Intelligent zero-carbon-emission park building energy-saving control method and system and medium

Abstract

The invention provides an intelligent zero-carbon-emission park building energy-saving control method and system and a medium, which are used for optimal control of window light transmittance, illumination intensity and air conditioner operation power of a building. According to the system, future environment parameters (such as temperature and illumination intensity) are predicted through an LSTM model, and current environment data and historical control signals are combined and input into an MLP model to generate a real-time control strategy, so that the running states of windows, illumination and air conditioners are dynamically adjusted. The system can significantly improve the energy efficiency of a building, reduce carbon emission and optimize the indoor environment comfort, and has adaptive learning and automatic fault detection functions. The system is suitable for the fields of green buildings, intelligent buildings and the like, intelligent and automatic environment control can be achieved, and energy conservation, emission reduction and sustainable development of the buildings are promoted.

Description

Technical Field

[0001] The present invention relates to the field of environmental control technology, and in particular to an intelligent zero-carbon emission park building energy-saving control method, system and medium. Background Art

[0002] With the intensification of global environmental problems and the improvement of environmental protection awareness, building energy management and environmental control have become key links in modern building design. Traditional environmental control systems usually rely on preset rules or simple feedback control, which is difficult to cope with complex and changing external environments and user needs. These systems are often energy-efficient and lack adaptive capabilities. They cannot make optimal adjustments according to real-time environmental changes, resulting in energy waste and reduced user comfort.

[0003] In recent years, the development of intelligent control technology has provided new solutions for building environmental control. By introducing artificial intelligence and machine learning technologies, the building's environmental control system can make more accurate control decisions based on historical data and real-time environmental parameters. Among them, multi-layer perceptron (MLP) and long short-term memory network (LSTM), as typical models in deep learning, have gradually been widely used in intelligent control systems due to their powerful nonlinear modeling and time series prediction capabilities.

[0004] However, most existing intelligent control systems only optimize a single variable (such as temperature, light or air quality) and fail to achieve comprehensive control of multiple related variables (such as window transmittance, lighting intensity and air conditioning power). In addition, the existing system lacks real-time monitoring of the operating status of the equipment and fault self-detection functions, resulting in complex and costly system maintenance. Therefore, there is an urgent need for an intelligent control system that integrates LSTM and MLP technologies, which can comprehensively consider multiple environmental control variables, realize dynamic adjustment and adaptive optimization, and further improve the energy efficiency and environmental comfort of buildings. Summary of the invention

[0005] The present invention aims to at least solve the technical problems existing in the prior art, and particularly innovatively proposes an intelligent zero-carbon emission campus building energy-saving control system, system and medium, which utilizes an intelligent environmental control system based on LSTM and MLP to achieve intelligent and automated environmental regulation in buildings through multivariable joint control and adaptive learning, thereby greatly improving energy efficiency, reducing carbon emissions and enhancing user experience.

[0006] In order to achieve the above-mentioned object of the present invention, the present invention provides an intelligent zero-carbon emission park building energy-saving control system, comprising:

[0007] Data processing module: including temperature sensors, light sensors, smart window sensors, and power consumption metering equipment;

[0008] Deep learning module: Use multi-layer neural network combined with convolutional neural network to learn and train historical data to predict future indoor environment requirements;

[0009] Intelligent control module: Based on the prediction results of the deep learning model, it adjusts the light transmittance and thermal conductivity of the window in real time;

[0010] Central control system: Through the intelligent control module, it connects with the park's energy management system to optimize overall energy utilization.

[0011] In a preferred embodiment of the present invention, the data processing module includes:

[0012] Data collection: Deploy temperature sensors, light intensity sensors, and power metering sockets to continuously collect indoor temperature T(t), light intensity I(t), and energy consumption data P(t);

[0013] Data preprocessing: Clean and normalize the collected data to eliminate noise and outliers, and provide high-quality data sets for deep learning model training, including:

[0014] Use Z-score normalization:

[0015]

[0016]

[0017]

[0018] Among them, Z T (t), Z I (t) and Z P (t) are the processed temperature, light intensity and energy consumption data, which are used to fill missing values ​​and replace abnormal values;

[0019] μ T , μ I and μ P are the average values ​​of temperature, light intensity and energy consumption data, σ T , σ I and σ P are the standard deviations of temperature, light intensity, and energy consumption data, respectively;

[0020] The processed indoor temperature T(t), light intensity I(t) and energy consumption data P(t) are combined into the environmental data vector x t :

[0021]

[0022] x tis the environmental data vector at time t.

