A new energy power generation and energy storage management method based on power peaks

Abstract

The application discloses a new energy power generation and energy storage management method based on power peak, and belongs to the technical field of new energy power generation and smart grid. The method constructs a dynamic peak identification model of regional power load, and establishes an interval prediction model of new energy power generation by combining meteorological data and historical power generation data; based on the two models, two dynamic decision factors of peak overlap degree and energy storage urgency are introduced to generate a forward-looking and adaptive energy storage charging and discharging strategy. The method fuses static load curve analysis and dynamic and uncertain power generation prediction depth, realizes optimal operation of energy storage under double targets of "peak load shifting" and "smooth fluctuation", and significantly improves local consumption capacity of new energy and stability of the power grid.

Description

Technical Field

[0001] This invention relates to the fields of new energy power generation and smart grid technology, and in particular to a new energy power generation and energy storage management method based on peak electricity demand. Background Technology

[0002] With the advancement of the "dual carbon" target, the penetration rate of new energy sources, represented by photovoltaics and wind power, in the power grid is constantly increasing. However, new energy power generation is intermittent, volatile, and uncertain, and its output often does not match the electricity load in time, posing a huge challenge to the stable operation of the power grid. Energy storage systems are considered a key means to solve this problem, as they can smooth out the fluctuations in new energy output and realize the spatiotemporal translation of electricity.

[0003] Existing methods for managing renewable energy generation and energy storage primarily focus on the generation side. For example, they control the charging and discharging of energy storage based on predicted power generation to smooth grid connection power. Alternatively, some methods rely solely on simple time-of-use pricing mechanisms for "low storage, high generation," failing to adequately consider the dynamic peak characteristics of regional electricity load. The limitations of these methods are:

[0004] Static strategy: The traditional "peak shaving and valley filling" strategy is based on a fixed division of peak and valley periods, which cannot adapt to the daily dynamic changes in the load curve, especially in areas that are greatly affected by factors such as weather and holidays.

[0005] Insufficient coupling with new energy sources: The dynamic prediction of load peaks has not been deeply coupled with the probabilistic prediction of new energy power generation. When a peak in electricity demand is predicted, but new energy power generation is subject to significant uncertainty, existing strategies lack forward-looking, quantitative decision-making basis to reserve sufficient energy in advance.

[0006] Single objective: Most of them only focus on the single objective of "smoothing power generation" or "peak shaving and valley filling", making it difficult to optimize the overall operating efficiency of the system under multiple constraints.

[0007] Therefore, there is an urgent need for a new energy storage management method that can finely couple dynamic load peaks with uncertain new energy power generation in order to improve the utilization efficiency of energy storage systems and the intelligence level of the power grid. Summary of the Invention

[0008] The purpose of this invention is to propose a new energy power generation and energy storage management method based on electricity demand peaks. This method aims to solve the problem that existing energy storage management strategies fail to finely couple the characteristics of regional electricity load peaks with the uncertainty of multi-source new energy power generation, resulting in low energy storage efficiency and high grid peak-shaving pressure.

[0009] The present invention is implemented as follows: a new energy power generation and energy storage management method based on peak electricity consumption, the method comprising: S1 collects real-time electricity load data, meteorological data, and historical and real-time power generation data of related new energy power plants in the target area, and preprocesses the collected data. S2, based on the electricity load data, predict the load curve within the future period T through a dynamic peak identification model, identify and quantify the peak characteristics based on the load curve, and construct the peak membership function for each peak according to the peak characteristics. S3. Based on the meteorological data and historical power generation data, the predicted range of new energy power generation within the future period T is obtained through the interval prediction model of new energy power generation. S4, calculate the peak overlap and energy storage urgency at each moment within the future period T; based on the peak overlap, energy storage urgency and the current state of charge of the energy storage system, generate a set of charging and discharging power instructions for the energy storage system within the period T according to preset rules. S5, execute the generated charging and discharging power instruction set, and monitor the actual power load and new energy power generation in real time, compare them with their corresponding predicted values, and calculate the prediction deviation respectively; if any deviation exceeds the corresponding set threshold, the strategy regeneration process is triggered, and the process returns to step S2.

