Zero-carbon park energy efficiency optimization method and system based on machine learning, and medium

Abstract

The invention provides a zero-carbon park energy efficiency optimization system based on machine learning, and aims to realize a zero-carbon emission target of a park through a multi-level energy efficiency optimization strategy. The system includes short-term, medium-term, and long-term load prediction models for real-time load prediction, periodic load prediction, and carbon emission management, respectively. The short-term model is based on an ARIMA model, can carry out real-time load prediction, and quickly responds to the fluctuation of the load in the park. The medium-term model adopts a Prophet model to identify periodic demand changes and provide support for load balance and energy storage management; a Bayesian structure time series model (BSTS) is utilized by the long-term model to realize carbon emission management and energy structure optimization. According to the system, historical and real-time data are analyzed through a machine learning algorithm, an energy consumption scheme of a park is dynamically adjusted, and an optimal energy scheduling strategy is generated in combination with short-term and long-term carbon emission targets.

Description

Technical Field

[0001] The present invention relates to the technical field of system optimization, and in particular to a method, system and medium for optimizing energy efficiency of a zero-carbon park based on machine learning. Background Art

[0002] With the intensification of global climate change and the continuous increase in energy consumption, zero carbon emissions have become an important development goal worldwide. As one of the energy-intensive areas, industrial parks often include multiple energy-consuming units such as office buildings, industrial facilities, and residential buildings. Their energy needs are complex, diverse, and volatile. Therefore, achieving zero carbon emissions in industrial parks has extremely high economic and social value.

[0003] In the process of building a zero-carbon park, energy efficiency optimization is the key to achieving the zero-carbon goal. Current energy efficiency optimization systems usually rely on traditional energy management methods, such as load forecasting, energy consumption monitoring, and basic demand response strategies. However, most of these methods are based on preset rules and fixed models, which are difficult to cope with complex load changes and the dynamic impact of the external environment, and cannot achieve the ideal energy efficiency optimization effect. Summary of the invention

[0004] The present invention aims to at least solve the technical problems existing in the prior art, and in particular innovatively proposes a zero-carbon park energy efficiency optimization system based on machine learning, comprising:

[0005] Data acquisition module: collects data from sensors and stores the data in the form of time series;

[0006] Short-term model module: Real-time load forecasting using ARIMA model;

[0007] Medium-term model module: Use prophet model for periodic load forecasting;

[0008] Long-term model module: constructing Bayesian structural time series models for carbon emission management;

[0009] Hybrid scheduling module: combines the prediction components of various models to achieve comprehensive optimization in the short, medium and long term.

[0010] In a preferred embodiment of the present invention, the data acquisition module includes:

[0011] Collect data from sensors, including load data L(t), environmental data E(t), electricity price data V(t), and equipment status data S(t):

[0012] Data(t)={L(t),E(t),V(t),S(t)};

[0013] Where Data(t) is the combined data vector;

[0014] After collecting the data, store it in time series form and concatenate it into a long matrix:

[0015]

[0016] Where Data is a data matrix, which is composed of combined data vectors, t1, t2, ..., t T is the sampling timestamp.

[0017] In a preferred embodiment of the present invention, the short-term model module comprises:

[0018] The data matrix Data is input into the ARIMA model for real-time load forecasting. The model can be expressed as:

[0019]

[0020] c(t) is the real-time load value at time t, f ARIMA (*) is the ARIMA model, Data is the data matrix;

[0021] is the autoregressive part, indicating the past p t time steps, φ i is the autoregressive coefficient, which indicates the influence of the load value at time step ti on time t, and L(ti) indicates the load value at the i-th time step in the past;

[0022] is the moving average part, which indicates the past q t The influence of the error term of time steps on the current load, θ j is the moving average coefficient, which represents the weight of the error at time step j on time step t, ρ(tj) is the prediction error of the jth time step in the past, that is, the difference between the actual load value and the predicted load value, and μ(t) is the random error term at time t.

[0023] In a preferred embodiment of the present invention, the medium-term model module comprises:

[0024] Using Fourier series to capture the seasonality of load data, the seasonal component can be expressed as:

[0025]

[0026] s month (t) is the monthly seasonal component at time t, P is the seasonal period, N is the order of the Fourier series, and a n and b n are the Fourier coefficients;

[0027] Calculate holiday effects:

[0028]

[0029] h(t) is the holiday effect at time t, I u is the impact intensity of the u-th holiday, U is the total number of holidays, indicator(t∈holiday u ) is an indicator function, which is 1 when t is the uth holiday and 0 otherwise.

