Glass factory load green electricity intelligent regulation and control method based on green certificate

Abstract

The invention discloses a glass factory load green electricity intelligent regulation and control method based on green certificates, and the method comprises the steps: calculating the available number of the green certificates through constructing a multi-scale cross convolution network model; and constructing a regulation and control optimization model, comprehensively considering load cost, power cost, certificate revenue and performance penalty, realizing full-time-sequence dynamic optimization of load regulation and energy storage operation, and outputting a future control strategy on the premise of meeting equipment operation and green certificate constraints. And finally, proposing a load real-time control and green performance closed-loop strategy, executing an optimization strategy in a production system, carrying out rolling feedback, carrying out dynamic comparison on the actual green electricity absorption amount and the target performance progress, and if a deviation is found, triggering a re-optimization mechanism or green certificate supplementary purchase so as to guarantee the performance compliance. According to the invention, intelligentization, economization and compliance of green electricity regulation and control of the glass factory are realized, improvement of the green manufacturing level is promoted, and the active adaptability of enterprises in the carbon market and the green certificate market is enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of power system regulation and control, and in particular to a green certificate-based intelligent regulation method for green electricity load in a glass factory. Background Art

[0002] Green electricity regulation of glass factory loads can effectively improve the matching utilization rate of green electricity in high-energy-consuming processes (such as melting furnaces, annealing sections, etc.), reduce the carbon emission intensity per unit product, and promote the green manufacturing transformation of enterprises. By regulating the load in response to green electricity fluctuations, the efficient absorption of renewable energy can be achieved, while providing a regulatory basis and economic benefit space for enterprises to participate in carbon quota management and green electricity trading. The existing green electricity regulation methods for glass factory loads have many shortcomings in terms of green electricity utilization accuracy, response mechanism flexibility, and green compliance closed loop, making it difficult to meet the current "dual carbon" goals and green manufacturing needs, such as:

[0003] 1) Inadequate accuracy in identifying green electricity availability. Existing methods often rely on simple time-slot divisions or average PV / wind power forecasts. These methods fail to consider the glass factory's energy consumption structure, the volatility of distributed energy resources, and the spatiotemporal coupling characteristics. This makes it difficult to accurately determine whether matching green electricity is available for a specific load, leading to blind spots and waste in green electricity utilization.

[0004] 2) Green electricity regulation optimization lacks multi-objective integration. Most regulation strategies simply aim for load to follow green electricity fluctuations, failing to fully consider the linkage constraints of the glass factory's complex process flow (such as melting furnace continuity and annealing section temperature control rigidity), as well as factors such as green certificate acquisition and compliance costs. This makes regulation strategies difficult to implement and lacks profitability and feasibility.

[0005] 3) Lack of a closed-loop green compliance mechanism. Current methods typically separate load regulation from green certificates, making it impossible to achieve real-time monitoring, dynamic adjustment, and verification of green electricity utilization. This results in a failure to form a closed loop between regulation and carbon asset management and policy compliance, resulting in low green compliance efficiency and high costs.

[0006] To this end, we designed a green certificate-based intelligent control method for glass factory load green electricity to solve the above problems. Summary of the Invention

[0007] The purpose of this invention is to solve the shortcomings of the existing technology in the process of regulating the green electricity access of glass factory loads, such as insufficient accuracy in identifying green electricity availability, lack of multi-objective fusion in regulation optimization, and lack of a closed-loop green compliance mechanism. A method for intelligent green electricity regulation of glass factory loads based on green certificates is proposed to improve the precise response of glass factory loads to green electricity and their green compliance capabilities. The multi-scale cross-convolutional network is used to identify the green electricity availability and certificate acquisition potential. A regulation optimization model that takes into account both process constraints and certificate benefits is constructed by combining spatiotemporal optimization with differential equation constraints. A real-time load control and closed-loop compliance strategy is further proposed to achieve efficient coordination between green electricity regulation and green certificate certification, thus facilitating the green transformation and carbon asset management of glass factories.

[0008] In order to achieve the above object, the present invention adopts the following technical solutions:

[0009] A method for intelligently controlling green electricity load in a glass factory based on green certificates comprises the following steps:

[0010] Step 1: Based on a multi-scale cross-convolutional network, a joint multi-scale cross-convolutional network model is constructed that integrates the glass factory's load, power supply, energy storage, and meteorological information to predict the green electricity ratio, available green electricity power, and the amount of green certificates obtained.

[0011] Step 2: Build a regulation optimization model based on combining spatiotemporal optimization with differential equation constraints and green certificates. This model comprehensively considers load regulation costs, electricity costs, certificate revenue, and compliance penalties to dynamically optimize the full time sequence of load regulation and energy storage operations.

[0012] Step three: Execute the optimization strategy and conduct rolling feedback, dynamically comparing the actual green electricity absorption with the target compliance progress. If any deviation is found, re-optimization or green certificate purchase mechanism will be triggered.

