A method for optimizing demand-based electricity costs in a green electricity direct-connection system

Abstract

This invention discloses a method for optimizing demand charges in a green energy direct-connection system, comprising establishing a day-ahead MILP model and an intraday MILP model. The day-ahead MILP model uses next-day predicted renewable energy output data and a predetermined rigid load curve to complete global optimal planning and determine a baseline operating strategy. The objective function of the day-ahead MILP model is to minimize the combined cost of demand charges and electricity charges. The rigid load curve has fixed power at each time period and is not optimized or adjusted. The intraday MILP model continuously updates real-time data and modifies the operation based on the baseline operating strategy determined by the day-ahead MILP model, thus smoothing out fluctuations in renewable energy output and load disturbances. This invention is adaptable to the power supply architecture of dedicated green energy direct-connection lines, taking into account overall economic efficiency and avoiding real-time power spikes exceeding demand thresholds.

Description

Technical Field

[0001] This invention relates to the field of power system source-grid-load-storage dispatch optimization technology, specifically to a method for optimizing demand charges in a green electricity direct-connection system. Background Technology

[0002] With the advancement of my country's dual-carbon strategy, the green electricity direct connection model, which relies on dedicated transmission lines to enable new energy to be directly supplied to nearby load areas, has become an important form of implementation for the green transformation of the power industry. Compared with traditional public distribution network areas, green electricity direct connection areas have independent statistical methods for power output and capacity demand billing rules, and their power supply architecture and power coupling mechanisms are significantly different.

[0003] Rigid loads, such as data centers, lack the capacity for peak-shaving and demand-response adjustment, making it impossible to smooth out peak electricity demand through load-side peak shaving. Simultaneously, renewable energy output exhibits strong random fluctuations, and relying solely on direct supply of native green electricity can easily cause spikes in grid power output. To ensure power supply reliability, the grid must determine maximum demand capacity based on peak load, resulting in persistently high demand-based electricity costs for industrial parks. Energy storage, with its bidirectional rapid charging and discharging regulation capabilities, is the only flexible resource in this scenario that can be actively controlled.

[0004] In summary, while existing research has verified the feasibility of energy storage dispatching to reduce peak demand and lower demand charges, it is based on traditional public grid power supply scenarios and has not adapted to green electricity direct-connection power supply architectures. Furthermore, existing research has not focused on the rigid load characteristics under green electricity direct-connection scenarios, nor has it taken the smoothing of power fluctuations and reduction of demand charges for rigid loads as its core objectives. Therefore, there is an urgent need to propose a day-ahead / intra-day two-layer MILP dynamic dispatching method adapted to green electricity direct-connection scenarios and addressing the unadjustable characteristics of rigid loads, to achieve precise control of maximum demand and optimal demand charges. Summary of the Invention

[0005] The technical problem to be solved by this invention is to provide a method for optimizing the demand charge of a green electricity direct connection system, more specifically, a method adapted to the demand charge optimization of a green electricity direct connection system under rigid loads represented by data centers. This includes a two-layer dynamic optimization MILP model adapted to the green electricity direct connection power supply architecture and the non-adjustable characteristics of rigid loads, which solves the problems of weak anti-disturbance capability, low accuracy of maximum demand nonlinear solution, and inconsistency with the two-part electricity pricing optimization logic of traditional single-layer scheduling, as well as overcoming the difficulties of high peak power of green electricity direct connection parks, excessive monthly maximum demand, and high demand charge.

[0006] According to the technical solution of the present invention, the present invention provides a method for optimizing demand electricity charges in a green electricity direct connection system, including establishing a day-ahead MILP model and an intraday MILP model; Among them, the day-ahead MILP model is used to complete the global optimal planning and determine the baseline operation strategy based on the next day's predicted renewable energy output data and the established rigid load curve; the objective function of the day-ahead MILP model is to minimize the comprehensive cost of demand and electricity charges; the power of the rigid load curve is fixed in each period and is not optimized or adjusted. The intraday MILP model is used to continuously update real-time data and adjust the operation based on the baseline operating strategy determined by the day-ahead MILP model to smooth out fluctuations in renewable energy output and load disturbances. The objective function of the intraday MILP model is: , Where α is the real-time demand control weight, β is the day-ahead benchmark weight for grid power tracking, and γ is the energy storage output stability constraint weight; P g (τ) represents the real-time power purchase by the power grid during the τ time period; the intraday MILP model adopts a 15-minute rolling refresh mechanism and a sliding time window, T R Let τ be the set of intraday rolling window periods, where τ∈T R ; P represents the benchmark value of the power purchased by the power grid during the τ time period; c (τ) represents the real-time energy storage charging power during the τ period; This is the baseline value for energy storage charging power during the τ time period.