[0030] In a preferred embodiment of the present invention, the long-term model module comprises:

[0031] By introducing nonlinear functions

[0032] g(t)=r0+r1t+r2t 2 +...+r m t m ;

[0033] Among them, g(t) is the trend component at time t, which is used to represent the long-term trend, r0, r1, r2...r m is the trend parameter, m represents the order of the polynomial, and such a model can adapt to nonlinear trend changes;

[0034] Combine the long-term model and the medium-term model of seasonal components, that is, the annual seasonal component s in the long-term model year (t) and the quarterly seasonal component s quarter (t) and the monthly seasonal component s in the medium-term model month (t) Combined:

[0035] s(t)=s year (t)+s quarter (t)+s month (t);

[0036] Among them, s(t) is the combined periodic component, s year (t) is the annual seasonal component at time t, s quarter (t) is the quarterly seasonal component at time t, s month (t) is the monthly seasonal component at time t;

[0037] Introducing dynamic regressors:

[0038]

[0039] X(t) is the regression component at time t, Z k (t) is the kth external regressor, K is the total number of external regressors, δk (t) is the dynamic weight of the external regressor, which determines the impact strength of different external review variables;

[0040] in,

[0041] δ k (t+1) is the external regressor Z at time t+1 k (t), ψ is the learning rate of the dynamic weight of the external regressor, which is used to control the step size of the weight update, p(X(t)|δ k ) is the likelihood function of the regression component X(t) at time t, indicating that given the weight δ k The probability of observing X(t) under the condition of

[0042] is the gradient of the log-likelihood function, which is used to indicate the sensitivity of the current weight to the prediction error.

[0043] In a preferred embodiment of the present invention, the hybrid scheduling module includes:

[0044] Combine the prediction components of the individual models:

[0045] y(t)=α t g(t)+β t ·s(t)+γ t c(t)+X(t)+h(t)+∈ t ;

[0046] Among them, y(t) is the hybrid scheduling framework at time t, which is obtained by weighted combination of multiple prediction components, g(t) is the trend component at time t, indicating the long-term trend, s(t) is the combined periodic component, indicating periodic fluctuations, c(t) is the real-time load value at time t, indicating short-term prediction, X(t) is the regression component at time t, indicating the impact related to other variables, h(t) is the holiday effect at time t, ∈ t is the error term, α t , β t and γ t is the adaptive weight;

[0047] The weights are adjusted by regularizing Bayesian updates to adapt to dynamic changes in data:

[0048]

[0049] Among them, α t+1 , β t+1 and γ t+1 represents the updated weight value at time t+1, λα t , λβt and λγ t It is a regularization term that prevents the weight value from being too large or too small and is used to maintain the stability of the update;

[0050] as well as is the gradient of the likelihood function, which is used to express the sensitivity of the current weight to the prediction error;

[0051] Log-likelihood logp(y(t)|α,β,γ) is the optimization objective function, which is used to maximize the explanatory power of y(t) under the model parameters α,β,γ, thereby improving the model's fit to the data;

[0052] α, β, γ are the corresponding weights α t ,β t ,γ t The base weight of , η is the learning rate of the regularized Bayesian update, and the weight is adjusted by the Bayesian posterior probability to enable it to adapt to changes.

[0053] The present invention also discloses a computer system, comprising:

[0054] a memory for storing processor-executable instructions;

[0055] Among them, the processor is configured to implement the zero-carbon park energy efficiency optimization system based on machine learning when executing the executable instructions.

[0056] The present invention also discloses a computer-readable medium, comprising:

[0057] a memory having a computer program stored thereon;

[0058] A processor is used to execute the program in the memory to implement the zero-carbon park energy efficiency optimization system based on machine learning.

[0059] The present invention also discloses a zero-carbon park energy efficiency optimization method based on machine learning, which is characterized by comprising:

[0060] S1, collects data from sensors and stores the data in the form of time series;

[0061] S2, real-time load forecasting using ARIMA model;

[0062] S3, periodic load forecasting using prophet model;

[0063] S4, constructing a Bayesian structural time series model for carbon emission management;

[0064] S5, combines the prediction components of each model to achieve comprehensive optimization in the short, medium and long term.