[0010] Beneficial effects of the present invention This invention discloses a new energy power generation and energy storage management method based on electricity consumption peaks, belonging to the fields of new energy power generation and smart grid technology. The method constructs a dynamic peak identification model of regional electricity load and, combined with meteorological data and historical power generation data, establishes an interval prediction model for new energy power generation. Based on these two models, it introduces two dynamic decision factors: peak overlap and energy storage urgency, to generate a forward-looking and adaptive energy storage charging and discharging strategy. By deeply integrating static load curve analysis with dynamic and uncertain power generation prediction, this method achieves optimal operation of energy storage under the dual objectives of "peak shaving and valley filling" and "fluctuation smoothing," significantly improving the local absorption capacity of new energy and the stability of the power grid. Attached Figure Description

[0011] Figure 1 This is a flowchart of a preferred embodiment of the present invention for a new energy power generation and energy storage management method based on peak electricity consumption. Detailed Implementation

[0012] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. For ease of explanation, only the parts related to the embodiments of this invention are shown. It should be understood that the specific embodiments described herein are merely for explaining this invention and are not intended to limit this invention.

[0013] This invention discloses a new energy power generation and energy storage management method based on electricity consumption peaks. The method constructs a dynamic peak identification model of regional electricity load and, combined with meteorological data and historical power generation data, establishes an interval prediction model for new energy power generation. Based on these two models, it introduces two dynamic decision factors: peak overlap and energy storage urgency, to generate a forward-looking and adaptive energy storage charging and discharging strategy. By deeply integrating static load curve analysis with dynamic and uncertain power generation prediction, this invention achieves optimized operation of energy storage under the dual objectives of "peak shaving and valley filling" and "fluctuation smoothing," significantly improving the local absorption capacity of new energy and the stability of the power grid.

[0014] Figure 1 This is a flowchart of a preferred embodiment of the present invention for a new energy power generation and energy storage management method based on peak electricity demand; the method includes the following steps: S1: Real-time collection of electricity load data, meteorological data, and historical and real-time power generation data of associated new energy power plants in the target area, and preprocessing of the collected data; In this embodiment of the invention, the electricity load data is collected at intervals (usually one data point every 15 minutes or 1 hour) from smart meters or energy management systems in the target area (such as an industrial park, commercial building complex, or distribution network node); the electricity load data is the total active power consumption data. The meteorological data is obtained from local meteorological stations or authoritative meteorological service agencies, including forecast meteorological data within the future period T and historical meteorological data for the same period. The meteorological data includes, but is not limited to, irradiance, ambient temperature, wind speed and wind direction.

[0015] The new energy power generation data is collected in real time from grid-connected new energy power plants (such as photovoltaic inverters and wind turbine monitoring systems); the new energy power generation data includes actual power generation and historical power generation data; The preprocessing of the collected data includes: Data cleaning: This involves verifying the quality of the collected raw data and handling issues such as missing data and abrupt changes. For example, interpolation can be used to repair short-term missing data, or moving average filtering can be used to smooth out abnormal fluctuations.

[0016] Data alignment: unifying data from different sources and with different sampling frequencies to the same timestamp and sampling interval to form a regular time series dataset.