[0013] Furthermore, before constructing the multi-scale cross-convolutional network joint model, the historical time series of the glass factory load, power supply, energy storage, and meteorological information data are integrated, the historical window length is selected, and the input historical time series vector is constructed. The representation of the historical time series vector is as follows:

[0014]

[0015] Where, X t is the input tensor at time t, L is the length of the historical input window, P load (t-L+1:t) is the total load power of the glass factory, P res (t-L+1:t) is the output of distributed renewable energy, is the discharge power of the energy storage system, SOC(t-L+1:t) is the state of charge of the energy storage, T(t-L+1:t) is the ambient temperature, and W(t-L+1:t) is the wind speed or light intensity;

[0016] The expression of the feature compression layer using multi-scale convolution kernels for parallel modeling is as follows:

[0017]

[0018] Where, LayerNorm(·) represents the multi-scale convolution kernel function, W proj represents the multi-scale convolution kernel coefficient, The size of the feature matrix extracted for all convolution kernels.

[0019] Furthermore, the process of constructing a multi-scale cross convolutional network joint model is as follows:

[0020] The trend and short-term disturbance characteristics of the input historical time series vector are extracted through a multi-scale convolutional network. After splicing, the comprehensive multi-scale load-green power-meteorological characteristics are obtained. A multi-scale cross-convolutional network joint model is constructed to deeply predict the future multi-variable time series. The expression of the multi-scale cross-convolutional network joint model is as follows:

[0021] F1=Conv1D k＝3 (X′ t ), F2 = Conv1D k＝5 (X′ t ), F3 = Conv1D k＝7 (X′ t )

[0022] F MICN =Concat(F1,F2,F3)

[0023]

[0024] In the formula, F1 is the local time series feature with a scale of 3, F2 is the medium-term time series feature with a scale of 5, F3 is the long-term time series feature with a scale of 7, and Conv1D k＝x (·) is a one-dimensional convolution operation, x is the convolution kernel size, F MICN is the output after multi-scale feature fusion, Concat(·) is the concatenation operation according to feature dimension; X′ t It is the feature compressed by the convolution operation; is the predicted total load power, For the predicted green power output, To predict the discharge power of the energy storage system, H is the total length of the prediction time step, MICN decoder (·) is a multi-scale cross convolutional neural network for time series prediction.

[0025] Furthermore, based on the multi-scale cross convolutional network joint model, the green electricity ratio, available green electricity power, and green certificate acquisition are predicted:

[0026] Calculate the proportion of green electricity that can be absorbed by unit load and the green electricity absorption power at each moment in the future to measure the system's green electricity utilization capacity:

[0027]

[0028] Where, is the predicted green electricity factor, which indicates the proportion of green electricity in the unit load; is the total load power at the predicted time t+τ, For the predicted green power output, is the predicted discharge power of the energy storage system, τ is the prediction step index;

[0029] Convert the green electricity power that can be absorbed per unit time into the number of certificates that can be obtained, and quantify the green performance potential of each period.

[0030]

[0031] Where, The maximum green power that the load can absorb is used. The min function is used to limit the green power absorption to not exceed the actual load demand.

[0032] The total number of certificates in the forecast period is accumulated, the amount of certificates obtained in the forecast period is accumulated over time, and the total acquisition capacity of green certificates in the future cycle is estimated.

[0033]

[0034] Where, is the number of green certificates at the predicted time t+τ, Δt is the duration of a single time step, and the constant 1000 is the green certificate unit conversion constant;

[0035]

[0036] Where, is the predicted total number of green certificates.

[0037] Furthermore, based on the prediction results of the multi-scale cross-convolutional network joint model, a regulation optimization model was constructed based on the combination of spatiotemporal optimization, differential equation constraints, and green certificates. With the optimization goals of minimizing load regulation costs, minimizing electricity costs, maximizing certificate benefits, and minimizing compliance penalties, the adjustable load and energy storage charging and discharging power are controlled. The process is as follows:

[0038] The following formula is used to define the control variables, and the power consumption structure of the glass factory is dynamically adjusted by adjusting the power up and down of the adjustable load, and the charging power and discharging power of the energy storage:

[0039]

[0040] Where u(t+τ) is the control variable at time t+τ, is the upper and lower adjustment value of the adjustable load at time t+τ, is the charging power of the energy storage system at time t+τ, is the discharge power of the energy storage system at time t+τ, τ is the prediction step index;

[0041] The objective function of the glass factory load green electricity dispatch is constructed to minimize electricity expenditure, load regulation costs, and certificate compliance penalties, while maximizing the benefits of green certificates. The expression of the control optimization model is as follows:

[0042]

[0043] Where J is the optimization objective function value, H is the total length of the prediction time step, and C elec (t+τ) is the electricity cost at time t+τ, C adj (t+τ) is the load regulation cost at time t+τ, R cert (t+τ) is the income from green certificates at time t+τ, P penalty (t+τ) is the penalty for not completing the certificate performance target at time t+τ, λ1 and λ2 are the weighting coefficients;

[0044]

[0045] In the above formula, To predict load power, is the upper and lower adjustment value of the adjustable load, P res To predict green power output, The charging power of the energy storage system, is the discharge power of the energy storage system, E(t+τ) is the electricity price at time t+τ, α is the load regulation cost coefficient, p cert The income price of a single green certificate, The target number of performance certificates, is the number of green certificates predicted at time t+τ, μ is the penalty price for missing certificates, and max{0,·} indicates that the penalty will only be incurred when the actual number of certificates is less than the target.