[0007] In some implementations, the objective function of the day-ahead MILP model is: , Among them, C demand For demand-based electricity pricing, c d P is the unit maximum demand electricity price. g,max This represents the maximum monthly demand limit; C energy For electricity bills, c e For industrial and commercial electricity prices, P g (t) represents the day-ahead forecast of grid power purchases during time period t, where Δt is the duration of a single time period, T D This is the set of scheduling cycles for the day before.

[0008] In some implementations, the monthly maximum demand limit Pg,max satisfies the following formula: , Where M is a constant, and the value of M is greater than the theoretical upper limit of the power purchased by the power grid; y(t) is a binary auxiliary variable, y(t)=1 when the power purchased by the power grid in time period t is at its peak, and y(t)=0 when the power purchased by the power grid in time period t is not at its peak.

[0009] In some implementations, the constraints of the day-ahead MILP model include: system power balance constraints, rigid load constraints, energy storage charging and discharging power and mutual exclusion constraints, energy storage state of charge timing constraints, and grid power purchase constraints. The system power balance constraint is as follows: , Among them, P pv (t) represents the day-ahead forecast of renewable energy output for time period t, P d (t) represents the predicted day-ahead energy storage discharge power for time period t, P L (t) represents the day-ahead forecast rigid load power for time period t, P c (t) represents the day-ahead predicted energy storage charging power for time period t, λ loss Active power loss in direct green electricity connection lines; Rigid load constraints are: ; The energy storage charging and discharging power and mutual exclusion constraints are as follows: the energy storage charging and discharging power shall not exceed the corresponding rated limit, and simultaneous charging and discharging are prohibited at the same time. The timing constraints for the state of charge of energy storage are: , , Where SOC(t) is the day-ahead predicted state of charge of energy storage for time period t, SOC(t+1) is the day-ahead predicted state of charge of energy storage for time period t+1, and η c For energy storage charging efficiency, η d For energy storage discharge efficiency, E N For the rated capacity of energy storage, SOC min As the limit of the state of charge of energy storage, SOC max This represents the upper limit of the energy storage state of charge. The power purchase limit of the power grid is: .

[0010] In some implementations, the baseline operating strategy determined by the day-ahead MILP model includes P obtained after solving the objective function of the day-ahead MILP model. c (t), P d (t), P g (t), SOC(t), P g,max .

[0011] In some implementations, the objective function of the intraday MILP model includes, P is determined by the day-ahead MILP model. g (t) is obtained through mapping and matching. P is determined by the day-ahead MILP model. c (t) is obtained through mapping and matching.

[0012] In some implementations, the constraints of the intraday MILP model include: real-time power balance constraints, rigid load constraints, energy storage charging and discharging power and mutual exclusion constraints, and demand upper limit hard constraints. The real-time power balance constraint is as follows: , Among them, P pv (τ) represents the real-time daily renewable energy output during the τ time period, P d (τ) represents the intraday real-time energy storage discharge power during the time period τ, P L (τ) represents the intraday real-time rigid load power during the τ time period; Rigid load constraints are: ; The energy storage charging and discharging power and mutual exclusion constraints are as follows: the energy storage charging and discharging power shall not exceed the corresponding rated limit, and simultaneous charging and discharging are prohibited at the same time. In addition, the following conditions must also be met: , Where, ΔP g,max ΔP is the maximum allowable correction for grid power. c,max ΔP is the maximum allowable correction amount for energy storage charging power. d,max This is the maximum allowable correction amount for energy storage discharge power; The reference value for energy storage discharge power during time period τ is P, determined by the day-ahead MILP model. d (t) is obtained through mapping and matching; The upper limit of demand is a hard constraint: .

[0013] In some implementations, the output P is obtained after the intraday MILP model is solved. g (τ), P c (τ), P d (τ) and based on P c (τ) and P d The intraday energy storage state of charge trajectory SOC(τ) is obtained from (τ).

[0014] In some implementations, the objective function of the intraday MILP model has α+β+γ=1 and α,β,γ>0.

[0015] In some implementations, the objective function of the intraday MILP model has α=0.65, β=0.25, and γ=0.10.