[0065] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0066] The data acquisition module collects multi-dimensional data such as load, environment, and electricity price from sensors and stores it in the form of time series, providing complete and reliable basic data for subsequent load forecasting and energy efficiency optimization. This real-time data collection method, combined with the organizational structure of the time series, ensures the continuity and integrity of the data, thereby improving the prediction accuracy of the model.

[0067] The short-term model module performs real-time load forecasting based on the ARIMA model, which can quickly respond to load fluctuations in the park. The ARIMA model uses autoregression and moving average methods, combined with short-term load historical data, to accurately predict real-time loads, which is particularly suitable for handling rapid changes in daily energy consumption in the park. This function greatly improves the system's scheduling flexibility in a short period of time.

[0068] The medium-term model module uses the Prophet model for periodic load forecasting, effectively capturing seasonal, periodic patterns, and energy consumption characteristics at special time points such as holidays. Through medium-term forecasting, the system can formulate targeted energy scheduling strategies in advance, such as optimizing the charging and discharging plans of energy storage equipment, providing more forward-looking support for energy efficiency management.

[0069] The long-term model module realizes the long-term management and trend analysis of carbon emissions by constructing a Bayesian structural time series model (BSTS). This module can capture the long-term trend of the park's energy structure and carbon emissions, formulate strategic recommendations for optimizing the energy structure, support the realization of zero-carbon goals, and thus meet the sustainable development needs of the park.

[0070] The hybrid scheduling module achieves comprehensive optimization on a time scale by organically combining the prediction components of short-term, medium-term and long-term models. By integrating the real-time performance of the short-term model, the periodic prediction of the medium-term model and the trend analysis of the long-term model, the system can provide accurate energy efficiency optimization solutions in different time dimensions. This multi-model collaborative scheduling mechanism enables the system to quickly respond to sudden load demands and optimize the overall energy utilization efficiency according to medium- and long-term goals, finding a balance between economic efficiency and carbon emission targets.

[0071] 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

[0072] 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:

[0073] Figure 1 It is a system schematic diagram of the present invention. DETAILED DESCRIPTION

[0074] 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.

[0075] like Figure 1 As shown, the present invention provides a method, system and medium for optimizing energy efficiency of a zero-carbon park based on machine learning, including:

[0076] 1. Data Acquisition

[0077] Collect data from sensors, including load data L(t), environmental data E(t), electricity price data V(t), and equipment status data S(t):

[0078] Data(t)={L(t),E(t),V(t),S(t)};

[0079] Where Data(t) is the combined data vector;

[0080] After collecting the data, store it in time series form and concatenate it into a long matrix:

[0081]

[0082] Where Data is a data matrix, which is composed of combined data vectors, t1, t2, ..., t T is the sampling timestamp.

[0083] 2. Short-term model

[0084] The data matrix Data is input into the ARIMA model for real-time load forecasting. The model can be expressed as:

[0085]

[0086] c(t) is the real-time load value at time t, f ARIMA (*) is the ARIMA model, Data is the data matrix;

[0087] is the autoregressive part, indicating the past p t time steps, φ i is the autoregressive coefficient, which indicates the influence of the load value at time step ti on time t, and L(ti) indicates the load value at the i-th time step in the past;

[0088] is the moving average part, which indicates the past q t The influence of the error term of time steps on the current load, θ j is the moving average coefficient, which represents the weight of the error at time step j on time step t, ρ(tj) is the prediction error of the jth time step in the past, that is, the difference between the actual load value and the predicted load value, and μ(t) is the random error term at time t.

[0089] 3 mid-term model

[0090] Use prophet model for periodic load forecasting.

[0091] 3.1 Using Fourier series to capture the seasonality of load data, the seasonal component can be expressed as:

[0092]

[0093] s month (t) is the seasonal component at time t, P is the seasonal period, N is the order of the Fourier series, and a n and b n are the Fourier coefficients.