[0017] Data normalization: Load data, power generation data, etc., are normalized and mapped to the [0,1] interval, for example, using the minimum-maximum normalization method. S2: Based on the electricity load data, the load curve L(t) within a future period T (preset period T) is predicted using a dynamic peak identification model. Based on the load curve L(t), peak characteristics {Peak_i} are identified and quantified, and a peak membership function is constructed for each peak according to the peak characteristics {Peak_i}. This includes:

[0018] S21, employs a hybrid forecasting model combining time series decomposition and machine learning to predict the load curve for future periods T (e.g., 24 hours or 48 hours); The STL (Seasonal-Trend decomposition using Loess) algorithm is used to decompose historical load data into a trend term, a seasonal term (including daily / weekly cycles), and a residual term. The trend term sequence is extrapolated using polynomial fitting to obtain the predicted value of the trend term in the future period T; Fourier series analysis is applied to the seasonal term sequence to extract the fundamental and harmonic components of its inherent daily and weekly fixed cyclic patterns, and the predicted value of the future seasonal term is obtained by directly copying the cyclic pattern to the future period T; the residual term sequence is input into a gated recurrent unit network for training and prediction to obtain the prediction of the future residual term. By superimposing the predicted components, the load forecast curve for the future period T is obtained; The formula for the load forecast curve is: L(t) = T_trend(t) + S_seasonal(t) + R_residual(t); Where L(t) represents the predicted total electricity load at time t; T_trend(t) represents the predicted value of the load trend term at time t within a future period T; S_seasonal(t) represents the predicted value of the load seasonal term at time t within a future period T; and R_residual(t) represents the predicted value of the load residual term at time t within a future period T.

[0019] The dynamic peak identification model adopts a hybrid model that combines time series decomposition and machine learning; S22, based on the predicted load curve L(t), identify the peaks and quantify their characteristics; including: The load prediction curve L(t) is applied to a peak-finding algorithm (such as scipy.signal.find_peaks) based on the change of the first derivative and the detection of local maxima. By setting the height, distance and prominence parameters, the main peaks are located.

[0020] For each peak Peak_i identified by the peak-finding algorithm, the following features are extracted: Peak period: the start time T_start_i and end time T_end_i of the peak; specifically, the boundary is determined by finding the point on both sides of the peak where the power value drops to a certain proportion (e.g., 85%) of the peak value, or the critical point where the first derivative of the power curve changes from positive to negative. Crest intensity: The peak load L_peak_i of this peak; In this embodiment of the invention, the output of this stage is a set of all identified peaks {Peak_i}, where each peak Peak_i contains its T_start_i, T_end_i, and L_peak_i information.

[0021] In this embodiment of the invention, in order to achieve a smooth transition from peak to non-peak and to quantify the degree to which the peak is at any time t, step S22 is followed by step S23. Step S23 defines a peak membership function I_peak(t). This functionalization process makes peak judgment no longer a binary decision of 0 / 1, but a continuous measurement, laying the foundation for generating a smoother and more reasonable energy storage strategy.

[0022] S23, based on peak feature extraction, output the set of all identified peaks {Peak_i}. For each identified peak Peak_i, define the peak membership function I_peak(t) within its peak time period [T_start_i, T_end_i]. During the peak core period [T_start_i+Δ, T_end_i-Δ] (where Δ is the preset period offset, such as 15 minutes), the membership value I_peak(t) = 1; During the forward slope transition period [T_start_i, T_start_i+Δ] and the backward slope transition period [T_end_i-Δ, T_end_i], the membership value I_peak(t) smoothly decays from 1 to 0 using a Gaussian or quadratic function. For example, during the forward slope period, a quadratic function is used for decay to provide a smooth curve transition. I_peak(t) = [(t - T_start_i) / Δ]^2.

[0023] For non-peak periods, I_peak(t) = 0.

[0024] If the transition periods of multiple peaks overlap, the maximum value among all overlapping function values ​​is taken as the final I_peak(t) value at that moment.

[0025] In this embodiment of the invention, through the above steps, the load forecast curve L(t), peak feature set {Peak_i}, and continuous peak membership function I_peak(t) within the future period T are finally output, providing a dynamic and refined decision-making basis for the subsequent generation of energy storage strategies.

