Comprehensive energy self-adaptive scheduling method based on meteorological prediction

Abstract

The invention provides a comprehensive energy self-adaptive scheduling method based on meteorological prediction, and aims to solve the problems that an existing energy system lags in response to meteorological disturbance and is insufficient in load regulation precision, and the comprehensive energy self-adaptive scheduling method comprises the following steps: fusing second-level meteorological disturbance data, electrical load power data, energy storage device charge state data and power grid boundary power flow data; constructing a disturbance response feature set, and establishing a time correlation mapping structure between a disturbance variable and a load variable; according to the method, a disturbance trend is predicted based on a mapping structure, an index set of disturbance direction, intensity and sensitive sections is formed, a dynamic regulation capability vector is generated in combination with a current load state, a scheduling optimization model meeting multiple operation constraints is constructed on the basis, a cooperative control strategy set is output, and the dynamic regulation capability vector is obtained. Finally, the control strategy is converted into a control instruction, the control instruction is accurately acted on a power generation device, an energy storage device and a load device, second-level dynamic self-adaptive scheduling of the comprehensive energy system on meteorological disturbance is achieved, and the system response capacity, stability and energy utilization efficiency are improved.

Description

Technical Field

[0001] This invention relates to the field of integrated energy system scheduling and control technology, and in particular to an integrated energy adaptive scheduling method based on meteorological forecasting. Background Technology

[0002] Existing integrated energy system dispatching methods mainly rely on static optimization strategies based on daily or hourly plans. These methods typically generate a unified source-load-storage coordinated control scheme by combining historical load data and medium- to long-term meteorological forecasts with pre-defined rules or linear models. While these methods are feasible in engineering implementation, their control response granularity is relatively coarse, and they generally lack the ability to adapt in real time to high-frequency disturbances or rapidly changing environmental conditions.

[0003] With the large-scale integration of distributed renewable energy, the transient impact of meteorological factors on power generation output and load characteristics has become increasingly significant. Traditional dispatching methods are unable to fully reflect the coupling effect of second-level disturbances such as wind speed and solar irradiance on the system's operating status, resulting in reduced energy utilization, increased system volatility, and severe lag in dispatching decisions, especially lacking specificity and robustness on short time scales.

[0004] To overcome the above problems, it is urgent to propose a comprehensive energy dispatching method that can respond to second-level meteorological disturbance trends and achieve real-time adaptive optimization control. Summary of the Invention

[0005] This application provides a comprehensive energy adaptive scheduling method based on meteorological forecasting to achieve rapid response and dynamic adaptive scheduling to second-level meteorological disturbance trends, thereby improving the stability and energy utilization efficiency of the comprehensive energy system.

[0006] This application provides a comprehensive adaptive energy scheduling method based on meteorological forecasting, including: It collects second-level meteorological disturbance data, including wind speed, solar irradiance, temperature and humidity, within the target area, and simultaneously collects power load data, energy storage device state of charge data and power flow data at the grid boundary. By integrating the second-level meteorological disturbance data, power load data, energy storage device state of charge data, and power grid boundary flow data, a disturbance response feature set is constructed, and a disturbance response mapping structure characterizing the time correlation between disturbance variables and load variables is constructed based on the disturbance response feature set. Based on the disturbance response mapping structure, the changing trend of the disturbance variable within a second-level period is predicted, and a set of disturbance trend indicators including disturbance direction, disturbance intensity and response sensitive segment is generated. By combining the set of disturbance trend indicators with the current power load data, a dynamic adjustment capability vector for power load is generated. The dynamic adjustment capability vector is used to characterize the adjustment elasticity and adjustment range of power load under different disturbance trends. A multi-objective scheduling optimization model is constructed based on the dynamic adjustment capability vector and the state of charge data of the energy storage device, and a set of coordinated control strategies that meet the requirements of grid operation stability, renewable energy output utilization rate and energy storage device safety threshold are output. The set of collaborative control strategies is transformed into a set of control commands, which are then sent to the power generation unit, energy storage unit, and adjustable load unit respectively, to perform second-level dynamic adaptive scheduling control of the integrated energy system.