[0046] Furthermore, differential equation constraints are introduced to describe the dynamic process of the glass factory's operating status changing with the control variables, as shown in the following equation:

[0047]

[0048] Where SOC(t+τ) is the energy storage state of charge at time t+τ, η c is the energy storage charging efficiency, η d is the energy storage discharge efficiency, P grid (t+τ) is the real-time power purchased by the glass factory from the power grid at time t+τ, is the predicted load power, P res To predict green power output, For the upper and lower adjustment values ​​of the adjustable load, The charging power of the energy storage system, is the discharge power of the energy storage system;

[0049] The conditions of the constraint control optimization model are as follows:

[0050]

[0051] SOC min ≤SOC(1+τ)≤SOC max

[0052]

[0053] Where, P i min and P i max is the minimum and maximum safe operating power of load equipment i, is the predicted power of load device i, δ ramp is the maximum adjustment rate allowed by the system, is the adjustable load adjustment value at the next moment t+τ, SOC min and SOC max The allowable range of the state of charge of the energy storage device, is the rated maximum charge and discharge power of the energy storage system.

[0054] Furthermore, the optimization strategy is executed and rolling feedback is performed. The process is as follows:

[0055] The control variables output by the control optimization model at each time step are converted into actual control instructions and sent to the device layer, as shown in the following formula:

[0056]

[0057] Where u * (t) is the optimal control variable at time t, The load change value after adjustment at time t, is the charging power change value after adjustment at time t, is the discharge power change value after adjustment at time t;

[0058] The glass factory status after implementation is monitored in real time, and key variables including actual load, green power output, energy storage discharge, actual green power absorption, and green power factor are updated. Feedback data is used to revise the actual estimates of green power utilization and the number of green certificates, providing dynamic input for performance accounting:

[0059]

[0060]

[0061] In the above formula, is the actual load power at time t, is the load power predicted at time t, ε load (t) is the execution error at time t, which is used for feedback correction; is the green power actually absorbed at time t, is the actual green power output at time t, is the optimal discharge power at time t, GPF real (t) is the actual green power factor at time t, is the number of green certificates actually obtained at time t, Δt is the time interval, and 1000 is the conversion constant.

[0062] Furthermore, the actual green electricity absorption amount is dynamically compared with the target compliance progress:

[0063] The actual number of green certificates obtained is calculated based on the feedback of green electricity absorption value, and compared with the target performance cumulatively to obtain the current performance deviation. This deviation value is a key indicator for judging whether it is necessary to adjust the regulatory strategy or enter the certificate trading market, ensuring the dynamic controllability of the performance process. The cumulative performance status of green certificates is defined as follows:

[0064]

[0065] Where, is the cumulative number of green certificates at the next moment, is the cumulative number of green certificates obtained up to time t;

[0066] Compare the cumulative target and actual achievement gap using the following formula:

[0067]

[0068] If ε cert (t) is greater than 0, indicating that the performance has been met;

[0069] If ε cert If (t) is less than 0, it means there is a performance gap, triggering compensation adjustment or green certificate purchase mechanism;

[0070] Where, ε cert (t) is the current performance deviation, The number of performance target certificates that should be completed currently.

[0071] Furthermore, ε cert When (t) is less than 0, the re-optimization or green certificate purchase mechanism is triggered as follows:

[0072] Start reoptimization by updating the control variable bounds:

[0073]

[0074] Where, is the adjusted value after correction at the next moment, RampUp(·) is the upward adjustment function in response to the performance deviation, and the deviation value ε cert (t) trigger; u *,new (t+1) is the new optimized control variable, J(t+1) is the objective function at the next moment, u is the initial control variable, and θ is the tolerable performance deviation threshold. If the deviation exceeds the tolerance range, re-optimization is triggered.

[0075] If the contract cannot be fulfilled through further optimization prediction, the green certificate market will be entered and the purchase of green certificates will be initiated as follows:

[0076]

[0077] Where C buy (t) is the purchase cost of green certificates, is the number of certificates that need to be purchased from the market. It is the current market price of green certificates.