[0016] Compared with the prior art, the beneficial technical effects of the present invention are as follows: 1. Strong scenario adaptability: It is specially adapted to the green electricity direct connection dedicated line power supply architecture, which is in line with its downstream power statistics and two-part demand billing rules. It is different from the traditional distribution network park scheduling method. This method has been applied to the first data center green electricity direct connection project in the country and has achieved the expected results. 2. Adaptable to rigid loads: It follows the non-adjustable constraints of rigid loads throughout the process, does not depend on load-side demand response, and is suitable for high-rigidity scenarios such as data centers and industrial constant loads; 3. Economic optimization is more aligned with engineering: Demand cost + electricity cost are jointly optimized to take into account the overall economic efficiency, and the maximum demand is accurately solved through MILP linearization; 4. Multi-timescale disturbance resistance: The demand upper limit is locked a day-ahead, and the fluctuation is smoothed out every 15 minutes during the day to avoid real-time power peaks exceeding the demand threshold and to control demand electricity costs in the long term. 5. Practical engineering applications: The model has stable solutions and clear physical meanings of parameters, providing direct technical solutions for energy storage scheduling and demand management in parks with direct green electricity connection to high rigid loads. Attached Figure Description

[0017] Figure 1 This is a method architecture block diagram of an embodiment of the present invention. Detailed Implementation

[0018] This invention provides a method for optimizing demand-based electricity costs in a green electricity direct connection system. Specifically, it is a method for optimizing energy storage scheduling, maximum demand control, and demand-based electricity cost savings across multiple time scales for non-adjustable rigid loads, such as data centers, under a dedicated green electricity direct connection line park. This includes a two-layer dynamic optimization MILP model adapted to the green electricity direct connection power supply architecture and the non-adjustable characteristics of rigid loads. This solves the problems of weak anti-disturbance capability, low accuracy of nonlinear maximum demand calculation, and inconsistency with the two-part electricity pricing optimization logic of traditional single-layer scheduling. It also overcomes the difficulties of high peak power output, exceeding monthly maximum demand limits, and high demand-based electricity costs in green electricity direct connection parks.

[0019] The basic structure of the green electricity direct connection system targeted by this invention is based on existing technology, and mainly includes: a green electricity direct connection park, rigid loads, a new energy power station, a dedicated power supply line, an energy storage system, and a grid common connection point. The new energy power station supplies power to the green electricity direct connection park through a dedicated power supply line. The park's load is a rigid load represented by data centers and steel plants, which is connected to the grid through the grid common connection point. The green electricity direct connection park is equipped with an energy storage system.

[0020] Please see Figure 1In terms of park system modeling, this invention strictly adheres to the green electricity direct connection park construction model under rigid load. It explicitly defines the rigid load curve (time-series curve) as fixed and unadjustable, the day-ahead output of new energy as predictable, and only energy storage participating in power regulation as boundary conditions. It establishes a day-ahead MILP model and an intraday MILP model, forming a two-layer dynamic optimization architecture of day-ahead global optimization + intraday rolling correction. Both layers are based on mixed integer linear programming (MILP) modeling, achieving dynamic optimization through multi-timescale nesting. Specifically, the day-ahead MILP model uses next-day predicted data and a given rigid load curve to complete the global optimal planning and determine the baseline operating strategy. The power of the rigid load curve is fixed at each time period and is not optimized or adjusted. The objective function of the day-ahead MILP model is to minimize the combined cost of demand and electricity charges. The intraday MILP model employs a 15-minute rolling refresh mechanism and a sliding time window, updating real-time data in 15-minute sliding windows. It performs minor adjustments based on the baseline operating strategy determined by the day-ahead MILP model, smoothing out fluctuations in new energy output and load disturbances, while balancing global economic efficiency and real-time operational reliability. In this way, the data for the next day is predicted every day, the baseline operating strategy is determined based on the prediction, and then minor adjustments are made on the next day based on the actual real-time data.

[0021] Specifically, the day-ahead MILP model includes the following:

[0022] A typical day-ahead MILP model uses a 24-hour scheduling cycle and a 15-minute time resolution as inputs, including the day-ahead output forecast of new energy sources, a predetermined rigid load curve, energy storage equipment parameters, electricity prices and demand billing standards (including the maximum demand price per unit and the industrial and commercial electricity price), and the grid loss coefficient (i.e., the active power loss of green electricity direct connection lines). The objective function is to minimize the comprehensive cost of demand electricity charges and electricity charges. For example, the objective function is to minimize the comprehensive cost of monthly demand electricity charges and electricity charges, and the model solves for the daily energy storage charging and discharging plan and the grid purchase benchmark curve.

[0023] The objective function of the current layer MILP model is: , Among them, C demand For demand-based electricity pricing, c d P is the unit maximum demand electricity price. g,max This represents the maximum monthly demand limit; C energy For electricity bills, c e For industrial and commercial electricity prices, P g (t) represents the day-ahead forecast of grid power purchases during time period t; Δt represents the duration of a single time period, for example, Δt = 15 min; T D Let T be the set of day-ahead scheduling periods.D For example, a set formed by dividing the whole day into 96 15-minute intervals.