[0007] The beneficial effects of the technical solution provided in this application include: (1) Achieve second-level prediction and response to meteorological disturbance trends, significantly improve the dispatch system's ability to perceive and respond to rapidly changing factors such as wind speed and solar irradiance, and avoid the problem of decreased system stability caused by delayed dispatch. (2) By constructing a disturbance response mapping structure and a dynamic adjustment capability vector, accurately characterize the elasticity and adjustment potential of the power load to different disturbance trends, and realize refined linkage control between the load side and the source and storage side. (3) Introduce a multi-objective dispatch optimization model, comprehensively consider the grid operation stability, renewable energy utilization rate and energy storage device safety threshold, and improve the overall operating efficiency and economy of the energy system. (4) Adopt a control strategy set distribution mechanism to realize the coordinated control closed loop of the source-load-storage unit, and enhance the robustness and adaptability of the system in high-frequency disturbance scenarios. Attached Figure Description

[0008] Figure 1 This is a flowchart of a comprehensive energy adaptive scheduling method based on weather forecasting provided in the first embodiment of this application. Detailed Implementation

[0009] Many specific details are set forth in the following description to provide a full understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this application; therefore, this application is not limited to the specific embodiments disclosed below.

[0010] The first embodiment of this application provides a comprehensive adaptive energy scheduling method based on weather forecasting. Please refer to... Figure 1 This figure is a schematic diagram of the first embodiment of this application. The following is in conjunction with... Figure 1 The first embodiment of this application provides a detailed description of a comprehensive energy adaptive scheduling method based on meteorological forecasting.

[0011] Step S101: Collect second-level meteorological disturbance data, including wind speed, solar irradiance, temperature and humidity, within the target area, and simultaneously collect power load data, energy storage device state of charge data and power flow data at the grid boundary.

[0012] In the integrated energy adaptive scheduling method based on meteorological forecasting described in this invention, step S101 serves as the input basis for system operation and undertakes the core task of multi-source data acquisition and synchronization. Its implementation process must have a high degree of real-time performance, stability, and data consistency to ensure the effectiveness and reliability of subsequent modeling and scheduling processes.

[0013] Specifically, high-precision meteorological monitoring devices with second-level sampling capabilities need to be deployed in the target area to continuously collect environmental parameters, including at least wind speed, solar irradiance, and air temperature and humidity. Wind speed measurement can be achieved using ultrasonic anemometers, which feature no mechanical parts, high-frequency response, and high linearity, accurately reflecting the changing trends of wind energy resources. Solar irradiance can be collected by photoelectric radiation meters (such as silicon photovoltaic sensors), whose response bandwidth covers the entire spectrum and can adapt to fluctuations in solar intensity under different climatic conditions. Temperature and humidity data can be obtained through digital temperature and humidity modules, which have high integration and short response times, making them suitable for dynamic scenarios.

[0014] Simultaneously, within the same sampling period, it is necessary to acquire power system operation data that is completely synchronized with the aforementioned meteorological data, including three core types of data: power load data, energy storage device state of charge data, and grid boundary power flow data. Specifically, the power load data should originate from high-frequency energy metering units or smart terminals installed at major user sides or substation nodes, capable of outputting active and reactive power information in real time at the second level; the energy storage device state of charge data can be provided through a battery management system (BMS), mainly including SOC (State of Charge), SOH (State of Health), and operating parameters such as charging and discharging current and voltage; grid boundary power flow data is typically acquired by the SCADA system at the substation dispatching end, including power injection values, voltage levels, and power factors at the feeder ends.