[0078] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention is based on a glass factory load green electricity availability and certificate acquisition method based on a multi-scale cross convolutional network, and uses a multi-scale convolutional nested network to mine load behavior, distributed energy output and time coupling characteristics, and accurately identify the green electricity availability and green certificate obtainability of different process sections; constructs a glass factory load green electricity regulation optimization model based on the combination of spatiotemporal optimization, differential equation constraints and green certificates, integrates glass factory process constraints, green electricity output predictions and certificate income weights, designs a multi-objective flexible regulation mechanism, and improves green electricity absorption efficiency and certificate income; proposes a real-time control and green compliance closed-loop strategy for glass factory loads, constructs a real-time regulation-green electricity utilization-certificate verification closed-loop, realizes the integration of dynamic adjustment, real-time verification and compliance feedback, and improves the executability of regulation and compliance. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1This is a flow chart of a method for intelligently controlling green electricity load in a glass factory based on green certificates proposed in the present invention. DETAILED DESCRIPTION

[0080] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0081] A green electricity intelligent control method for glass factory load based on green certificate, such as Figure 1 As shown, the method mainly includes the following three steps:

[0082] Step 1: Based on the multi-scale cross convolutional network, a multi-scale cross convolutional network joint model is constructed that integrates the glass factory load, power supply, energy storage and meteorological information to predict the green electricity ratio, available green electricity power and the amount of green certificates obtained.

[0083] Step 2: Build a regulation optimization model based on combining spatiotemporal optimization with differential equation constraints and green certificates, comprehensively considering load regulation costs, electricity costs, certificate income and performance penalties, and dynamically optimize the full time sequence of load regulation and energy storage operations.

[0084] Step three: Execute the optimization strategy and conduct rolling feedback, dynamically comparing the actual green electricity absorption with the target compliance progress. If any deviation is found, re-optimization or green certificate purchase mechanism will be triggered.

[0085] This implementation further explains the above main steps.

[0086] Based on the multi-scale cross convolutional network, the process of building a multi-scale cross convolutional network joint model that integrates the glass factory load, power supply, energy storage and meteorological information is as follows:

[0087] First, multi-dimensional input feature construction: We propose information collection technology for evaluating the sustainable regulation capacity of glass factories to provide data support. We collect and construct historical time series of glass factory load, green power output, energy storage status, and meteorological data. We select the length of the historical window and construct the input historical time series vector to reflect the dynamic relationship between electricity consumption and energy supply. The historical time series vector is represented as follows:

[0088]

[0089] Where, X t is the input tensor at time t, L is the length of the historical input window (such as 24 represents the past 24 time points), P load (t-L+1:t) is the total load power of the glass factory (unit: kW), P res(t-L+1:t) is the output of distributed renewable energy (such as photovoltaic) (unit: kW), is the energy storage system discharge power (unit: kW), SOC(t-L+1:t) is the energy storage charge state, ranging from [0,1], T(t-L+1:t) is the ambient temperature (unit: °C), and W(t-L+1:t) is the wind speed or light intensity (the unit depends on the sensor).

[0090] Secondly, the multi-scale cross convolutional network encoding-decoding structure extracts the trend and short-term disturbance characteristics of the input historical time series through multi-scale convolution, captures the coupling pattern of load and green power, and realizes in-depth prediction of future multi-variable time series. The expression of the feature compression layer using multi-scale convolution kernel for parallel modeling is as follows:

[0091]

[0092] Where X′ t represents the features compressed by the convolution operation, LayerNorm(·) represents the multi-scale convolution kernel function, and W proj represents the multi-scale convolution kernel coefficient, The size of the feature matrix extracted for all convolution kernels.

[0093] The multi-scale convolutional network extracts the trend and short-term disturbance characteristics of the input historical time series vector. After splicing, the comprehensive multi-scale load-green power-meteorological characteristics are obtained. The multi-scale cross-convolutional network joint model is constructed to deeply predict the future multi-variable time series. The expression of the multi-scale cross-convolutional network joint model is as follows:

[0094] F1=Conv1D k＝3 (X′ t ), F2 = Conv1D k＝5 (X′ t ), F3 = Conv1D k＝7 (X′ t ) (3)

[0095] F MICN =Concat(F1,F2,F3) (4)

[0096]

[0097] In the above formula, F1 is the local time series feature (short-term trend) with a scale of 3, F2 is the medium-term time series feature with a scale of 5, and F3 is the long-term time series feature with a scale of 7. Conv1D k＝x (·) is a one-dimensional convolution operation, x is the convolution kernel size, F MICN is the output after multi-scale feature fusion, Concat(·) is the concatenation operation according to the feature dimension, and the result dimension is L×3dn . is the predicted total load power, For the predicted green power output, To predict the discharge power of the energy storage system, H is the total length of the prediction time step (e.g. H = 96 means the next 24 hours and 15-minute granularity), MICN decoder (·) is a multi-scale cross convolutional neural network for time series prediction.

[0098] Based on the multi-scale cross convolutional network joint model, the green electricity ratio, available green electricity power, and green certificate acquisition are predicted:

[0099] Calculate the proportion of green electricity that can be absorbed by unit load and the green electricity absorption power at each moment in the future to measure the system's green electricity utilization capacity:

[0100]

[0101] Where, is the predicted green electricity factor, which represents the proportion of unit load that comes from green electricity (dimensionless, 0–1), the numerator represents the total green electricity supply (photovoltaic + energy storage), and the denominator is the predicted load power; is the total load power at the predicted time t+τ (unit: kW), is the predicted green power output (unit: kW), is the predicted discharge power of the energy storage system (unit: kW), τ is the prediction step index, which represents the future time point relative to the current t.