[0094] 3.2 Holiday Effect

[0095] The holiday effect is introduced as one of the correction factors of the mid-term model, and the special energy consumption impact caused by holidays is expressed by the following formula:

[0096]

[0097] h(t) is the holiday effect at time t, I u is the impact intensity of the u-th holiday, U is the total number of holidays, indicator(t∈holiday u ) is an indicator function, which is 1 when t is the uth holiday and 0 otherwise.

[0098] Through this modeling approach, the system can incorporate special load demands during holidays into forecasts, capture abnormal load fluctuation characteristics during holidays, and provide more accurate support for energy efficiency optimization and energy scheduling.

[0099] 4 Long-term model

[0100] Constructing a Bayesian structural time series model for carbon emissions management.

[0101] 4.1 Nonlinear trend modeling:

[0102] By introducing nonlinear functions to capture more complex long-term trends, the model's ability to capture nonlinear trends can be further improved.

[0103] g(t)=r0+r1t+r2t 2 +...+r m t m ;

[0104] Among them, g(t) is the trend component at time t, which is used to represent the long-term trend, r0, r1, r2...r m is the trend parameter, m represents the order of the polynomial, and such a model can adapt to nonlinear trend changes.

[0105] 4.2 Combining medium-term and long-term seasonal components:

[0106] Combine the long-term model and the medium-term model of seasonal components, that is, the annual seasonal component s in the long-term model year (t) and the quarterly seasonal component s quarter (t) and the monthly seasonal component s in the medium-term model month (t) Combined:

[0107] s(t)=s year (t)+s quarter (t)+s month (t);

[0108] Among them, s(t) is the combined periodic component, s year (t) is the annual seasonal component at time t, s quarter (t) is the quarterly seasonal component at time t, s month (t) is the monthly seasonal component at time t.

[0109] The seasonal components of the long-term model and the medium-term model are combined, and the characteristics of the annual seasonal component, the quarterly seasonal component, and the monthly seasonal component are extracted respectively, and the periodic laws on different time scales are integrated into a unified model to construct a comprehensive periodic component. This combination method fully utilizes the advantages of the long-term model in capturing macro trends and the accuracy of the medium-term model in identifying shorter periodic laws. Through the collaboration between models, the complex seasonal characteristics in time can be more comprehensively characterized, so that the model can not only make strategic plans for long-term trends, but also respond to short-term demand fluctuations in the medium term, thereby providing more timely and robust forecasting support for optimized scheduling.

[0110] 4.3 Dynamic regression components

[0111] Introduce dynamic regression variables that can be continuously updated over time. Consider variables such as temperature and economic indicators that may have an impact on load demand in long-term forecasts. Dynamic regression components can update weights as these external factors change.

[0112]

[0113] X(t) is the regression component at time t, Z k (t) is the kth external regressor, K is the total number of external regressors, δ k (t) is the dynamic weight of the external regressor, which determines the impact strength of different external review variables;

[0114] Such dynamic regression components enable the model to flexibly adapt to changes in the external environment.

[0115] in,

[0116] δ k (t+1) is the external regressor Z at time t+1 k (t), ψ is the learning rate of the dynamic weight of the external regressor, which is used to control the step size of the weight update, p(X(t)|δ k ) is the likelihood function of the regression component X(t) at time t, indicating that given the weight δ k The probability of observing X(t) under the condition of

[0117] It is the gradient of the log-likelihood function, which is used to indicate the sensitivity of the current weight to the prediction error.

[0118] 5. Hybrid Scheduling

[0119] Short-term models are used for real-time load forecasting, with an emphasis on fast response and real-time adjustments.

[0120] The medium-term model is used for periodic load forecasting and focuses on capturing the medium-term periodic characteristics.

[0121] The long-term model focuses on long-term trend forecasting and carbon emission management, supporting long-term optimization and strategic planning of the system.

[0122] 5.1 Combining the prediction components of individual models

[0123] y(t)=α t g(t)+β t ·s(t)+γ t c(t)+X(t)+h(t)+∈ t ;

[0124] Among them, y(t) is the hybrid scheduling framework at time t, which is obtained by weighted combination of multiple prediction components, g(t) is the trend component at time t, indicating the long-term trend, s(t) is the combined periodic component, indicating periodic fluctuations, c(t) is the real-time load value at time t, indicating short-term prediction, X(t) is the regression component at time t, indicating the impact related to other variables, h(t) is the holiday effect at time t, ∈ t is the error term, α t , β t and γ t is the adaptive weight.