[0024] For solving the objective function of the day-ahead MILP model, this invention adopts the two-part electricity pricing logic for industrial and commercial users, using demand-based electricity pricing and energy-based electricity pricing as joint multi-objective optimization. The maximum demand (i.e., the monthly maximum demand ceiling Pg,max) is essentially the maximum power purchased by the grid in each time period, exhibiting a nonlinear extreme value relationship. Here, the Big M method is used to transform this nonlinear extreme value relationship into a linear MILP solvable form (i.e., the monthly maximum demand ceiling Pg,max). g,max (satisfies the following formula): , Where M is a constant, and to ensure that the feasible region is not lost during the above transformation, the value of M is strictly greater than the theoretical upper limit of the grid power purchase capacity; y(t) is a binary auxiliary variable, indicating whether the current grid power purchase capacity is at its peak. When the grid power purchase capacity is at its peak in time period t, y(t) = 1; when the grid power purchase capacity is not at its peak in time period t, y(t) = 0. The first formula P... g,max ≥P g (t) expresses the maximum demand constraint relationship.

[0025] The constraints of the current layer MILP model include: system power balance constraints, rigid load constraints, energy storage charging and discharging power and mutual exclusion constraints, energy storage state of charge timing constraints, and grid power purchase constraints (i.e., maximum demand maximum value constraints).

[0026] The system power balance constraint is: , Among them, P pv (t) represents the day-ahead forecast of renewable energy output for time period t, P g (t) represents the day-ahead forecast of grid power purchases (in other words, the day-ahead planned grid power purchases) for time period t, P d (t) represents the predicted day-ahead energy storage discharge power for time period t, P L (t) represents the day-ahead forecast rigid load power for time period t, P c (t) represents the day-ahead predicted energy storage charging power for time period t, λ loss Active power loss in direct green electricity connection lines; The rigid load constraint follows the principle of constant load rigidity. The optimization process strictly adopts a predetermined rigid load curve, and the rigid load power is fixed at each time period without optimization adjustment. .

[0027] The energy storage charging and discharging power and mutual exclusion constraints are as follows: the energy storage charging and discharging power shall not exceed the corresponding rated limit, and simultaneous charging and discharging are prohibited at the same time.

[0028] The energy storage state of charge (SOC) timing constraint is that the SOC is updated recursively in time intervals and always operates within a safe range, i.e.: , , Where SOC(t) is the day-ahead predicted state of charge of energy storage for time period t, SOC(t+1) is the day-ahead predicted state of charge of energy storage for time period t+1, and η c For energy storage charging efficiency, η d The energy storage discharge efficiency is given by E, where Δt is the duration of a single time period. N For the rated capacity of energy storage, SOC min As the limit of the state of charge of energy storage, SOC max This represents the upper limit of the energy storage state of charge.

[0029] The power purchase limit for the power grid is that the power purchase capacity of the power grid in any time period must not exceed the preset maximum demand limit, that is: .

[0030] The MILP solver is invoked to solve the objective function and constraints of the day-ahead MILP model. The baseline operating strategy determined by the day-ahead MILP model includes P obtained after solving the objective function of the day-ahead MILP model. c (t), P d (t), P g (t), SOC(t), P g,max In other words, the output of the daytime MILP model after solution is complete is: the daily energy storage charge and discharge time sequence plan (i.e., P...). c (t) and P d (t)), the grid power purchase baseline curve (i.e., P) g (t)), energy storage state of charge trajectory (i.e., SOC(t)), global demand control upper limit (i.e., P g,max The daily energy storage charging and discharging sequence plan strictly adheres to energy storage power limits and charging / discharging mutual exclusion constraints, serving as the daily energy storage output benchmark. The grid power purchase benchmark curve is controlled by system power balance constraints, serving as a reference standard for daily power tracking deviation. The energy storage state-of-charge trajectory is obtained by solving the SOC time-series recursive formula, satisfying the state-of-charge safety range constraint and providing initial conditions for daily energy storage power updates. The global demand control upper limit is the output result of the maximum demand linearization constraint, serving as both a parameter for calculating the day-ahead demand charge and a hard threshold that the daily grid power cannot exceed. These outputs collectively serve as the benchmark and constraint boundaries for the daily MILP model rolling optimization.

[0031] The intraday MILP model includes the following components.

[0032] A typical intraday MILP model employs a 15-minute rolling refresh mechanism and a sliding time window. With a 15-minute scheduling period, the sliding window receives real-time data on actual renewable energy output to identify prediction biases and minor load disturbances. Under the premise of strictly adhering to the baseline operating strategy obtained from the day-ahead MILP model and using a predetermined rigid load curve, minor adjustments are made to energy storage output to smooth power fluctuations and prevent temporary power spikes from exceeding demand thresholds. This is the core of dynamic optimization.