[0026] S3: Based on the meteorological data and historical power generation data, the prediction interval [P_min(t), P_max(t)] of new energy power generation within the future period T is obtained through the interval prediction model of new energy power generation; Preferably, in this embodiment of the invention, a quantile regression forest (QRF) or convolutional neural network (CNN) algorithm is used to construct an interval prediction model; Interval prediction model training: The model is trained using historical meteorological data (features) and historical power generation data (labels). The model learns the complex nonlinear mapping relationship from meteorological features to power generation.

[0027] Interval forecasting: For any future time t, the interval forecasting model outputs a forecast interval [P_min(t), P_max(t)], which represents the range in which the actual power generation may fall at a specified confidence level (e.g., 90%). P_forecast(t) is the median or expected value of the forecast interval [P_min(t), P_max(t)].

[0028] The interval prediction in this embodiment of the invention contains more information than traditional point prediction. A wider interval indicates high prediction uncertainty (e.g., photovoltaics under cloudy weather), while a narrower interval indicates high prediction confidence (e.g., photovoltaics under sunny weather), which can provide a basis for subsequent assessment of energy shortage risks; S4: Calculate the peak overlap O(t) and energy storage urgency U(t) at each moment within the future period T; Based on the peak overlap O(t), energy storage urgency U(t), and the current state of charge (SOC) of the energy storage system, generate a set of charging and discharging power instructions for the energy storage system within period T according to preset rules; The formula for calculating the peak overlap O(t) is:

[0029] O(t) = (L(t) - L_avg) / (L_peak - L_avg) × I_peak(t); where O(t) represents the peak overlap at time t, reflecting the degree of peak electricity consumption at time t; L(t) is the predicted load at time t; L_avg is the average load within period T; L_peak is the intensity of the most significant peak; and I_peak(t) is the peak membership degree. The calculation formula of said energy storage urgency U(t) is: U(t)= O(t)×(1-R(t)); wherein, U(t) represents the energy storage urgency at time t, which is an indicator reflecting the urgency of requiring the energy storage system to perform discharge compensation at time t; R(t) is the new energy power generation confidence level at time t, R(t)=(P_forecast(t)- P_min(t)) / (P_max(t)-P_min(t)). P_min(t) represents the lower forecast limit of new energy power generation power at time t, and P_max(t) represents the upper forecast limit of new energy power generation power at time t.

[0030] Wherein, the lower R(t) is, the greater the risk that the actual power generation power is close to the lower edge P_min(t) of the prediction interval, that is, the worse the power supply reliability. Therefore, U(t) obtains a high value when the load is high (O(t) is high) and the new energy power generation reliability is low (R(t) is low), and energy storage discharge is required.

[0031] Said preset rules are: If the energy storage urgency U(t)>U_high (high urgency threshold) and SOC(t)>SOC_low (discharge lower limit), then discharge is triggered; the discharge power P_discharge(t) can be in a proportional relationship or a piecewise function relationship with U(t).

[0032] If the energy storage urgency U(t)<U_low (low urgency threshold) and SOC(t)<SOC_high (charging upper limit), and there is a current new energy power generation surplus (that is, P_actual(t)>L(t)), then charging is triggered; the charging power can be obtained from the surplus power; Standby / floating state: if U_low ≤ U(t)≤ U_high, the energy storage system is instructed to stand by or perform small power compensation to smooth the tie-line power; Wherein, U_high is a preset high urgency threshold; U_low is a preset low urgency threshold; SOC_high and SOC_low are the upper and lower safety limits of the state of charge of the energy storage system; SOC(t) represents the current state of charge of the energy storage system, which is a measured value representing the percentage of the remaining battery capacity to the total capacity; S5: Execute the generated charge and discharge power instruction set, monitor the actual power load and new energy power generation power in real time, compare them with their corresponding predicted values respectively, and calculate the prediction deviation separately; if any deviation exceeds the corresponding set threshold, trigger the strategy regeneration process and return to step S2. It includes:

[0033] Continuously monitor the actual power load L_real(t) and the actual new energy power generation power P_real(t); Calculate the absolute deviation of load forecast and the absolute deviation of new energy power generation forecast; ΔL=|L_real(t)-L(t)|, ΔP=|P_real(t)-P_forecast(t)|.