[0015] To ensure time consistency among data from different sources, a unified timestamp mechanism (such as GPS clock synchronization) should be adopted to ensure that all data is sampled and transmitted using the same time base, avoiding model errors and control deviations caused by data asynchrony. All types of data need to be preprocessed and standardized in format through edge gateway devices to remove outliers, fill in missing items, and upload to the central scheduling and computing platform according to a unified data structure.

[0016] In addition, to avoid information delays caused by data transmission latency, it is recommended to deploy a data transmission channel with local caching and dynamic packet loss compensation mechanisms, such as using low-latency communication protocols like MQTT or IEC 61850, and setting a data refresh window at the second or sub-second level to meet the system's requirements for rapid response to disturbances.

[0017] In summary, step S101 provides a high-resolution, highly consistent, and representative basic input for subsequent disturbance response modeling and dynamic scheduling optimization by constructing a synchronous multi-source real-time dataset composed of wind speed, solar irradiance, temperature and humidity, power load, energy storage device state of charge, and power flow data at the grid boundary.

[0018] Step S102: Integrate the second-level meteorological disturbance data, power load data, energy storage device state of charge data and power grid boundary flow data to construct a disturbance response feature set, and construct a disturbance response mapping structure based on the disturbance response feature set to characterize the time correlation between disturbance variables and load variables.

[0019] In the integrated energy adaptive scheduling method based on meteorological forecasting described in this invention, the core function of step S102 is to perform structured fusion processing on the multi-source real-time data collected in the previous stage, thereby extracting the coupling characteristics between meteorological disturbances and load response, and constructing a time-correlated mapping structure to support the basic model for subsequent trend prediction and scheduling optimization calculations.

[0020] First, the second-level meteorological disturbance data (including wind speed, solar irradiance, temperature, and humidity) obtained in step S101, along with the power system-related operational data (including power load data, energy storage device state of charge data, and grid boundary power flow data), need to undergo unified preprocessing. This includes data format standardization, sampling period alignment, outlier identification and removal, and missing value imputation. A sliding window mechanism can be used in this stage to package data from different times into time segments to preserve their continuous evolution characteristics while reducing the noise impact of short-term abrupt changes.

[0021] Next, based on the completed data preprocessing, a high-dimensional disturbance response feature set is constructed to clarify the short-time response relationship between meteorological disturbance variables (such as wind speed change rate, irradiance fluctuation amplitude, etc.) and response variables (such as load power change, energy storage charging and discharging rate, power grid flow swing). This feature set can adopt a vector or tensor structure, arranged in the order of sampling time, so that each data point simultaneously possesses spatial distribution and temporal evolution. For example, statistical features can be extracted from the sampled data within a certain minute to obtain indicators such as maximum value, minimum value, mean, standard deviation, and gradient of change, which are used to characterize the local intensity and directionality of the disturbance variable.

[0022] After constructing the disturbance response feature set, high-frequency coupling relationships are further extracted using multidimensional correlation analysis to establish a time-series response mapping structure between disturbance variables and load variables. This mapping structure not only considers the linear correlation coefficients between variables but also introduces indicators such as mutual information, time lag coefficients, and Granger causality to describe the time dependence and directionality between variables. This allows for a clear depiction of how a meteorological disturbance causes the lag path and intensity changes in the response of downstream loads or energy storage equipment.

[0023] To improve the predictability and generalization ability of the mapping structure, a trainable regression framework (such as a lightweight neural network or support vector machine) can be used to model and train the structure, enabling it to adaptively learn the system's response behavior under different meteorological disturbance scenarios. Simultaneously, it should be ensured that the mapping structure is updatable; that is, when new disturbance data is input in subsequent runs, the parameters can be adjusted in a timely manner through an online learning mechanism, allowing the model to maintain high adaptability to the current environment.

[0024] The mapping structure should ultimately output a clear and identifiable disturbance-response path map. Its core function is to provide a quantitative basis for subsequent disturbance trend prediction and response elasticity calculation, ensuring that the trend indicators and adjustment capability results generated in subsequent steps have real physical meaning and high response accuracy.