[0033] For the objective function of the intraday MILP model (which is a multi-objective optimization function), intraday optimization integrates three objectives: strictly controlling real-time demand, adhering to the day-ahead benchmark plan (i.e., grid power tracking the day-ahead benchmark), and smoothly adjusting energy storage output. To ensure the linearity of the model, the deviation term is linearized. The objective function of the intraday MILP model is as follows: , Where α is the real-time demand control weight, β is the day-ahead benchmark weight for grid power tracking, and γ is the energy storage output stability constraint weight; P g (τ) represents the real-time power purchase capacity of the power grid during the time period τ. The real-time energy storage charging power during the τ period; the intraday MILP model adopts a 15-minute rolling refresh mechanism and a sliding time window, T R Let τ be the set of intraday rolling window periods, where τ∈T R ; P represents the benchmark value of the power purchased by the power grid during the τ time period; c (τ) represents the real-time energy storage charging power during the τ period; This is the baseline value for energy storage charging power during the τ time period.

[0034] Furthermore, P is determined by the day-ahead MILP model. g (t) is obtained through mapping and matching. P is determined by the day-ahead MILP model. c (t) is obtained through mapping matching. Mapping matching refers to substituting the intraday time period τ into the corresponding curve determined before the previous day (such as P). g In (t), more specifically, for a certain moment within a day, the corresponding time period t from the previous day is looked up, and the corresponding value on the curve is obtained as the corresponding baseline value. In a preferred embodiment, T D and T R The two sets are established in the same way, which are formed by dividing the whole day into 96 15-minute time intervals.

[0035] The day-ahead MILP model completes the global optimal planning for the entire day, determining the system's baseline operating curve and control thresholds. The intraday layer performs rolling corrections with a 15-minute cycle, only mitigating minor power output disturbances from renewable energy sources without overturning the established day-ahead plan. The day-ahead and intraday layers share common variables, unified constraints, and progressive objectives, ultimately achieving multi-timescale collaborative optimization of "determining the optimal boundary day-ahead and ensuring real-time execution intraday," stably reducing maximum demand and controlling demand-based electricity costs without adjusting rigid loads.

[0036] Regarding the values ​​and sensitivity analysis of the aforementioned weights, in this invention, α, β, and γ are assigned values ​​based on engineering scheduling priority and normalization principles: α, β, and γ correspond to the real-time demand control weight, the day-ahead benchmark weight for grid power tracking, and the energy storage output stability constraint weight, respectively. This invention preferably performs range normalization on the intraday three-layer optimization objectives, then determines the benchmark values ​​of the weights based on the scheduling priority of green electricity direct connection park projects, and finally verifies the rationality of the weights through sensitivity analysis.

[0037] First, the weight coefficients satisfy the normalization constraint, namely: α+β+γ=1, α,β,γ>0.

[0038] Then, based on the actual operation and scheduling of green electricity directly connected to rigid load parks, the priority order was established as follows: strict control of the real-time maximum demand of the power grid takes priority, followed by adherence to the current global dispatch plan, and smoothing out fluctuations in energy storage output last, i.e., α>β>γ. Based on this, the baseline weights were set as follows: α=0.65, β=0.25, γ=0.10.

[0039] Finally, a weight sensitivity analysis is performed. To verify the influence of weight coefficient values ​​on the optimization results, the following three sets of weight gradient scenarios are set up for comparative testing: Scenario 1: α=0.70, β=0.20, γ=0.10; Scenario 2: α=0.65, β=0.25, γ=0.10; Scenario 3: α=0.60, β=0.25, γ=0.15.

[0040] By comparing four indicators—monthly maximum demand, demand electricity cost, grid benchmark tracking error, and energy storage output fluctuation—under different weights, the analysis shows that: increasing α can further reduce grid peak value and demand electricity cost, but will slightly increase the day-ahead plan tracking deviation; increasing β can make the intraday trajectory closer to the day-ahead benchmark, but the peak shaving and demand reduction effect is slightly weakened; increasing γ can significantly smooth energy storage output and reduce frequent converter adjustments, but will crowd out the priority of demand control. Taking into account both the economic efficiency of demand electricity cost and the stability of dispatch operation, the benchmark weight of Scenario 2 is finally selected as the fixed value for intraday optimization.

[0041] The constraints of the intraday MILP model include: real-time power balance constraints, rigid load constraints, energy storage charging and discharging power and mutual exclusion constraints, and demand upper limit hard constraints (i.e., intraday maximum power correction range constraints).

[0042] The real-time power balance constraint is to maintain the active power balance of the system based on the real-time output of new energy sources and the established rigid load curve, that is: , Among them, P pv (τ) represents the real-time daily renewable energy output during the τ time period, P d (τ) represents the intraday real-time energy storage discharge power during the time period τ, P L (τ) represents the intraday real-time rigid load power during the τ period (the load curves for each period are fixed and do not participate in optimization adjustments).