[0102] Convert the green electricity power that can be absorbed per unit time into the number of certificates that can be obtained, and quantify the green performance potential of each period.

[0103]

[0104] Where, The maximum green power that the load can absorb (unit: kW). The min function is used to limit the green power absorption to not exceed the actual load demand.

[0105] The total number of certificates during the forecast period is accumulated, and the amount of certificates obtained during the forecast period is accumulated over time to estimate the total green certificate acquisition capacity in the future cycle:

[0106]

[0107] Where, is the number of green certificates predicted at time t+τ (unit: pieces), Δt is the duration of a single time step (unit: hours, such as 15 minutes is 0.25), and The product represents the actual amount of green electricity absorbed during the period (kWh); the constant 1000 is the green certificate unit conversion constant (1 certificate = 1000kWh);

[0108]

[0109] Where, is the total number of green certificates in the forecast period from t+1 to t+H (unit: pieces).

[0110] In step 2, based on the prediction results of the multi-scale cross-convolutional network joint model, a regulation optimization model is constructed based on the combination of spatiotemporal optimization, differential equation constraints, and green certificates. The optimization objectives are to minimize load regulation costs, minimize electricity costs, maximize certificate benefits, and minimize compliance penalties. The control variables include adjustable load and energy storage charging and discharging power. The process is as follows:

[0111] Define control variables: By adjusting the power of the load, the charging power and the discharging power of the energy storage, the power structure of the glass factory is dynamically adjusted. These variables serve as the optimization inputs in the control optimization model, determining the green power absorption level, energy consumption distribution method and ultimate performance capacity at each moment, and are the basis for calculating the optimization objective function. Use the following formula to define the control variables:

[0112]

[0113] Where u(t+τ) is the control variable at time t+τ, is the upper and lower adjustment value of the adjustable load at time t+τ (unit: kW), is the charging power of the energy storage system at time t+τ, is the discharge power of the energy storage system at time t+τ, τ is the prediction step index, and represents the future time point relative to the current t.

[0114] A control optimization model was constructed to minimize electricity costs, load regulation costs, and certificate compliance penalties for the glass factory's load green electricity scheduling. This model integrates the economic efficiency and compliance of green electricity, ensuring that the model maximizes green value while ensuring process stability. This is the core basis for calculating the control strategy. The expression of the control optimization model is as follows:

[0115]

[0116] Where J is the optimization objective function value (total operating cost), C elec (t+τ) is the electricity cost at time t+τ (unit: yuan), C adj (t+τ) is the load regulation cost at time t+τ (unit: yuan), R cert(t+τ) is the income from green certificates at time t+τ (unit: yuan), P penalty (t+τ) is the penalty (unit: yuan) generated by failure to complete the certificate performance target at time t+τ, λ1 and λ2 are weighting coefficients, which respectively control the influence weight of benefits and penalties in the total cost.

[0117] in:

[0118]

[0119] In the above formula, is the predicted load power (kW), is the upper and lower adjustment value of the adjustable load, P res is the predicted green power output (kW), E(t+τ) is the electricity price at time t+τ (yuan / kWh), α is the load adjustment cost coefficient (yuan / kW), ||·|| is the absolute value, indicating that both upward and downward adjustments are included in the cost, p cert is the income price of a single green certificate (yuan / piece), is the target number of performance certificates (pieces), μ is the penalty price for missing certificates (yuan / piece), and max{0,·} indicates that the penalty is only incurred when the actual number of certificates is less than the target. Furthermore, differential equation constraints are introduced to describe the dynamic process of the glass factory's operating state as the control variables change, including the charge and discharge rate of the energy storage SOC and the dynamic response of total power consumption. This reflects the control optimization model's ability to characterize time continuity and physical processes, enabling it to describe the smooth evolution of the process system over time series, ensuring the engineering feasibility of the control strategy, as shown in the following formula: The above formulas respectively represent the charge and discharge change rate of energy storage SOC, the energy storage SOC differential equation; the dynamic response of total power consumption, and the change rate of total purchased power. Among them, SOC(t+τ) is the energy storage charge state at time t+τ, η c is the energy storage charging efficiency, η d is the energy storage discharge efficiency, P grid (t+τ) is the real-time power purchased by the glass factory from the power grid at time t+τ, is the predicted load power, P res To predict green power output, For the upper and lower adjustment values ​​of the adjustable load, The charging power of the energy storage system, is the discharge power of the energy storage system; the conditions of the constraint control optimization model are as follows: SOC min ≤SOC(1+τ)≤SOC max (20) Where, P i min and P imax is the minimum and maximum safe operating power of load equipment i, is the predicted power of load device i, |·| is the absolute value, indicating that both the upward and downward adjustments are included in the cost, δ ramp The maximum adjustment rate allowed by the system is used to protect the stability of the equipment. is the adjustable load adjustment value at the next moment t+τ, SOC min and SOC max The allowable range of the state of charge of the energy storage device, is the rated maximum charge and discharge power of the energy storage system.