[0023] In a preferred embodiment of the present invention, the deep learning module includes:

[0024] Model training:

[0025] Use the combination of LSTM model and MLP model to train the data.

[0026] The LSTM model is used to predict the changes in light and temperature trends in the future, so as to help the system make adjustment decisions in advance. The MLP model is used to receive the prediction input of the LSTM model and the current system status, and generate control signals to optimize energy efficiency.

[0027] Data input:

[0028] X t =[x t-n ,x t-n+1 ,…,x t ];

[0029] Among them, X t The input time series data includes the environmental data vector x at the past n time points. t ;

[0030] x t-i is the environmental data vector at the ti-th moment, where the value range of the integer i is 0≤i≤n;

[0031] Forget Gate:

[0032] f t =σ(W f (t)·[h t-1 ,x t ]+b f );

[0033] f t is the forget gate at time t, W f (t) is the forget gate weight matrix at time t, h t-1 is the hidden state at time t-1, x t is the environmental data vector at time t, b f is the bias of the forget gate, σ(*) is the activation function;

[0034] Among them, the forget gate weight matrix W f (t) is updated as follows:

[0035]

[0036] Among them, ω is the angular frequency, which controls the period of the periodic term, φ is the phase offset, which determines the initial value of the periodic term, and cos(ωt+φ) is the periodic term, which prevents the update process from continuing in a fixed direction, thereby preventing the model from falling into a local optimal solution;

[0037] W f (t+1) is the weight matrix of the forget gate at time t+1, W f (t) is the weight matrix of the forget gate at time t, W i (t) is the weight matrix of the input gate at time t, W C (t) is the weight matrix of the candidate memory at time t, and ε is a constant used to prevent the denominator from being zero;

[0038] Input Gate:

[0039] i t =σ(W i (t)·[h t-1 ,x t ]+b i );

[0040]

[0041] Among them, i t is the input gate at time t, is the candidate memory at time t, W i (t) is the weight matrix of the input gate at time t, W C (t) is the weight matrix of the candidate memory at time t;

[0042] b i is the bias of the input gate, b C is the bias of the candidate memory;

[0043] tanh(*) is the hyperbolic tangent activation function;

[0044] Memory Update:

[0045]

[0046] C t is the cell state at time t, C t-1 is the cell state at time t-1;

[0047] Output Gate:

[0048] o t =σ(W o (t)·[h t-1 ,x t ]+b o );

[0049] h t=o t tanh(C t );

[0050] o t is the input gate at the current time t, W o (t) is the weight matrix of the output gate at time t, b o is the bias of the output gate, h t It is the output of the LSTM model, indicating the hidden state at the current moment;

[0051] The prediction output of the LSTM model:

[0052]

[0053] It represents the temperature and light prediction results at the t+kth moment, and LSTM(*) is the prediction output of the LSTM model.

[0054] In a preferred embodiment of the present invention, the deep learning module includes:

[0055] The MLP model receives the prediction results of LSTM And the environmental data vector x at time t t and historical control output to generate control signals to control air conditioning power, smart window transmittance and lighting intensity;

[0056] enter:

[0057]

[0058] M t is the input vector of MLP;

[0059] is the predicted output of the LSTM model at the t+kth moment;

[0060] x t is the environmental data vector at time t;

[0061] u t-1 is the control signal at time t-1;

[0062] MLP calculation formula:

[0063] h1=tanh(W1·M t +b1);

[0064] h2=tanh(W2·h1+b2);

[0065] u t =W3·h2+b3+γΔu t-1 ;

[0066] Where h1 and h2 represent the activation outputs of the first hidden layer and the second hidden layer, respectively; W1 and W2 are the weight matrices of the first hidden layer and the second hidden layer, respectively; b1 and b2 are the bias vectors of the first hidden layer and the second hidden layer, respectively;

[0067] γ is an adaptive function used to adjust the influence of the change in the control signal, Δu t-1 is the change in the control signal at time t-1;

[0068] Among them, the calculation method of the adaptive function γ is:

[0069]

[0070] W3 is the weight matrix of the output layer, b3 is the bias vector of the output layer, and u t is the generated control signal vector, including window transmittance, lighting intensity, and air conditioning power.