[0043] The rigid load constraint is (continuing the rigid load constraint in the daytime MILP model), following the principle of constant load rigidity. The optimization process strictly adopts the predetermined rigid load curve, and the rigid load power is fixed in each time period without optimization adjustment. .

[0044] The energy storage charging and discharging power and mutual exclusion constraints are consistent with the day-ahead rules, namely, the energy storage charging and discharging power shall not exceed the corresponding rated limit, and simultaneous charging and discharging are prohibited at the same time, thus restricting the energy storage operating state. The actual state of charge of the energy storage at the current moment is used as the initial value of the rolling window, and the power is updated hourly. Furthermore, to avoid local corrections from destroying the global optimal result, the maximum deviation between the intraday operating parameters and the day-ahead benchmark is limited, that is, the following conditions must also be met: , Where, ΔP g,max ΔP is the maximum allowable correction for grid power. c,max ΔP is the maximum allowable correction amount for energy storage charging power. d,max This is the maximum allowable correction amount for energy storage discharge power; The reference value for grid power purchase during time period τ is P, determined by the day-ahead MILP model. g (t) is obtained through mapping and matching; The reference value for energy storage charging power during time period τ is P, determined by the day-ahead MILP model. c (t) is obtained through mapping and matching. The reference value for energy storage discharge power during time period τ is P, determined by the day-ahead MILP model. d (t) is obtained through mapping and matching.

[0045] The hard constraint on the upper limit of demand is that the real-time power purchase capacity of the power grid must not exceed the maximum demand limit determined before the day before, that is: .

[0046] After the intraday MILP model is solved, the real-time grid power purchase capacity for the entire intraday period (i.e., P) is output. g (τ)), energy storage real-time charge and discharge timing (i.e., P) c (τ) and P d (τ)), and based on P c (τ) and P d The intraday energy storage state-of-charge trajectory SOC(τ) is obtained from (τ). The intraday MILP model uses the grid and energy storage power time series obtained from the day-ahead optimization as the tracking benchmark. It relies on the power deviation penalty term in the objective function to constrain the adjustment range, and only slightly corrects the energy storage charging and discharging operation commands. It also captures the fluctuations in new energy output and small disturbances in rigid loads in real time in the rolling time domain. It relies on the bidirectional charging and discharging level of energy storage to suppress the power difference, and at the same time, it uses 0≤P g (τ)≤P g The maximum power purchase limit is rigidly imposed, and the overall optimal framework is not overturned throughout the process. Short-term power disturbances are mitigated without exceeding the daily demand limit, and temporary power spikes are avoided from pushing up the monthly maximum demand.

[0047] To verify the effectiveness of the present invention, the following three comparative scenarios were set up for simulation comparison: Scenario 1: No energy storage is invested, only green electricity is directly connected to the native power supply; Scenario 2: Single-layer day-ahead scheduling only, without intraday rolling correction; Scenario 3: The present invention's day-to-day two-layer MILP dynamic scheduling method.

[0048] Simulation results show that the method of the present invention can effectively reduce the maximum monthly demand of the power grid without changing the electricity consumption behavior of rigid loads, significantly reduce the demand electricity cost, improve the local consumption rate of new energy, and ensure that the energy storage SOC operates stably and meets the daily cycle closed-loop requirements, thus demonstrating strong engineering practicality.

[0049] In summary, this invention boasts strong scenario adaptability, specifically tailored for direct green power connection scenarios with rigid loads. It aligns with the power statistics and two-part demand billing rules of these scenarios, differing from traditional distribution network park scheduling methods. It adheres to the non-adjustable constraints of rigid loads, eliminating reliance on load-side demand response, and is suitable for high-rigidity scenarios such as data centers and industrial constant loads. Its economic optimization is more aligned with engineering practices, employing joint optimization of demand-based electricity charges and energy-based electricity charges to consider overall economic efficiency. MILP linearization enables accurate calculation of maximum demand. It exhibits multi-timescale disturbance resistance, locking the demand upper limit day-ahead and smoothing fluctuations every 15 minutes within the day, preventing real-time power spikes from exceeding demand thresholds and effectively controlling demand-based electricity charges in the long term. It is readily applicable to engineering projects, with stable model solutions and clearly defined physical meanings of parameters, providing a direct technical solution for energy storage scheduling and demand management in parks with direct green power connections to high-rigidity loads.