[0034] Where ΔL represents the absolute deviation of the load forecast at time t; L_real(t) represents the actual total power load at time t, which in a specific implementation is usually the real active power value obtained by real-time measurement of a smart meter; L(t) represents the predicted total power load at time t, which is output by the dynamic peak identification model. ΔP represents the absolute deviation of the new energy power generation prediction at time t; P_real(t) represents the actual new energy power generation at time t, which is usually collected in real time from the photovoltaic inverter or wind turbine monitoring system in specific implementations; P_forecast(t) represents the median or expected value of the prediction interval [P_min(t), P_max(t)] output by the interval prediction model (i.e., the predicted value of the new energy power generation at time t). When any deviation (ΔL or ΔP) exceeds its respective preset dynamic threshold (which may be related to the width of the prediction interval), the current prediction is determined to be significantly inaccurate, the execution of the current instruction set is immediately interrupted, and the process returns to step S2. In a specific implementation, the generated charge and discharge power command set is sent to the energy storage converter (PCS) to control the energy storage battery to perform the corresponding charge and discharge actions.

[0035] The embodiments of the present invention can dynamically adapt to sudden situations (such as sudden weather changes or sudden load increases) and always maintain the effectiveness of decision-making.

[0036] The following uses a photovoltaic-storage system in an industrial park as an example to illustrate the implementation process of the method of this invention. An industrial park has a 1 MW rooftop photovoltaic power station, equipped with a 500 kW / 1000 kWh lithium iron phosphate energy storage system. Decisions are made every 15 minutes for the next 24 hours (period T=24).

[0037] Data acquisition and processing: Total active power data is collected in real-time from smart meters in the park's power distribution room at 15-minute intervals. The data range covers the past 30 days to the present. Forecast data of total horizontal irradiance (W / m²) and ambient temperature (°C) for the park's location are obtained from a professional meteorological service interface, every 15 minutes. Historical power generation (past 30 days) and real-time power generation are collected from the photovoltaic power station monitoring system. All data undergoes cleaning, alignment, and normalization.

[0038] Dynamic power consumption peak identification: The historical load data processed over the past 30 days was decomposed using the STL algorithm to obtain the trend term, seasonal term (daily cycle), and residual term.

[0039] The system identified a slight linear upward trend in the park's load due to capacity ramp-up. A first-order polynomial fitting was used to extrapolate and obtain the trend term T_trend(t) for a gradual increase over the next 24 hours. A stable daily cyclical pattern (high during the daytime working hours and low during the night and lunch break) was extracted using Fourier series. This pattern was then replicated to the prediction day to obtain the seasonal term S_seasonal(t). The recent residual sequence was input into a pre-trained GRU network (64 hidden layer units) to predict the random fluctuation residual term R_residual(t) for the next 24 hours.

[0040] By superimposing the three terms, the predicted load curve L(t) is obtained. The prediction results show that there are two main electricity consumption peaks the next day: the morning peak (09:00-11:30) and the afternoon peak (14:00-17:30).

[0041] The scipy.signal.find_peaks function was applied to the predicted load curve L(t), with distance=12 (corresponding to the minimum 3-hour interval) and prominence=0.1 (the minimum peak prominence after normalization), and the two peaks mentioned above were successfully identified.

[0042] Extract the peak feature set {Peak_1, Peak_2}: Peak_1: T_start_1=09:00, T_end_1 =11:30, peak value L_peak_1=850kW.

[0043] Peak_2: T_start_2=14:00, T_end_2=17:30, peak value L_peak_2=900kW.