[0025] To facilitate understanding, the following example illustrates how the disturbance response feature set is constructed. For instance, within a certain area, the system continuously collects second-level changes in wind speed, solar irradiance, and temperature and humidity over 10 seconds. Simultaneously, it records the changes in electrical load power of a building, the charging and discharging current changes of the energy storage system, and the power flow fluctuations at the grid connection point during this time period. The system can divide this data into time slices and extract the rate of change, extreme points, local slopes, variance, and covariance of each variable per second. This forms a multi-dimensional vector set containing multiple pairs of causal relationships, such as "wind speed change rate - load power change" and "solar irradiance fluctuation - energy storage response rate," which serves as sample data for the disturbance response feature set and is used for subsequent modeling.

[0026] Furthermore, when constructing a disturbance-response mapping structure based on this feature set, the system can identify the lag time window and its direction and intensity by analyzing the delay correlation between the same disturbance variable (such as wind speed) and the corresponding response variable (such as load power change) in multiple time slices. For example, the system can determine that a 2 m / s increase in wind speed will typically cause a 3% increase in load in the area after 8 seconds, while a 30 W / m² decrease in solar irradiance will cause the energy storage system to enter discharge mode within 5 seconds. These mapping relationships are ultimately integrated into a graph structure, where nodes represent disturbance and response variables, and edges represent the direction and intensity of causal influence between variables, forming a clear and visual disturbance-response path topology, providing a reliable data foundation for trend prediction and control optimization.

[0027] Furthermore, the process of fusing the second-level meteorological disturbance data, power load data, energy storage device state-of-charge data, and grid boundary power flow data to construct a disturbance response feature set, and then constructing a disturbance response mapping structure based on the disturbance response feature set to characterize the temporal correlation between disturbance variables and load variables, includes: A three-dimensional sliding tensor containing time index, disturbance variable index, and load variable index is constructed using continuous second-level time windows. The sliding tensor is used to extract the joint evolution characteristics between each disturbance variable and each load variable within each time window, forming a set of disturbance response features including disturbance change rate, disturbance standard deviation, load response gradient, and power change consistency factor. When constructing the disturbance response mapping structure that characterizes the temporal correlation between the disturbance variable and the load variable, a causal response path between the disturbance variable and the load variable is constructed based on the disturbance lifetime range of each disturbance variable in the disturbance response feature set, using a causal determination method based on time delay cross-correlation, and the disturbance lifetime is used as the constraint boundary. Based on the causal response path, a dynamically updatable bidirectional weighted causal graph structure is constructed between each disturbance variable and the load variable. The weight of each side in the bidirectional weighted causal graph structure is jointly calculated based on the disturbance intensity index, the disturbance stability factor and the standardized length of the disturbance lifetime. An adaptive signal-to-noise ratio adjustment mechanism is introduced to compress and correct the causal edge weights in the low-confidence data window. The bidirectional weighted causal graph structure is used as the disturbance response mapping structure.

[0028] In this invention, to achieve high-resolution and high-sensitivity dynamic modeling of the relationship between meteorological disturbances and load response, the collected second-level multi-source data must first undergo structured reconstruction. The collected data includes meteorological disturbance variables such as wind speed, solar irradiance, temperature, and humidity; power load data (usually collected by terminal or circuit); energy storage device state of charge data (such as SOC, SOH, and charge / discharge rate); and power flow data at the grid boundary (such as feeder power input / output and voltage levels). Before entering the subsequent analysis process, the above data must be standardized, cleaned, interpolated, and time-aligned through a unified data interface to ensure that all variables are aligned on the same time reference, thus ensuring the temporal consistency of the joint analysis.