[0120] In step three, a real-time load control and green compliance closed-loop strategy for the glass factory is proposed, the control results generated by STGODE are executed in the production control system, and closed-loop control of green electricity consumption and green certificate compliance throughout the entire process is achieved.

[0121] Execute the optimization strategy and provide rolling feedback. The process is as follows:

[0122] Generate real-time control instructions: The control variables output by the control optimization model at each time step are converted into actual control instructions and sent to the equipment layer to drive the operation of the glass factory system. The execution results directly affect the subsequent green electricity utilization and certificate acquisition, and are the execution entry point for green response behavior.

[0123]

[0124] Where u * (t) is the optimal control variable at time t, The load change value after adjustment at time t, is the charging power change value after adjustment at time t, is the discharge power change value after adjustment at time t.

[0125] Then, adjustment instructions are issued to the adjustable load; charging and discharging control instructions are issued to the energy storage system; and the prediction between load, green electricity and certificate is converted into an action plan.

[0126] Status tracking and real-time feedback updates: The status of the glass factory after execution is monitored in real time, and key variables including actual load, green power output, energy storage discharge, actual green power absorption, and green power factor are updated. Feedback data is used to correct the actual estimates of green power utilization and the number of green certificates, providing high-precision dynamic input for performance accounting.

[0127]

[0128] In the above formula, is the actual load power at time t, is the load power predicted at time t, εload (t) is the execution error at time t, which is used for feedback correction; is the green power actually absorbed at time t, is the actual green power output at time t, is the optimal discharge power at time t, GPF real (t) is the actual green power factor at time t, is the number of green certificates actually obtained at time t, Δt is the time interval, and 1000 is the conversion constant.

[0129] Dynamically compare the actual green electricity absorption with the target compliance progress:

[0130] Cumulative performance status update and deviation estimation: Calculate the actual number of green certificates obtained based on the feedback of green electricity absorption value, and compare it with the target performance cumulatively to obtain the current performance deviation. This deviation value is a key indicator for judging whether to adjust the regulation strategy or enter the certificate trading market, ensuring the dynamic controllability of the performance process. The cumulative performance status of green certificates is defined as follows:

[0131]

[0132] Where, is the cumulative number of green certificates at the next moment, is the cumulative number of green certificates obtained up to time t;

[0133] Compare the cumulative target and actual achievement gap using the following formula:

[0134]

[0135] If ε cert (t) is greater than 0, indicating that the performance has been met;

[0136] If ε cert If (t) is less than 0, it means there is a performance gap, triggering compensation adjustment or green certificate purchase mechanism;

[0137] Where, ε cert (t) is the current performance deviation, The number of performance target certificates that should be completed currently.

[0138] A closed-loop strategy adjustment mechanism triggers strategy adjustments based on contract performance deviations. When contract performance is insufficient, the system prioritizes adjusting the load and energy storage scheduling strategies for the next period to achieve re-optimization. If re-optimization fails to achieve contract performance targets, a green certificate market purchase order is triggered to ensure contract compliance. This mechanism establishes a closed-loop control strategy of prediction-execution-feedback-correction.

[0139] ε certWhen (t) is less than 0, the re-optimization or green certificate purchase mechanism is triggered as follows:

[0140] Start reoptimization by updating the control variable bounds:

[0141]

[0142] Where, is the adjusted value after correction at the next moment, RampUp(·) is the upward adjustment function in response to the performance deviation, and the deviation value ε cert (t) trigger; u *,new (t+1) is the new optimized control variable, J(t+1) is the objective function at the next moment, u is the initial control variable, and θ is the tolerable performance deviation threshold. If the deviation exceeds the tolerance range, re-optimization is triggered.

[0143] If the contract cannot be fulfilled through further optimization prediction, the green certificate market will be entered and the purchase of green certificates will be initiated as follows:

[0144]

[0145] Where C buy (t) is the purchase cost of green certificates, is the number of certificates that need to be purchased from the market. It is the current market price of green certificates.

[0146] The closed-loop compliance update logic forms a complete closed-loop control chain, integrating forecasting, control optimization, real-time execution, and compliance tracking and feedback mechanisms to achieve dynamic management of the entire process of green electricity regulation and green certificate compliance. Through periodic rolling updates, the system has adaptive, self-correcting, and self-optimizing capabilities, enhancing the level of intelligent green operations.

[0147] It should be noted that the parts not covered by the present invention are the same as the existing technology or can be implemented by using the existing technology. The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art who, within the technical scope disclosed by the present invention, makes equivalent substitutions or changes based on the technical solution and inventive concept of the present invention shall be covered by the scope of protection of the present invention.