[0071] In a preferred embodiment of the present invention, the intelligent control module comprises:

[0072] Control signal vector u t The data structure is:

[0073] u t =[V t ,I t ,P t ];

[0074] V t Represents the voltage signal at the current time t, which is used to control the light transmittance B(V) of the smart window, I t represents the lighting control signal at the current time t, which is used to adjust the indoor lighting intensity, P t represents the air conditioning power command signal at the current time t, and adjusts the air conditioning power in percentage;

[0075] The relationship between the window transmittance B(V) and the applied voltage V is:

[0076]

[0077] Where, B(V) is the transmittance when voltage V is applied, B max is the maximum transmittance, V max Indicates the maximum voltage that the material can withstand. By adjusting the transmittance B(V), the air conditioning power P is adjusted. t and lighting control signal I t , saving power consumption without affecting actual operation;

[0078] α is the coefficient of transmittance change, which depends on the characteristics of the material itself and the specific electrochromic reaction. C is an exponential factor that describes the nonlinear degree of transmittance change with voltage.

[0079] The present invention also discloses an intelligent zero-carbon emission park building energy-saving control method, comprising the following steps:

[0080] S1, collects data from temperature sensors, light sensors, smart window sensors, and power consumption metering devices, and processes the data;

[0081] S2, uses a multi-layer neural network combined with a convolutional neural network to learn and train historical data to predict future indoor environmental requirements;

[0082] S3, based on the prediction results of the deep learning model, adjusts the light transmittance and thermal conductivity of the window in real time;

[0083] S4, through the intelligent control module, connects with the park’s energy management system to optimize overall energy utilization.

[0084] The present invention also discloses a computer system, comprising:

[0085] processor;

[0086] a memory for storing processor-executable instructions;

[0087] Among them, the processor is configured to implement the intelligent zero-carbon emission campus building energy-saving control system when executing the executable instructions.

[0088] The present invention also discloses a computer-readable storage medium, comprising:

[0089] a memory having a computer program stored thereon;

[0090] A processor is used to execute the program in the memory to implement the intelligent zero-carbon emission campus building energy-saving control system.

[0091] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0092] In the smart building energy-saving control system, the combination of long short-term memory network (LSTM) and multi-layer perceptron (MLP) has the following advantages in promoting the goal of zero carbon emissions: LSTM network, with its unique memory unit design, is very good at processing time series data and can capture long-term dependencies in the data. In the smart building environment, this means that LSTM accurately predicts the changing trend of environmental parameters such as temperature and light in the future, thereby reflecting the accuracy of time series prediction, while the MLP model is good at making decisions quickly based on current input data. In the smart building control system, MLP dynamically adjusts the control strategy based on the prediction results of LSTM and real-time environmental parameters such as temperature and light intensity. The flexibility of dynamic control is enhanced, further improving energy-saving efficiency.

[0093] By combining LSTM and MLP, more refined management of energy consumption can be achieved. LSTM provides accurate future environmental predictions, while MLP generates precise control instructions based on these predictions and real-time data. This refined management helps reduce energy waste, thereby promoting green certification and sustainable development of buildings.

[0094] In summary, the combination of LSTM and MLP plays an important role in the energy-saving control system of intelligent buildings. It has jointly promoted the realization of zero carbon emission goals by improving prediction accuracy, enhancing dynamic control flexibility, realizing refined management of energy consumption, extending equipment life, improving user comfort and promoting sustainable development.

[0095] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0096] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0097] Figure 1 It is a schematic diagram of the system of the present invention.

[0098] Figure 2 This is a schematic diagram of the power consumption ratio of each part.

[0099] Figure 3 This is an energy efficiency comparison chart after optimization of this system. DETAILED DESCRIPTION

[0100] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0101] like Figure 1 As shown, the present invention discloses an intelligent zero-carbon emission park building energy-saving control system, including the following modules:

[0102] The system architecture is divided into the following key modules:

[0103] Data processing module: including temperature sensors, light sensors, smart window sensors (transmittance adjustment), power consumption metering equipment, etc.

[0104] Deep learning module: A multi-layer neural network (MLP) combined with a convolutional neural network (CNN) is used to learn and train historical data to predict future indoor environmental requirements.