[0044] Set the offset Δ = 30 minutes, and define membership functions f1(t) and f2(t) for Peak_1 and Peak_2 respectively. During their respective core time periods [09:30, 11:00] and [14:30, 17:00], the function value is 1; during the transition periods of 30 minutes before and after, the function value smoothly decays from 1 to 0 according to a quadratic function.

[0045] Forecast of new energy power generation range: A pre-trained quantile regression forest model is used. The input consists of weather forecast features for the next 24 hours (irradiance, temperature, hour, month, etc.).

[0046] The model outputs the 90% confidence interval [P_min(t), P_max(t)] and median P_forecast(t) for photovoltaic power generation. Taking 15:00 as an example: the forecast is cloudy with high uncertainty. P_forecast(15:00) = 320kW (median forecast); P_min(15:00) = 100kW (5th percentile); P_max(15:00) = 550kW (95th percentile);

[0047] Energy storage strategies are generated dynamically: Taking 15:00 as an example, the average load within period T is L_avg = 600 kW, the intensity of the most significant peak Peak_2 is L_peak = 900 kW, and L(15:00) = 880 kW.

[0048] Crest overlap: O(15:00) = (880 - 600) / (900 - 600) * 1 ≈ 0.93.

[0049] Confidence level of power generation: R(15:00) =(320 - 100) / (550 - 100) ≈ 0.49.

[0050] Energy storage urgency: U(15:00) = 0.93 * (1 - 0.49) ≈ 0.47.

[0051] Command generation (taking 15:00 as an example): Preset parameters U_high = 0.5, U_low = 0.2, SOC_high = 0.9, SOC_low = 0.2. Current energy storage SOC(15:00) = 0.7. Since U(15:00) = 0.47 is slightly lower than U_high but greater than U_low, according to the rules, the energy storage system should be in standby or low-power compensation state. The command generated at 15:00 is: {Mode: "Standby", Target Power: 0 kW}.

[0052] The system calculation found that at 10:15 AM, the predicted value of U(10:15) = 0.65 > U_high was 0.65, and there was a surplus of renewable energy generation at that time. Therefore, in the generated instruction set for the whole day, it was planned to use the photovoltaic surplus to charge the energy storage to a higher SOC during the period from 9:00 AM to 10:00 AM in preparation for the high urgent discharge demand around 10:15 AM.

[0053] Instruction execution and feedback correction: The energy management system sends a set of charging and discharging commands for the next 24 hours, including {15:00: Standby, 0kW}, to the energy storage converter (PCS). The system controls the energy storage system to standby at 15:00. Real-time monitoring of the actual data feedback at 15:00 shows: L_real(15:00) = 890 kW, P_real(15:00) = 120 kW.

[0054] ΔL = |890 - 880| = 10 kW, the deviation rate is 1.1%, which does not exceed the threshold (assumed to be 10%).

[0055] ΔP = |120 ​​- 320| = 200 kW, with a deviation rate of 62.5%, exceeding the preset threshold (assumed to be 30%). This is because the actual cloud cover was thicker, resulting in photovoltaic output far below the predicted median.

[0056] Due to ΔP exceeding the limit, the execution of subsequent original instructions is immediately interrupted, and the strategy regeneration process is triggered. The process returns to the "Dynamic Electricity Peak Identification" step. The dynamic peak identification model takes the actual load at 15:00, L_real(15:00) = 890kW, as the latest historical data point into consideration; the interval prediction model also takes the actual photovoltaic output at 15:00, P_real(15:00) = 120kW, and worse weather conditions as inputs to re-predict the load curve and photovoltaic output range for the future (e.g., from 15:15 to 15:00 the next day). Based on the new and more accurate predictions, the urgency is recalculated and an updated energy storage charging and discharging instruction set is generated, thereby dynamically adapting to sudden weather changes and ensuring the rationality of subsequent time-period scheduling strategies.