Claims

1. A glass factory load green electricity intelligent control method based on green certificates, characterized in that: The following steps are involved: Step 1: Based on a multi-scale cross-convolutional network, a joint multi-scale cross-convolutional network model is constructed that integrates the glass factory's load, power supply, energy storage, and meteorological information to predict the green electricity ratio, available green electricity power, and the amount of green certificates obtained. Step 2: Build a regulation optimization model based on combining spatiotemporal optimization with differential equation constraints and green certificates. This model comprehensively considers load regulation costs, electricity costs, certificate revenue, and compliance penalties to dynamically optimize the full time sequence of load regulation and energy storage operations. Step three: Execute the optimization strategy and conduct rolling feedback, dynamically comparing the actual green electricity absorption with the target compliance progress. If any deviation is found, re-optimization or green certificate purchase mechanism will be triggered.

2. The method for intelligently controlling green electricity load in a glass factory based on green certificates according to claim 1 is characterized in that: Before building the multi-scale cross-convolutional network joint model, the historical time series of the glass factory's load, power supply, energy storage, and meteorological information data were integrated, the historical window length was selected, and the input historical time series vector was constructed. The historical time series vector is represented as follows: Where, X t is the input tensor at time t, L is the length of the historical input window, P load (t-L+1:t) is the total load power of the glass factory, P res (t-L+1:t) is the output of distributed renewable energy, is the discharge power of the energy storage system, SOC(t-L+1:t) is the state of charge of the energy storage, T(t-L+1:t) is the ambient temperature, and W(t-L+1:t) is the wind speed or light intensity; The expression of the feature compression layer using multi-scale convolution kernels for parallel modeling is as follows: Where, LayerNorm(·) represents the multi-scale convolution kernel function, W proj represents the multi-scale convolution kernel coefficient, The size of the feature matrix extracted for all convolution kernels.

3. The method for intelligently controlling green electricity load in a glass factory based on green certificates according to claim 2 is characterized in that: The process of building a multi-scale cross convolutional network joint model is as follows: The trend and short-term disturbance characteristics of the input historical time series vector are extracted through a multi-scale convolutional network. After splicing, the comprehensive multi-scale load-green power-meteorological characteristics are obtained. A multi-scale cross-convolutional network joint model is constructed to deeply predict the future multi-variable time series. The expression of the multi-scale cross-convolutional network joint model is as follows: F1=Conv1D k＝3 (X t '),F2=Conv1D k＝5 (X t '),F3=Conv1D k＝7 (X t ') F MICN =Concat(F1,F2,F3) In the formula, F1 represents the local time series feature with a scale of 3, F2 represents the medium-term time series feature with a scale of 5, and F3 represents the long-term time series feature with a scale of 7. Conv1D k＝x (·) is a one-dimensional convolution operation, x is the convolution kernel size, F MICN is the output after multi-scale feature fusion, Concat(·) is the concatenation operation according to feature dimension; X t ' is the feature compressed by the convolution operation; is the predicted total load power, For the predicted green power output, To predict the discharge power of the energy storage system, H is the total length of the prediction time step, MICN decoder (·) is a multi-scale cross convolutional neural network for time series prediction.

4. The method for intelligently controlling green electricity load in a glass factory based on green certificates according to claim 3 is characterized in that: Based on the multi-scale cross convolutional network joint model, the green electricity ratio, available green electricity power, and green certificate acquisition are predicted: Calculate the proportion of green electricity that can be absorbed by unit load and the green electricity absorption power at each moment in the future to measure the system's green electricity utilization capacity: Where, is the predicted green electricity factor, which indicates the proportion of green electricity in the unit load; is the total load power at the predicted time t+τ, For the predicted green power output, is the predicted discharge power of the energy storage system, τ is the prediction step index; Convert the green electricity power that can be absorbed per unit time into the number of certificates that can be obtained, and quantify the green performance potential of each period. Where, The maximum green power that the load can absorb is used. The min function is used to limit the green power absorption to not exceed the actual load demand. The total number of certificates in the forecast period is accumulated, the amount of certificates obtained in the forecast period is accumulated over time, and the total acquisition capacity of green certificates in the future cycle is estimated. Where, is the number of green certificates at the predicted time t+τ, Δt is the duration of a single time step, and the constant 1000 is the green certificate unit conversion constant; Where, is the predicted total number of green certificates.