[0105] Intelligent control module: Based on the prediction results of the deep learning model, the light transmittance and thermal conductivity of the windows are adjusted in real time to adjust the indoor temperature and light in the best way; according to the real-time natural light intensity data, the brightness of the indoor lighting is adjusted through the deep learning algorithm to avoid unnecessary energy consumption.

[0106] Central control system: Through the intelligent control module, it connects with the park's energy management system to optimize overall energy utilization.

[0107] 1. Data processing module

[0108] Data collection: Deploy temperature sensors, light intensity sensors, and power metering sockets in campus buildings to continuously collect indoor temperature T(t), light intensity I(t), and energy consumption data P(t).

[0109] Data preprocessing: Clean and normalize the collected data to eliminate noise and outliers, and provide high-quality data sets for the training of deep learning models.

[0110] Use Z-score normalization

[0111]

[0112]

[0113]

[0114] Among them, Z T (t), Z I (t) and Z P(t) are the processed temperature, light intensity and energy consumption data, which are used to fill missing values ​​and replace outliers; the energy consumption data includes air conditioning power data, lighting power data, etc.

[0115] μ T , μ I and μ P are the average values ​​of temperature, light intensity and energy consumption data, σ T , σ I and σ P are the standard deviations of temperature, light intensity and energy consumption data respectively.

[0116] The processed indoor temperature T(t), light intensity I(t) and energy consumption data P(t) are combined into the environmental data vector x t :

[0117] x t =[T(t),I(t),P(t)];

[0118] x t is the environmental data vector at time t.

[0119] 2. Deep learning module,

[0120] First perform model training:

[0121] Use the combination of LSTM model and MLP model to train the data.

[0122] The LSTM model is used to predict the changes in light and temperature trends in the future, so as to help the system make adjustment decisions in advance. The MLP model is used to receive the prediction input of the LSTM model and the current system status, and generate control signals (transmittance, lighting intensity, air conditioning power) to optimize energy efficiency.

[0123] Data input:

[0124] X t =[x t-n ,x t-n+1 ,…,x t ];

[0125] Among them, X t The input time series data includes the environmental data vector x at the past n time points. t ;

[0126] x t-i is the environmental data vector at the tith moment

[0127] Forget Gate:

[0128] f t =σ(Wf (t)·[h t-1 ,x t ]+b f );

[0129] f t is the forget gate at time t, W f (t) is the forget gate weight matrix at time t, h t-1 is the hidden state at time t-1, x t is the environmental data vector at time t, b f is the bias of the forget gate, σ(*) is the activation function;

[0130] Among them, the forget gate weight matrix W f (t) is updated as follows:

[0131]

[0132] Among them, ω is the angular frequency, which controls the period of the periodic term, φ is the phase offset, which determines the initial value of the periodic term, and cos(ωt+φ) is the periodic term, which prevents the update process from continuing in a fixed direction, thereby preventing the model from falling into a local optimal solution;

[0133] W f (t+1) is the weight matrix of the forget gate at time t+1, W f (t) is the weight matrix of the forget gate at time t, W I (t) is the weight matrix of the input gate at time t, W C (t) is the weight matrix of the candidate memory at time t, and ε is a constant used to prevent the denominator from being zero;

[0134] Input Gate:

[0135] i t =σ(W i (t)·[h t-1 ,x t ]+b i );

[0136]

[0137] Among them, i t is the input gate at time t, is the candidate memory at time t, W i (t) is the weight matrix of the input gate at time t, W C (t) is the weight matrix of the candidate memory at time t;

[0138] b i is the bias of the input gate, b C is the bias of the candidate memory

[0139] tanh(*) is the hyperbolic tangent activation function;

[0140] Memory Update:

[0141]

[0142] C t is the cell state at time t, C t-1 is the cell state at time t-1;

[0143] Output Gate:

[0144] o t =σ(W o (t)·[h t-1 ,x t ]+b o );

[0145] h t =o t tanh(C t );

[0146] o t is the input gate at the current time t, W o (t) is the weight matrix of the output gate at time t, b o is the bias of the output gate, h t It is the output of the LSTM model, indicating the hidden state at the current moment;

[0147] The prediction output of the LSTM model:

[0148]

[0149] Represents the temperature and light prediction results at time t+k (the energy consumption results are included in the previous X). LSTM(*) is the prediction output of the LSTM model.