[0057] As can be seen from this embodiment, the method of the present invention successfully achieves coordinated management of dynamic load peaks and uncertain photovoltaic power generation, and ensures the adaptability and robustness of the strategy in real environment through a closed-loop feedback mechanism, effectively improving the utilization efficiency of the energy storage system and the stable operation level of the power grid.

[0058] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by program instructions and related hardware. The program can be stored in a computer-readable storage medium, such as ROM, RAM, disk, optical disk, etc.

[0059] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A new energy power generation and energy storage management method based on peak electricity consumption, characterized in that, Includes the following steps: S1 collects real-time electricity load data, meteorological data, and historical and real-time power generation data of related new energy power plants in the target area, and preprocesses the collected data. S2, based on the electricity load data, predict the load curve within the future period T through a dynamic peak identification model, identify and quantify the peak characteristics based on the load curve, and construct the peak membership function for each peak according to the peak characteristics. S3. Based on the meteorological data and historical power generation data, the predicted range of new energy power generation within the future period T is obtained through the interval prediction model of new energy power generation. S4, calculate the peak overlap and energy storage urgency at each moment within the future period T; based on the peak overlap, energy storage urgency and the current state of charge of the energy storage system, generate a set of charging and discharging power instructions for the energy storage system within the period T according to preset rules. S5, execute the generated charging and discharging power instruction set, and monitor the actual power load and new energy power generation in real time, compare them with their corresponding predicted values, and calculate the prediction deviation respectively; if any deviation exceeds the corresponding set threshold, the strategy regeneration process is triggered, and return to step S2. The formula for calculating the overlap of wave crests is: O(t)= (L(t)-L_avg) / (L_peak-L_avg)×I_peak(t); Where O(t) represents the peak overlap at time t; L(t) is the predicted load at time t; L_avg is the average load within period T; L_peak is the intensity of the most significant peak; and I_peak(t) is the peak membership degree. The formula for calculating the urgency of energy storage is as follows: U(t) = O(t) × (1 - R(t)); Where U(t) represents the energy storage urgency at time t; R(t) is the confidence level of new energy power generation at time t, R(t)=(P_forecast(t)- P_min(t)) / (P_max(t)-P_min(t)); P_forecast(t) is the median or expected value of the prediction interval [P_min(t),P_max(t)], P_min(t) represents the lower limit of the predicted new energy power generation at time t, and P_max(t) represents the upper limit of the predicted new energy power generation at time t.

2. The new energy power generation and energy storage management method based on peak electricity consumption as described in claim 1, characterized in that, Step S2 includes: A hybrid forecasting model combining time series decomposition and machine learning is used to predict the load curve for future periods T. Based on the predicted load curve, identify the peaks and quantify their characteristics; The output is a set of all identified peaks based on peak feature extraction. For each identified peak, the peak membership function is defined within its peak time period.

3. The new energy power generation and energy storage management method based on peak electricity consumption as described in claim 2, characterized in that, The hybrid prediction model, which combines time series decomposition and machine learning, is used to predict the load curve for future periods T, including: The STL algorithm is used to decompose historical load data into trend, seasonal and residual terms. The trend term sequence is extrapolated using polynomial fitting to obtain the predicted value of the trend term for the future period T; Fourier series analysis is applied to the seasonal term sequence to extract the fundamental and harmonic components of its inherent daily and weekly fixed cyclic patterns, and the predicted value of the future seasonal term is obtained by directly copying the daily and weekly cyclic patterns to the future period T; the residual term sequence is input into a gated recurrent unit network for training and prediction to obtain the predicted value of the future residual term. By superimposing the predicted components, the load forecast curve for the future period T is obtained; The formula for the load forecast curve is: L(t) = T_trend(t) + S_seasonal(t) + R_residual(t); Where L(t) represents the predicted total electricity load at time t; T_trend(t) represents the predicted value of the load trend term at time t within a future period T; S_seasonal(t) represents the predicted value of the load seasonal term at time t within a future period T; and R_residual(t) represents the predicted value of the load residual term at time t within a future period T.