[0125] The weights are adjusted through Bayesian updating to adapt to the dynamic changes of data:

[0126]

[0127] Among them, α t+1 , β t+1 and γ t+1 represents the updated weight value at time t+1, λα t , λβ t and λγ t It is a regularization term that prevents the weight value from being too large or too small and is used to maintain the stability of the update;

[0128] as well as is the gradient of the likelihood function, which is used to express the sensitivity of the current weight to the prediction error;

[0129] Log-likelihood logp(y(t)|α,β,γ) is the optimization objective function, which is used to maximize the explanatory power of y(t) under the model parameters α,β,γ, thereby improving the model's fit to the data;

[0130] α, β, γ are the corresponding weights α t ,β t ,γ t The base weight of , η is the learning rate of the regularized Bayesian update, and the weight is adjusted by the Bayesian posterior probability to enable it to adapt to changes.

[0131] The focus of hybrid scheduling is to combine the advantages of various models to create a forecasting model that can address different time requirements;

[0132] Its short-term model focuses on real-time load forecasting, emphasizing rapid response and real-time adjustment capabilities. By analyzing real-time data and using the ARIMA model to quickly capture short-term fluctuation trends in load, it provides accurate real-time support for load scheduling. This level of model ensures the system's flexible adaptability and rapid response to short-term demand changes.

[0133] The medium-term model aims to identify the cyclical characteristics of the load, focusing on load forecasting and energy planning in the medium term. The Prophet model is used to capture the seasonal patterns and cyclical fluctuations of the load, providing a reference for energy reserves and scheduling. Through medium-term forecasting, the system can reasonably adjust seasonal load demand and improve energy efficiency.

[0134] The long-term model focuses on capturing the long-term trend of the load, while managing carbon emissions and optimizing the energy structure. The Bayesian Structural Time Series Model (BSTS) is used to model the long-term trend of the load and formulate a long-term optimization planning strategy in combination with carbon emission data. By analyzing the correlation between carbon emissions and energy efficiency, the model supports the strategic energy management of the system and helps achieve the long-term zero-carbon goal of the park.

[0135] 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. A zero-carbon park energy efficiency optimization system based on machine learning, characterized in that: include: Data acquisition module: collects data from sensors and stores the data in the form of time series; Short-term model module: Real-time load forecasting using ARIMA model; Medium-term model module: Use prophet model for periodic load forecasting; Long-term model module: constructing Bayesian structural time series models for carbon emission management; Hybrid scheduling module: combines the prediction components of various models to achieve comprehensive optimization in the short, medium and long term.

2. According to the machine learning-based zero-carbon park energy efficiency optimization system of claim 1, it is characterized in that: The data acquisition module comprises: Collect data from sensors, including load data L(t), environmental data E(t), electricity price data V(t), and equipment status data S(t): Data(t)={L(t),E(t),V(t),S(t)}; Where Data(t) is the combined data vector; After collecting the data, store it in time series form and concatenate it into a long matrix: Where Data is a data matrix, which is composed of combined data vectors, t1, t2, ..., t T is the sampling timestamp.

3. According to the machine learning-based zero-carbon park energy efficiency optimization system of claim 1, it is characterized in that: The short-term model module includes: The data matrix Data is input into the ARIMA model for real-time load forecasting. The model can be expressed as: c(t) is the real-time load value at time t, f ARIMA (*) is the ARIMA model, Data is the data matrix; is the autoregressive part, indicating the past p t time steps, φ i is the autoregressive coefficient, which indicates the influence of the load value at time step ti on time t, and L(ti) indicates the load value at the i-th time step in the past; is the moving average part, which indicates the past q t The influence of the error term of time steps on the current load, θ j is the moving average coefficient, which represents the weight of the error at time step j on time step t, ρ(tj) is the prediction error of the jth time step in the past, that is, the difference between the actual load value and the predicted load value, and μ(t) is the random error term at time t.