[0150] MLP Model: Control Decision Generation

[0151] The MLP model receives the prediction results of LSTM And the environmental data vector x at time t t and historical control output to generate control signals to control air conditioning power, smart window transmittance and lighting intensity;

[0152] enter:

[0153]

[0154] M t is the input vector of MLP;

[0155] is the predicted output of the LSTM model at the t+kth moment;

[0156] x t is the environmental data vector at time t;

[0157] u t-1 is the control signal at time t-1 (such as light transmittance, lighting intensity, air conditioning power);

[0158] MLP calculation formula:

[0159] h1=tanh(W1·M t +b1);

[0160] h2=tanh(W2·h1+b2);

[0161] u t =W3·h2+b3+γΔu t-1 ;

[0162] Where W1 and W2 are the weight matrices of the first hidden layer and the second hidden layer, respectively, and b1 and b2 are the bias vectors of the first hidden layer and the second hidden layer, respectively;

[0163] γ is an adaptive function used to adjust the influence of the change in the control signal, Δu t-1 is the change in the control signal at time t-1;

[0164] Among them, the calculation method of the adaptive function γ is:

[0165]

[0166] W3 is the weight matrix of the output layer, b3 is the bias vector of the output layer, and u t is the generated control signal vector, including window transmittance, lighting intensity, and air conditioning power.

[0167] 3. Intelligent control module:

[0168] Control signal vector u t The data structure is:

[0169] u t =[V t ,I t ,P t ];

[0170] V t Represents the voltage signal at the current time t, which is used to control the light transmittance B(V) of the smart window, I t represents the lighting control signal at the current time t, which is used to adjust the indoor lighting intensity, P tIndicates the air conditioning power command signal at the current time t, and adjusts the air conditioning power in percentage. The transmittance and lighting intensity are between 0 and 1, and the air conditioning power is between 0% and 100%;

[0171] Assume that the control signal vector output by the final calculation MLP model is u t =[0.7,0.5,60%], which means:

[0172] The window transmittance should be set to 70% (according to V t =0.7 converted).

[0173] Light intensity should be set to 50% (according to I t =0.5 conversion).

[0174] The air conditioning power should be set to 60%.

[0175] After training is completed, the model is integrated into the system for real-time prediction and control.

[0176] Window transmittance control:

[0177] According to the predicted light and temperature change trends, the control signal vector u generated by the deep learning model is t , automatically adjusts the light transmittance and thermal conductivity of the windows to make the indoor temperature and natural light reach comfortable standards, reducing dependence on air conditioning and artificial lighting.

[0178] like Figure 2 As shown, in summer, air conditioning accounts for 60% of the total energy consumption of buildings, while lighting energy consumption accounts for 15% of the total energy consumption, and other energy consumption accounts for 25% of the total energy consumption. By adjusting the window transmittance and intelligently controlling temperature and light, energy consumption can be effectively reduced and zero carbon emissions can be achieved.

[0179] Among them, the park uses electrochromic glass as the material for the smart windows to control the voltage V in the signal vector;

[0180] Working principle: This glass changes its transparency by applying voltage, which causes the material in the glass to undergo an electrochemical reaction, thereby changing the light transmittance. This change is reversible, and users can adjust the transparency manually or through intelligent programs as needed.

[0181] The relationship between the window transmittance B(V) and the applied voltage V is:

[0182]

[0183] Where, B(V) is the transmittance when voltage V is applied, B maxis the maximum transmittance, V max Indicates the maximum voltage that the material can withstand;

[0184] α is the coefficient of transmittance change, which depends on the material properties and the specific electrochromic reaction, and n is an exponential factor that describes the nonlinear degree of transmittance change with voltage;

[0185] Electrochromic glass is commonly used in building windows, skylights, car windshields, etc. It can darken when the sun is strong, reducing glare and the entry of solar radiation heat, thereby reducing the load on air conditioning.

[0186] Adjustable light transmittance: It can be adjusted from completely transparent to a darker state, and the adjustable light transmittance range is usually between 10% and 80%.

[0187] For example, when the sun is strong in the summer, the windows will automatically reduce their light transmittance, reducing the entry of solar radiation heat, thereby reducing the cooling load of the air conditioner.

[0188] 4. Central control system

[0189] Energy saving effect: By predicting environmental changes (such as temperature and light intensity) in real time and dynamically adjusting window transmittance, lighting and air conditioning power, the system can effectively reduce unnecessary energy consumption, especially in temperature regulation and artificial lighting.