[0029] After data preparation, the system time is divided into continuous, non-overlapping, or partially overlapping time windows on the order of seconds. For example, each time window is defined as 5 seconds long, sliding once every 1 second to form a sliding window sequence with a time step of 1 second. Each time window contains multiple sampling points, and statistical analysis is performed on the one-to-one correspondence between the disturbance variable and the load variable within each time window, forming a three-dimensional tensor. The three dimensions of this tensor correspond to: time window index, disturbance variable index, and load variable index, respectively.

[0030] For each time window, a set of joint evolutionary characteristic indices is calculated for each combination of disturbance variable and load variable, forming a disturbance response characteristic set. Specifically, for disturbance variables, their average rate of change within the current time window is extracted (e.g., if wind speed changes from 3.2 m / s to 4.1 m / s in 5 seconds, the rate of change is 0.18 m / s²), standard deviation (reflecting fluctuation intensity), maximum and minimum values ​​(characterizing instantaneous extreme response), etc.; for load variables, their response slope (first derivative of power change), instantaneous fluctuation amplitude (difference between local maximum and minimum values), power change entropy (measuring the dispersion of power change), etc. are extracted; simultaneously, co-evolutionary indices between disturbance variables and load variables, such as Pearson correlation coefficient and mutual information, are also calculated for initial screening in subsequent causal analysis.

[0031] These multiple sets of statistical features are standardized and embedded into the corresponding positions of the tensor. Since each time window can be processed independently, this structure is suitable for parallel construction in edge nodes or multi-threaded environments, improving model computation efficiency and meeting the requirements of second-level response.

[0032] Next, to further extract the temporal causal relationship between the disturbance variable and the load variable, it is necessary to identify the disturbance lifetime of the disturbance variable. The disturbance lifetime refers to the period during which the disturbance variable undergoes a sustained and significant change within multiple consecutive time windows. For example, if the wind speed rapidly increases from 3.5 m / s to 5.6 m / s and continues to fluctuate above a set threshold (e.g., 0.5 m / s) for more than 8 seconds, this wind speed change can be determined to constitute a disturbance lifetime. The start and end times of the lifetime can be identified by comprehensively considering multiple factors, such as the rate of change of the disturbance variable and the duration of exceeding the standard deviation limit.

[0033] After identifying the lifecycle, a causal relationship analysis boundary is constructed based on this lifecycle to restrict the disturbance variable to only affect specific load variables within its lifecycle, avoiding the introduction of irrelevant disturbances into the modeling. Within this lifecycle, a causality judgment method based on time-delayed cross-correlation is adopted. This involves performing a delayed sliding motion on the sampling sequences of the disturbance and load variables, calculating the cross-correlation coefficients at different time lags, and selecting the lag with the highest correlation exceeding the significance threshold as the causal lag. It is important to exclude spurious correlations here, so a Granger causality test is used for secondary confirmation. Response paths are established for all disturbance-load pairs, and their causal delays, response strengths, and stability indices are recorded.

[0034] Based on this, a bidirectional weighted causal graph structure is constructed. The nodes of this graph structure consist of perturbation variables and load variables, and the edges represent causal paths. The weight of each edge is designed as a weighted product of three indicators: the perturbation intensity index (i.e., the normalized value of the rate of change of the perturbation variable per unit time), the perturbation stability factor (e.g., the inverse standard deviation ratio), and the standardized length of the perturbation lifetime. The formula for calculating the weight is as follows: ; in: Represents the disturbance variable With load variables The boundary weight between; The disturbance intensity index; For disturbance variables Standard deviation over the life cycle; The lifetime length of the disturbance; This is the maximum value over the lifetime of all disturbances; , , These are normalized weighting coefficients, which can be dynamically set according to the system scheduling objectives. Recommended values ​​are 0.5, 0.3, and 0.2.

[0035] The disturbance intensity index represents the disturbance variable. The average rate of change over the disturbance's lifespan reflects the intensity of the disturbance. The stronger the disturbance, the greater its impact on load fluctuations.