4. According to the machine learning-based zero-carbon park energy efficiency optimization system of claim 1, it is characterized in that: The mid-term model module includes: Using Fourier series to capture the seasonality of load data, the seasonal component can be expressed as: s month (t) is the monthly seasonal component at time t, P is the seasonal period, N is the order of the Fourier series, and a n and b n are the Fourier coefficients; Calculate holiday effects: h(t) is the holiday effect at time t, I u is the impact intensity of the u-th holiday, U is the total number of holidays, indicator(t∈holiday u ) is an indicator function, which is 1 when t is the uth holiday and 0 otherwise.

5. According to the machine learning-based zero-carbon park energy efficiency optimization system of claim 1, it is characterized in that: The long-term model module includes: By introducing nonlinear functions g(t)=r0+r1t+r2t 2 +…+r m t m ; Among them, g(t) is the trend component at time t, which is used to represent the long-term trend, r0, r1, r2…r m is the trend parameter, m represents the order of the polynomial, and such a model can adapt to nonlinear trend changes; Combine the long-term model and the medium-term model of seasonal components, that is, the annual seasonal component s in the long-term model year (t) and the quarterly seasonal component s quarter (t) and the monthly seasonal component s in the medium-term model month (t) Combined: s(t)=s year (t)+s quarter (t)+s month (t); Among them, s(t) is the combined periodic component, s year (t) is the annual seasonal component at time t, s quarter (t) is the quarterly seasonal component at time t, s month (t) is the monthly seasonal component at time t; Introducing dynamic regressors: X(t) is the regression component at time t, Z k (t) is the kth external regressor, K is the total number of external regressors, δ k (t) is the dynamic weight of the external regressor, which determines the impact strength of different external review variables; in, δ k (t+1) is the external regressor Z at time t+1 k (t), ψ is the learning rate of the dynamic weight of the external regressor, which is used to control the step size of the weight update, p(X(t)|δ k ) is the likelihood function of the regression component X(t) at time t, indicating that given the weight δ k The probability of observing X(t) under the condition of is the gradient of the log-likelihood function, which is used to indicate the sensitivity of the current weight to the prediction error.

6. The zero-carbon park energy efficiency optimization system based on machine learning according to claim 1 is characterized in that: The hybrid scheduling module includes: Combine the prediction components of the individual models: y(t)=α t ·g(t)+β t ·s(t)+γ t ·c(t)+X(t)+h(t)+∈ t ; Among them, y(t) is the hybrid scheduling framework at time t, which is obtained by weighted combination of multiple prediction components, g(t) is the trend component at time t, indicating the long-term trend, s(t) is the combined periodic component, indicating periodic fluctuations, c(t) is the real-time load value at time t, indicating short-term prediction, X(t) is the regression component at time t, indicating the impact related to other variables, h(t) is the holiday effect at time t, ∈ t is the error term, α t , β t and γ t is the adaptive weight; The weights are adjusted by regularizing Bayesian updates to adapt to dynamic changes in data: Among them, α t+1 , β t+1 and γ t+1 represents the updated weight value at time t+1, λα t , λβ t and λγ t It is a regularization term that prevents the weight value from being too large or too small and is used to maintain the stability of the update; as well as is the gradient of the likelihood function, which is used to express the sensitivity of the current weight to the prediction error; Log-likelihood logp(y(t)|α,β,γ) is the optimization objective function, which is used to maximize the explanatory power of y(t) under the model parameters α,β,γ, thereby improving the model's fit to the data; α, β, γ are the corresponding weights α t ,β t ,γ t The base weight of , η is the learning rate of the regularized Bayesian update, and the weight is adjusted by the Bayesian posterior probability to enable it to adapt to changes.

7. A computer system, characterized in that: include: a memory for storing processor-executable instructions; Wherein, the processor is configured to implement a zero-carbon park energy efficiency optimization system based on machine learning as described in any one of claims 1 to 6 when executing the executable instructions.

8. A computer-readable medium, characterized in that Said include: a memory having a computer program stored thereon; A processor, used to execute the program in the memory to implement a zero-carbon park energy efficiency optimization system based on machine learning as described in any one of claims 1 to 5.

9. A zero-carbon park energy efficiency optimization method based on machine learning, characterized in that: include: S1, collects data from sensors and stores the data in the form of time series; S2, real-time load forecasting using ARIMA model; S3, periodic load forecasting using prophet model; S4, constructing a Bayesian structural time series model for carbon emission management; S5, combines the prediction components of each model to achieve comprehensive optimization in the short, medium and long term.

Images