[0190] Smart Scheduling: The system can adjust the equipment operation mode in advance based on the prediction results to avoid energy waste. For example, when the sun is bright, the system automatically reduces the lighting intensity and optimizes the window transmittance to reduce the air conditioning load, thereby achieving energy saving and improving user comfort.

[0191] like Figure 3 As shown in the figure, through experiments, this system has achieved the following optimization effects:

[0192] Lighting energy consumption: After optimization, lighting energy consumption was reduced by 54%, indicating that the intelligent lighting control strategy significantly improved energy efficiency.

[0193] Air conditioning energy consumption: After optimization, air conditioning energy consumption was reduced by 31%, reflecting the role of air conditioning equipment upgrades and intelligent temperature control systems in improving energy efficiency.

[0194] Carbon emissions: After optimization, carbon emissions were reduced by 28%, showing the effectiveness of comprehensive system optimization in reducing carbon emissions.

[0195] This experiment shows that by optimizing the lighting system, air conditioning system and energy management system, the overall energy efficiency of the building can be significantly improved, and energy consumption and carbon emissions can be reduced. The specific optimization measures effectively reduced the energy consumption of the lighting and air conditioning systems and reduced carbon emissions, verifying the important role of intelligent control systems in energy conservation and emission reduction. The comparison results of the experimental data clearly show the difference before and after optimization, supporting the effectiveness of the system optimization measures.

[0196] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. An intelligent zero-carbon emission park building energy-saving control system, characterized in that: include: Data processing module: including temperature sensors, light sensors, smart window sensors, and power consumption metering equipment; Deep learning module: Use multi-layer neural network combined with convolutional neural network to learn and train historical data to predict future indoor environment requirements; Intelligent control module: Based on the prediction results of the deep learning model, it adjusts the light transmittance and thermal conductivity of the window in real time; Central control system: Through the intelligent control module, it connects with the park's energy management system to optimize overall energy utilization.

2. According to claim 1, an intelligent zero-carbon emission park building energy-saving control system is characterized in that: The data processing module comprises: Data collection: Deploy temperature sensors, light intensity sensors, and power metering sockets to continuously collect indoor temperature T(t), light intensity I(t), and energy consumption data P(t); Data preprocessing: Clean and normalize the collected data to eliminate noise and outliers, and provide high-quality data sets for deep learning model training, including: Use Z-score normalization: Among them, Z T (t), Z I (t) and Z P (t) are the processed temperature, light intensity and energy consumption data, which are used to fill missing values ​​and replace abnormal values; μ T , μ I and μ P are the average values ​​of temperature, light intensity and energy consumption data, σ T , σ I and σ P are the standard deviations of temperature, light intensity, and energy consumption data, respectively; The processed indoor temperature T(t), light intensity I(t) and energy consumption data P(t) are combined into the environmental data vector x t : x t is the environmental data vector at time t.

3. According to claim 1, an intelligent zero-carbon emission park building energy-saving control system is characterized in that: The deep learning module includes: Model training: Use the combination of LSTM model and MLP model to train the data. The LSTM model is used to predict the changes in light and temperature trends in the future, so as to help the system make adjustment decisions in advance. The MLP model is used to receive the prediction input of the LSTM model and the current system status, and generate control signals to optimize energy efficiency. Data input: X t =[x t-n ,x t-n+1 ,…,x t ]; Among them, X t The input time series data includes the environmental data vector x at the past n time points. t ; x t-i is the environmental data vector at the ti-th moment, where the value range of the integer i is 0≤i≤n; Forget Gate: f t =σ(W f (t)·[h t-1 ,x t ]+b f ); f t is the forget gate at time t, W f (t) is the forget gate weight matrix at time t, h t-1 is the hidden state at time t-1, x t is the environmental data vector at time t, b f is the bias of the forget gate, σ(*) is the activation function; Among them, the forget gate weight matrix W f (t) is updated as follows: Among them, ω is the angular frequency, which controls the period of the periodic term, φ is the phase offset, which determines the initial value of the periodic term, and cos(ωt+φ) is the periodic term, which prevents the update process from continuing in a fixed direction, thereby preventing the model from falling into a local optimal solution; W f (t+1) is the weight matrix of the forget gate at time t+1, W f (t) is the weight matrix of the forget gate at time t, W i (t) is the weight matrix of the input gate at time t, W C (t) is the weight matrix of the candidate memory at time t, and ε is a constant used to prevent the denominator from being zero; Input Gate: i t =σ(W i (t)·[h t-1 ,x t ]+b i ); Among them, i t is the input gate at time t, is the candidate memory at time t, W i (t) is the weight matrix of the input gate at time t, W C (t) is the weight matrix of the candidate memory at time t; b i is the bias of the input gate, b C is the bias of the candidate memory; tanh(*) is the hyperbolic tangent activation function; Memory Update: C t is the cell state at time t, C t-1 is the cell state at time t-1; Output Gate: o t =σ(W o (t)·[h t-1 ,x t ]+b o ); h t =o t ·tanh(C t ); o t is the input gate at the current time t, W o (t) is the weight matrix of the output gate at time t, b o is the bias of the output gate, h t It is the output of the LSTM model, indicating the hidden state at the current moment; The prediction output of the LSTM model: It represents the temperature and light prediction results at the t+kth moment, and LSTM(*) is the prediction output of the LSTM model.