[0036] By observing the disturbance lifetime Within this range, the maximum change in the value of the disturbance variable is calculated and divided by the lifetime length to obtain the result. ; in, Represents the disturbance variable In time The numerical value at any given time; : Disturbance lifetime (in seconds); the unit is the original disturbance variable unit per second, for example, wind speed is . .

[0037] Disturbance variables Standard deviation over the life cycle This represents the fluctuation intensity of a disturbance variable over its disturbance lifetime. A larger standard deviation indicates greater instability and lower reliability of the disturbance. It can be calculated using the following formula: ; in, Represents the disturbance variable The mean over the lifespan; This represents the number of sampling points within the lifecycle.

[0038] Disturbance lifetime length This represents the continuous duration from the moment a disturbance variable first exceeds a set disturbance intensity threshold until the disturbance intensity returns to below the threshold. A longer lifetime indicates a more stable impact from the disturbance. For example, defining a disturbance intensity threshold (such as the rate of change of wind speed). Record the moment when the limit is exceeded. and the moment of returning to normal The lifecycle length is .

[0039] To enhance the resilience of the causal structure to anomalous disturbances and low-confidence data windows, an adaptive signal-to-noise ratio (SNR) adjustment mechanism is introduced. This mechanism calculates the SNR for the disturbance data in each time window, specifically by analyzing the ratio of the signal rate of change to the fluctuation intensity (standard deviation). If the SNR is below a threshold, it indicates that the disturbance signal within that window may lack clear directionality or be affected by environmental noise. In this case, the corresponding edge weights in the graph are multiplied by a suppression coefficient (e.g., 0.5) to compress their impact on subsequent prediction structures and improve the robustness of the overall prediction results.

[0040] The resulting bidirectional weighted causal graph structure, as an implementation of the disturbance response mapping structure, has the following advantages: First, it can dynamically adapt to the time lag characteristics of different disturbance scenarios; second, the edge weight update mechanism can reflect the internal structural changes of the disturbance life cycle; third, the graph structure can be used as input for subsequent disturbance trend prediction models to directly call and predict the evolution of disturbance direction, intensity, and impact range within the framework of graph neural networks or graph propagation algorithms.

[0041] The output of the aforementioned disturbance response mapping structure will be used downstream to generate a set of disturbance trend indicators, including disturbance direction (positive or negative), disturbance intensity level (low, medium, high), disturbance acceleration, disturbance duration, and affected response-sensitive sections. These structured indicators will be further combined with current power load data to generate a dynamic regulation capability vector, serving as an important input to the scheduling optimization model. Thus, through tensor modeling, lifecycle identification, delayed causal analysis, and weighted graph structure construction, this technology achieves fine-grained, highly robust, and highly controllable modeling capabilities for the relationship between disturbances and load response, providing a reliable data support foundation for second-level intelligent scheduling of integrated energy systems.

[0042] The entire process not only avoids the one-sidedness of traditional static correlation analysis, but also overcomes the problem of insufficient adaptability of existing linear models to dynamic disturbances. It enables disturbance perception, response mapping and control execution to form a closed-loop and real-time evolving model chain, which has good system engineering deployability and edge control scalability.

[0043] Step S103: Based on the disturbance response mapping structure, predict the change trend of the disturbance variable within a second-level period, and generate a set of disturbance trend indicators including disturbance direction, disturbance intensity and response sensitive segment.

[0044] In the integrated energy adaptive scheduling method based on meteorological forecasting described in this invention, the core objective of step S103 is to use the established disturbance response mapping structure to predict the changing trend of disturbance variables on a second-level time scale, and generate a set of quantifiable disturbance trend indicators accordingly, so as to provide a high-resolution forward-looking judgment basis for the scheduling system.