4. The new energy power generation and energy storage management method based on peak electricity consumption as described in claim 2, characterized in that, The process of identifying peaks and quantifying their characteristics based on the predicted load curve includes: A peak-finding algorithm based on first-order derivative change and local maximum detection is applied to the load prediction curve L(t). By setting height, distance and prominence parameters, the main peaks are located. For each peak Peak_i identified by the peak-finding algorithm, the following features are extracted: Peak period: the start time T_start_i and the end time T_end_i of the peak; Crest intensity: The peak load of the peak L_peak_i.

5. The new energy power generation and energy storage management method based on peak electricity consumption as described in claim 2, characterized in that, The peak membership function I_peak(t) is defined as: During the peak core period [T_start_i+Δ, T_end_i-Δ], the membership value I_peak(t)=1, where Δ is the preset time period offset; During the transition period before the peak [T_start_i,T_start_i+Δ] and the transition period after the peak [T_end_i-Δ,T_end_i], the membership value I_peak(t) smoothly decays from 1 to 0 according to a Gaussian function or a quadratic function. For non-peak periods, I_peak(t) = 0; if the transition periods of multiple peaks overlap, the maximum value among all overlapping function values ​​is taken as the final I_peak(t) value at that moment.

6. The new energy power generation and energy storage management method based on peak electricity consumption as described in claim 1, characterized in that, Based on the meteorological data and historical power generation data, the predicted range of new energy power generation [P_min(t), P_max(t)] within the future period T is obtained through the interval prediction model of new energy power generation. The interval prediction model is constructed using quantile regression forest or convolutional neural network algorithms; The training method for the interval prediction model is as follows: the model is trained using historical meteorological data and historical power generation data, and the model learns the complex nonlinear mapping relationship from meteorological characteristics to power generation.

7. The new energy power generation and energy storage management method based on peak electricity consumption as described in claim 1, characterized in that, The preset rules are as follows: If the energy storage urgency U(t) > U_high and SOC(t) > SOC_low, then discharge is triggered; If the energy storage urgency U(t)<U_low and SOC(t)< SOC_high, and there is current new energy power generation surplus, charging is triggered; If U_low ≤ U(t)≤ U_high, the energy storage system is instructed to stand by or perform small power compensation to smooth the tie-line power; Wherein, U_high is a preset high urgency threshold; U_low is a preset low urgency threshold; SOC_high and SOC_low are the upper and lower safety limits of the state of charge of the energy storage system; SOC(t) represents the state of charge of the energy storage system at the current moment, which is expressed as a percentage of the remaining battery power to the total capacity.

8. The new energy power generation and energy storage management method based on peak electricity consumption as described in claim 1, characterized in that, Said step S5 includes: Continuously monitor the actual power load and actual new energy power generation; Calculate the absolute deviation of load forecast and the absolute deviation of new energy power generation forecast; If any deviation exceeds the corresponding set threshold, the strategy regeneration process is triggered, and the process returns to step S2.

9. The new energy power generation and energy storage management method based on peak electricity consumption as described in claim 8, characterized in that, The calculation of the absolute deviation of load forecast and the absolute deviation of new energy power generation forecast is as follows: ΔL=|L_real(t)-L(t)|, ΔP=|P_real(t)-P_forecast(t)|; Wherein, ΔL represents the absolute deviation of load forecast at time t; L_real(t) represents the actual total power load at time t; L(t) represents the forecast total power load at time t, output by the dynamic peak identification model; ΔP represents the absolute deviation of new energy power generation forecast at time t; P_real(t) represents the actual new energy power generation at time t; P_forecast(t) represents the median or expected value of the forecast interval [P_min(t), P_max(t)] output by the interval forecast model.

Images