4. According to claim 1, an intelligent zero-carbon emission park building energy-saving control system is characterized in that: The deep learning module includes: The MLP model receives the prediction results of LSTM And the environmental data vector x at time t t and historical control output to generate control signals to control air conditioning power, smart window transmittance and lighting intensity; enter: M t is the input vector of MLP; is the predicted output of the LSTM model at the t+kth moment; x t is the environmental data vector at time t; u t-1 is the control signal at time t-1; MLP calculation formula: h1=tanh(W1·M t +b1); h2=tanh(W2·h1+b2); u t =W3·h2+b3+γΔu t-1 ; Where h1 and h2 represent the activation outputs of the first hidden layer and the second hidden layer, respectively; W1 and W2 are the weight matrices of the first hidden layer and the second hidden layer, respectively; b1 and b2 are the bias vectors of the first hidden layer and the second hidden layer, respectively; γ is an adaptive function used to adjust the influence of the change in the control signal, Δu t-1 is the change in the control signal at time t-1; Among them, the calculation method of the adaptive function γ is: W3 is the weight matrix of the output layer, b3 is the bias vector of the output layer, and u t is the generated control signal vector, including window transmittance, lighting intensity, and air conditioning power.

5. According to claim 1, an intelligent zero-carbon emission park building energy-saving control system is characterized in that: The intelligent control module comprises: Control signal vector u t The data structure is: u t =[V t ,I t ,P t ]; V t Represents the voltage signal at the current time t, which is used to control the light transmittance B(V) of the smart window, I t represents the lighting control signal at the current time t, which is used to adjust the indoor lighting intensity, P t represents the air conditioning power command signal at the current time t, and adjusts the air conditioning power in percentage; The relationship between the window transmittance B(V) and the applied voltage V is: Where, B(V) is the transmittance when voltage V is applied, B max is the maximum transmittance, V max Indicates the maximum voltage that the material can withstand. By adjusting the transmittance B(V), the air conditioning power P is adjusted. t and lighting control signal I t , saving power consumption without affecting actual operation; α is the coefficient of transmittance change, which depends on the characteristics of the material itself and the specific electrochromic reaction. C is an exponential factor that describes the nonlinear degree of transmittance change with voltage.

6. An intelligent zero-carbon emission park building energy-saving control method, characterized in that: The following steps are involved: S1, collects data from temperature sensors, light sensors, smart window sensors, and power consumption metering devices, and processes the data; S2, uses a multi-layer neural network combined with a convolutional neural network to learn and train historical data to predict future indoor environmental requirements; S3, based on the prediction results of the deep learning model, adjusts the light transmittance and thermal conductivity of the window in real time; S4, through the intelligent control module, connects with the park’s energy management system to optimize overall energy utilization.

7. A computer system, characterized in that: include: processor; a memory for storing processor-executable instructions; Wherein, the processor is configured to implement an intelligent zero-carbon emission campus building energy-saving control system as described in any one of claims 1 to 5 when executing the executable instructions.

8. A computer-readable storage medium, characterized in that: include: a memory having a computer program stored thereon; A processor is used to execute the program in the memory to implement an intelligent zero-carbon emission campus building energy-saving control system as described in any one of claims 1 to 5.

Images