[0050] As a supplementary explanation, in the prior art, regarding the green electricity direct connection system targeted by this invention, in June 2025, the National Development and Reform Commission and the National Energy Administration issued Document No. 650 of 2025, which for the first time established the institutional framework for green electricity direct connection at the national level, clarifying the basic rules such as the definition, planning management, and operation requirements of single-user green electricity direct connection. This marked a turning point for green electricity direct connection from local exploration to national regulation. In July 2025, Document No. 1192 of 2025 further improved the price mechanism, clarifying the price formation rules for the local consumption of new energy power generation, providing institutional guarantees for the economic feasibility of green electricity direct connection. In May 2026, the National Development and Reform Commission and the National Energy Administration jointly issued the "Notice on Relevant Matters Concerning the Orderly Promotion of the Development of Multi-User Green Electricity Direct Connection" (Document No. 688 of 2026), expanding the scope of application from single users to multi-users, releasing the scale potential of green electricity direct connection. Three policy documents work together to continuously promote the implementation of new energy consumption, green electricity trading, and source-load coordination mechanisms, forming a clear top-level guidance framework. While direct green electricity connection can achieve local supply of green electricity and carbon reduction, it suffers from problems such as intermittent fluctuations in wind and solar power output, lack of buffering by the main power grid, lack of self-regulation capabilities for rigid loads such as data centers, supply and demand mismatch, and power peaks pushing up rated demand, eroding green electricity revenue due to electricity costs.

[0051] Therefore, this invention is proposed. The method of this invention is precisely adapted to this scenario. It takes bidirectional energy storage regulation as the core, links prediction and coordinated scheduling, smooths the power curve by peak shaving and valley filling, reduces demand electricity costs, and improves the local consumption rate of green electricity. Moreover, it is in line with policy guidance, takes into account power supply stability, economic cost and low carbon benefits, and can be scaled up to computing power and industrial parks, with broad application prospects.

[0052] Meanwhile, the AI-related industries are expanding rapidly, with a large number of supercomputing and intelligent computing centers being built at an accelerated pace. Computing clusters are high-energy-consuming and rigid loads, leading to a sharp increase in overall electricity demand and a surge in pressure for low-carbon energy supply. In March 2026, "computing-power synergy" was included in the government work report for the first time, rising to the level of a national strategy. Green electricity direct connection, as the core path for reducing carbon emissions and costs in computing parks, has become a key carrier for the implementation of computing-power synergy, enabling point-to-point direct matching of wind and solar power with computing loads.

[0053] This method perfectly aligns with the needs of computing and power synergy development and has been successfully applied to the nation's first data center green power direct connection demonstration project. It has been verified as mature and reliable under actual working conditions and has outstanding advantages in standardized implementation, large-scale promotion, and rapid replication in multiple regions. It can be widely adapted to major hub nodes of East Data and West Computing, computing power clusters in the east and west, and green computing power base construction scenarios. It builds a stable and efficient green power consumption and supply system for massive intelligent computing and supercomputing parks across the country. It is a standardized benchmark solution for implementing the national dual-carbon strategy, consolidating the foundation of green computing power, and deepening the implementation of the national strategy of computing and power synergy.

[0054] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; obviously, the described embodiments are some embodiments of the present invention, but not all embodiments; based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention; in the absence of conflict, the embodiments and features in the embodiments of the present invention can be combined with each other; modifications to the technical solutions described in the foregoing embodiments, or equivalent substitutions for some of the technical features, do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for optimizing demand-based electricity costs in a green electricity direct-connection system, characterized in that, This includes establishing a day-ahead MILP model and an intraday MILP model; Among them, the day-ahead MILP model is used to complete the global optimal planning and determine the baseline operation strategy based on the next day's predicted renewable energy output data and the established rigid load curve; the objective function of the day-ahead MILP model is to minimize the comprehensive cost of demand and electricity charges; the power of the rigid load curve is fixed in each period and is not optimized or adjusted. The intraday MILP model is used to continuously update real-time data and adjust the operation based on the baseline operating strategy determined by the day-ahead MILP model to smooth out fluctuations in renewable energy output and load disturbances. The objective function of the intraday MILP model is: ， Where α is the real-time demand control weight, β is the day-ahead benchmark weight for grid power tracking, and γ is the energy storage output stability constraint weight; P g (τ) represents the real-time power purchase by the power grid during the τ time period; the intraday MILP model adopts a 15-minute rolling refresh mechanism and a sliding time window, T R Let τ be the set of intraday rolling window periods, where τ∈T R ; P represents the benchmark value of the power purchased by the power grid during the τ time period; c (τ) represents the real-time energy storage charging power during the τ period; This is the baseline value for energy storage charging power during the τ time period.

