh(c t )

[0032] Step (4) The home energy control center uses the predicted electricity consumption value y at the current time step. t The Q-learning method is used to give action commands to control the on / off state and power of household appliances, thereby controlling household energy.

[0033] The present invention has the following advantages and effects compared with the prior art:

[0034] (1) Historical electricity consumption data and weather data are concatenated into an input sequence in chronological order. For sentiment analysis, an artificial sentiment Q-learning algorithm is used to obtain a sentiment representation vector. For smart home energy management prediction, a deep neural network and a long short-term memory network are used for sequence modeling to predict household energy consumption.

[0035] (2) The emotional representation vector is fused with the prediction results of family energy to improve the prediction accuracy. At the same time, the proposed method incorporates emotional factors into the deep neural network for sentiment analysis, aiming to better explore the intrinsic characteristics of family energy data and thus improve prediction accuracy.

[0036] (3) Long Short-Term Memory Networks, as a sequence model, can better capture the dependencies between time series and improve the accuracy of household energy prediction.

[0037] (4) With the combination of artificial emotion Q learning, emotion deep neural network and long short-term memory network, the proposed method can predict smart home energy management well. Attached Figure Description

[0038] Figure 1 This is a flowchart of the artificial emotion Q-learning method of the present invention.

[0039] Figure 2 This is a flowchart of the training process of the emotional deep neural network of the present invention.

[0040] Figure 3 This is a flowchart of the method of the present invention for predicting energy in a smart home using a long short-term memory network. Detailed Implementation

[0041] This invention proposes a family energy control method based on artificial emotional time series Q-learning, which is described in detail below with reference to the accompanying drawings:

[0042] Figure 1 This is a block diagram of the artificial emotion Q-learning method of this invention. The idea behind the Q-learning algorithm is to construct a Q-value sheet based on the state and actions, store the Q-values, and then select actions that maximize benefits based on the Q-values. Integrating the Q-learning algorithm into artificial emotions yields a smart home energy management scheme that is system-managed and imbued with human emotions. By rationally selecting the size of the Q-value, the allocation and utilization of family energy can be further optimized. The intelligent agent (user), environmental conditions, reward (user's emotional value), and actions can all be abstracted into a Markov decision process.

[0043] Figure 2 This is a flowchart illustrating the training process of the emotional deep neural network in this invention. The emotional deep neural network predicts the effectiveness of home energy control by modeling and analyzing users' emotional states and behaviors. Home energy control is an important task, as it can reduce environmental impact and lower energy costs by controlling home energy usage. The multi-layered neural network of the emotional deep neural network models the input emotional state and then optimizes the network parameters through a backpropagation algorithm, enabling it to better predict the effectiveness of smart home energy control.

[0044] Figure 3 This is a flowchart of the method for predicting smart home energy using a long short-term memory network (LSTM) according to the present invention. The basic process for predicting home energy using an LTM network is as follows: first, data is collected; then, data preprocessing is performed; then, feature values ​​are extracted; a LTM network model is established and trained; and finally, prediction and control are performed.

[0045] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A family energy control method based on artificial emotional time series Q-learning, characterized in that, This method combines artificial emotions, long short-term memory networks, deep learning, and Q-learning for the control of family energy; it can effectively extract features when processing time-series data and can build long short-term memory models to predict future data trends. The steps during use are as follows: Step (1) Preprocess the data by collecting the user's historical electricity consumption as the input sequence x. t ; Input sequence x t The input sequence is fed into a multilayer perceptron to extract features, resulting in a feature vector v. Input sequence x t The output v of the multilayer perceptron t for: in t =f(W1*x t +b1),1≤t≤T In the formula, W1 and b1 are the weights and bias parameters of the multilayer perceptron, and f represents the activation function; * indicates matrix multiplication; T is the length of the sequence; The outputs v of a total of T multilayer perceptrons t The feature vector v is formed; Step (2) Set up the sentiment classification, and feed the feature vector v into the sentiment classifier to classify the sentiment and obtain the sentiment representation vector e; A sentiment classifier is a classifier or a combination of classifiers, and the sentiment representation vector e is: e = g(W² * v + b²) In the formula, W2 and b2 are the weights and bias parameters of the sentiment classifier, and g represents the activation function; Step (3) combines the emotion representation vector e and the input sequence x t Input the data into the prediction model to predict electricity consumption; The input to the prediction model is the input sequence x. t Given the sentiment representation vector e, the prediction model outputs the predicted electricity consumption value y for the current time step. t ; Use h t c represents the hidden state vector of an artificial emotion long short-term memory network. t Represents the cell state vector of an artificial emotional long short-term memory network; Input gate i of artificial emotional long short-term memory network t for: I t =σ(W i *[h t-1 ,x t ,e]+b i ) In the formula, W i It is the weight matrix of the input gate, b i It is the bias vector of the input gate, σ represents the sigmoid function, [h t-1 ,x t [e] indicates that h t-1 ,x t Concatenate e to form a vector; Forgetting gates in artificial emotional long short-term memory networks t for: f t =σ(W f *[h t-1 ,x t ,e]+b f ) In the formula, W f This is the weight matrix of the forget gate, b f It is the bias vector of the forget gate; Cellular states of artificial emotion long short-term memory networks from cell state c t-1 Update to cell state c t The formula is: c t =f t *c t-1 +i t *tanh(W c *[h t-1 ,x t ,e]+b c ) In the formula, W c It is the weight matrix for cell state updates, b c It is the bias vector for cell state updates, and tanh represents the hyperbolic tangent function; Output gate of artificial emotional long short-term memory network t for: the t =σ(W O *[h t-1 ,x t ,e]+b o ) In the formula, W O It is the weight matrix of the output gate, b o It is the bias vector of the output gate; The hidden state of the artificial emotional long short-term memory network is updated as follows: h t = no t *fish(c t ) Step (4) The home energy control center uses the predicted electricity consumption value y at the current time step. t The Q-learning method is used to give action commands to control the on / off state and power of household appliances, thereby controlling household energy.

Images