5. The method for intelligently controlling green electricity load in a glass factory based on green certificates according to claim 1 is characterized in that: Based on the prediction results of the multi-scale cross-convolutional network joint model, a regulation optimization model was constructed based on the combination of spatiotemporal optimization, differential equation constraints, and green certificates. With the optimization goals of minimizing load regulation costs, minimizing electricity costs, maximizing certificate benefits, and minimizing compliance penalties, the adjustable load and energy storage charging and discharging power are controlled. The process is as follows: The following formula is used to define the control variables, and the power consumption structure of the glass factory is dynamically adjusted by adjusting the load power up and down, and the charging power and discharging power of the energy storage: Where u(t+τ) is the control variable at time t+τ, is the upper and lower adjustment value of the adjustable load at time t+τ, is the charging power of the energy storage system at time t+τ, is the discharge power of the energy storage system at time t+τ, τ is the prediction step index; The objective function of the glass factory load green electricity dispatch is constructed to minimize electricity expenditure, load regulation costs, and certificate compliance penalties, while maximizing the benefits of green certificates. The expression of the control optimization model is as follows: Where J is the optimization objective function value, H is the total length of the prediction time step, and C elec (t+τ) is the electricity cost at time t+τ, C adj (t+τ) is the load regulation cost at time t+τ, R cert (t+τ) is the income from green certificates at time t+τ, P penalty (t+τ) is the penalty for not completing the certificate performance target at time t+τ, λ1 and λ2 are the weighting coefficients; In the above formula, To predict load power, is the upper and lower adjustment value of the adjustable load, P res To predict green power output, The charging power of the energy storage system, is the discharge power of the energy storage system, E(t+τ) is the electricity price at time t+τ, α is the load regulation cost coefficient, p cert The income price of a single green certificate, The target number of performance certificates, is the number of green certificates predicted at time t+τ, μ is the penalty price for missing certificates, and max{0,·} indicates that the penalty is incurred only when the actual number of certificates is less than the target.

6. The method for intelligently controlling green electricity load in a glass factory based on green certificates according to claim 5 is characterized in that: Differential equation constraints are introduced to describe the dynamic process of the glass factory's operating status changing with the control variables, as shown in the following formula: Where SOC(t+τ) is the energy storage state of charge at time t+τ, η c is the energy storage charging efficiency, η d is the energy storage discharge efficiency, P grid (t+τ) is the real-time power purchased by the glass factory from the power grid at time t+τ, is the predicted load power, P res To predict green power output, For the upper and lower adjustment values ​​of the adjustable load, The charging power of the energy storage system, is the discharge power of the energy storage system; The conditions of the constraint control optimization model are as follows: SOCIETY min ≤SOC(1+τ)≤SOC max Where, P i min and P i max is the minimum and maximum safe operating power of load equipment i, is the predicted power of load device i, δ ramp is the maximum adjustment rate allowed by the system, is the adjustable load adjustment value at the next moment t+τ, SOC min and SOC max The allowable range of the state of charge of the energy storage device, is the rated maximum charge and discharge power of the energy storage system.

7. The method for intelligently controlling green electricity load in a glass factory based on green certificates according to claim 1 is characterized in that: Execute the optimization strategy and provide rolling feedback. The process is as follows: The control variables output by the control optimization model at each time step are converted into actual control instructions and sent to the device layer, as shown in the following formula: Where u * (t) is the optimal control variable at time t, The load change value after adjustment at time t, is the charging power change value after adjustment at time t, is the discharge power change value after adjustment at time t; The glass factory status after implementation is monitored in real time, and key variables including actual load, green power output, energy storage discharge, actual green power absorption, and green power factor are updated. Feedback data is used to revise the actual estimates of green power utilization and the number of green certificates, providing dynamic input for performance accounting: In the above formula, is the actual load power at time t, is the load power predicted at time t, ε load (t) is the execution error at time t, which is used for feedback correction; is the green power actually absorbed at time t, is the actual green power output at time t, is the optimal discharge power at time t, GPF real (t) is the actual green power factor at time t, is the number of green certificates actually obtained at time t, Δt is the time interval, and 1000 is the conversion constant.

8. The method for intelligently controlling green electricity load in a glass factory based on green certificates according to claim 7 is characterized in that: Dynamically compare the actual green electricity absorption with the target compliance progress: The actual number of green certificates obtained is calculated based on the feedback of green electricity absorption value, and compared with the target performance cumulatively to obtain the current performance deviation. This deviation value is a key indicator for judging whether it is necessary to adjust the regulatory strategy or enter the certificate trading market, ensuring the dynamic controllability of the performance process. The cumulative performance status of green certificates is defined as follows: Where, is the cumulative number of green certificates at the next moment, is the cumulative number of green certificates obtained up to time t; Compare the cumulative target and actual achievement gap using the following formula: If ε cert (t) is greater than 0, indicating that the performance has been met; If ε cert If (t) is less than 0, it means there is a performance gap, triggering compensation adjustment or green certificate purchase mechanism; Where, ε cert (t) is the current performance deviation, The number of performance target certificates that should be completed currently.

9. The method for intelligently controlling green electricity load in a glass factory based on green certificates according to claim 8 is characterized in that: ε cert When (t) is less than 0, the re-optimization or green certificate purchase mechanism is triggered as follows: Start reoptimization by updating the control variable bounds: Where, is the adjusted value after correction at the next moment, RampUp(·) is the upward adjustment function in response to the performance deviation, and the deviation value ε cert (t) trigger; u *,new (t+1) is the new optimized control variable, J(t+1) is the objective function at the next moment, u is the initial control variable, and θ is the tolerable performance deviation threshold. If the deviation exceeds the tolerance range, re-optimization is triggered. If the contract cannot be fulfilled through further optimization prediction, the green certificate market will be entered and the purchase of green certificates will be initiated as follows: Where C buy (t) is the purchase cost of green certificates, is the number of certificates that need to be purchased from the market. It is the current market price of green certificates.

Images