2. The method for optimizing demand-based electricity costs in a green electricity direct-connection system according to claim 1, characterized in that, The objective function of the current layer MILP model is: ， Among them, C demand For demand-based electricity pricing, c d P is the unit maximum demand electricity price. g,max This represents the maximum monthly demand limit; C energy For electricity bills, c e For industrial and commercial electricity prices, P g (t) represents the day-ahead forecast of grid power purchases during time period t, where Δt is the duration of a single time period, T D This is the set of scheduling cycles for the day before.

3. The method for optimizing demand-based electricity costs in a green electricity direct-connection system according to claim 2, characterized in that, The maximum monthly demand limit Pg,max satisfies the following formula: ， Where M is a constant, and the value of M is greater than the theoretical upper limit of the power purchased by the power grid; y(t) is a binary auxiliary variable, y(t)=1 when the power purchased by the power grid in time period t is at its peak, and y(t)=0 when the power purchased by the power grid in time period t is not at its peak.

4. The method for optimizing demand-based electricity costs in a green electricity direct-connection system according to claim 3, characterized in that, The constraints of the day-ahead MILP model include: system power balance constraints, rigid load constraints, energy storage charging and discharging power and mutual exclusion constraints, energy storage state of charge timing constraints, and grid power purchase constraints. The system power balance constraint is as follows: ， Among them, P pv (t) represents the day-ahead forecast of renewable energy output for time period t, P d (t) represents the predicted day-ahead energy storage discharge power for time period t, P L (t) represents the day-ahead forecast rigid load power for time period t, P c (t) represents the day-ahead predicted energy storage charging power for time period t, λ loss Active power loss in direct green electricity connection lines; Rigid load constraints are: ； The energy storage charging and discharging power and mutual exclusion constraints are as follows: the energy storage charging and discharging power shall not exceed the corresponding rated limit, and simultaneous charging and discharging are prohibited at the same time. The timing constraints for the state of charge of energy storage are: ， ， Where SOC(t) is the day-ahead predicted state of charge of energy storage for time period t, SOC(t+1) is the day-ahead predicted state of charge of energy storage for time period t+1, and η c For energy storage charging efficiency, η d For energy storage discharge efficiency, E N For the rated capacity of energy storage, SOC min As the limit of the state of charge of energy storage, SOC max This represents the upper limit of the energy storage state of charge. The power purchase limit of the power grid is: 。 5. The method for optimizing demand-based electricity costs in a green electricity direct-connection system according to claim 4, characterized in that, The baseline operating strategy determined by the day-ahead MILP model includes P obtained after solving the objective function of the day-ahead MILP model. c (t), P d (t), P g (t), SOC(t), P g,max .

6. The method for optimizing demand-based electricity charges in a green electricity direct-connection system according to claim 5, characterized in that, In the objective function of the intraday MILP model P is determined by the day-ahead MILP model. g (t) is obtained through mapping and matching. P is determined by the day-ahead MILP model. c (t) is obtained through mapping and matching.

7. The method for optimizing demand-based electricity costs in a green electricity direct-connection system according to claim 6, characterized in that, The constraints of the intraday MILP model include: real-time power balance constraints, rigid load constraints, energy storage charging and discharging power and mutual exclusion constraints, and demand upper limit hard constraints. The real-time power balance constraint is as follows: ， Among them, P pv (τ) represents the real-time daily renewable energy output during the τ time period, P d (τ) represents the intraday real-time energy storage discharge power during the time period τ, P L (τ) represents the intraday real-time rigid load power during the τ time period; Rigid load constraints are: ； The energy storage charging and discharging power and mutual exclusion constraints are as follows: the energy storage charging and discharging power shall not exceed the corresponding rated limit, and simultaneous charging and discharging are prohibited at the same time. In addition, the following conditions must also be met: ， Where, ΔP g,max ΔP is the maximum allowable correction for grid power. c,max ΔP is the maximum allowable correction amount for energy storage charging power. d,max This is the maximum allowable correction amount for energy storage discharge power; The reference value for energy storage discharge power during time period τ is P, determined by the day-ahead MILP model. d (t) is obtained through mapping and matching; The upper limit of demand is a hard constraint: 。 8. The method for optimizing demand-based electricity costs in a green electricity direct-connection system according to claim 7, characterized in that, After solving the intra-day MILP model, the output P is... g (τ), P c (τ), P d (τ) and based on P c (τ) and P d The intraday energy storage state of charge trajectory SOC(τ) is obtained from (τ).

9. The method for optimizing demand charges in a green electricity direct-connection system according to any one of claims 1 to 8, characterized in that, In the objective function of the intraday MILP model, α+β+γ=1, α,β,γ>0.

10. The method for optimizing demand charges in a green electricity direct-connection system according to any one of claims 1 to 8, characterized in that, In the objective function of the intraday MILP model, α=0.65, β=0.25, and γ=0.10.

Images