[0045] This step first extracts highly sensitive feature dimensions representing the dominant path of the disturbance based on the established temporal relationships between variables in the disturbance response mapping structure. This process typically relies on time series modeling methods to capture the evolution patterns of disturbance variables (such as wind speed, solar irradiance, etc.) over multiple consecutive sampling periods. Optional prediction methods include, but are not limited to, sliding window prediction, multi-scale autoregressive models, exponentially weighted moving average models, or lightweight predictors constructed using Long Short-Term Memory (LSTM) networks. Importantly, in the implementation of this invention, the prediction model should be customized according to parameters such as lag time and dominant factor weights identified in the disturbance response mapping structure, rather than using a fixed template or static weights, to ensure adaptability to specific application scenarios.

[0046] Through the above prediction process, trend prediction results for each disturbance variable can be generated within a few seconds in the future. The prediction result for each disturbance variable should include the disturbance direction (e.g., wind speed increase or decrease), disturbance intensity (e.g., the magnitude of change per unit time), disturbance acceleration (rate of change of the trend), and disturbance stability (e.g., coefficient of variation or trend consistency score). In practical applications, the system will comprehensively identify the most sensitive response channel in the system based on the changing trends of multiple disturbance variables and their influence weights in the disturbance response mapping structure, and extract the response-sensitive segment corresponding to that channel, usually represented by key equipment, specific load nodes, or specific time periods.

[0047] The generated set of disturbance trend indicators should be output in a structured data format, including the name of the disturbance variable, the direction of the disturbance (positive / negative), the level of disturbance intensity (e.g., divided into low, medium, and high intervals), the estimated duration of the disturbance, and the number and attribute identifier of the affected response-sensitive segments. This set should be generated with real-time update capability; that is, after each sampling period of data fusion and mapping calculation, the prediction results should be dynamically updated based on the latest data, enabling the system to have high-frequency forward-looking perception capabilities.

[0048] Furthermore, to enhance the reliability of the disturbance trend indicators, this invention combines the historical error distribution of disturbances with the model residual evaluation mechanism to assign a confidence level label to each prediction result. The confidence level label is used to determine the reliability of the trend during subsequent scheduling processes, deciding whether to introduce more redundant control or buffer response mechanisms.

[0049] In summary, step S103 not only completes the continuous prediction from the current disturbance state to the future disturbance trend, but also establishes a set of disturbance trend indicators through three dimensions: disturbance direction, disturbance intensity, and response sensitive section. This enables the subsequent scheduling model to make more targeted, responsive, and robust dynamic control strategies, and is a key bridging link to achieve adaptive scheduling capability.

[0050] To facilitate understanding of the specific implementation process of step S103, the following example illustrates its typical workflow. In a regional integrated energy system, the system continuously collected data over the past 30 seconds, showing a gradual increase in wind speed from 2.8 meters per second to 4.1 meters per second. Simultaneously, through a disturbance response mapping structure, it identified a strong correlation between this wind speed change and the increase in local load power, with a 6-second lag. The system uses a sliding window combined with an LSTM model to predict the current wind speed trend, determining that the wind speed will continue to increase at a rate of 0.15 meters per second within the next 10 seconds. This trend is labeled as a "positive disturbance," with a disturbance intensity level of "medium," and the disturbance duration is expected to be 12 seconds. Since the wind power output in the area corresponding to the increased wind speed will increase, and the load response lag in this area is high, the system further identifies a response-sensitive section that may generate transient voltage fluctuations, including load nodes numbered #3 and #5. Ultimately, the set of disturbance trend indicators generated by the system will include: "Disturbance variable: wind speed", "Direction: positive", "Intensity: medium", "Response sensitive sections: #3, #5", and "Confidence level: 0.91". This result will be used to optimize the control strategy during this cycle and will be passed on to subsequent scheduling stages.

[0051] Step S104: Combine the set of disturbance trend indicators with the current power load data to generate a dynamic adjustment capability vector for power load. The dynamic adjustment capability vector is used to characterize the adjustment elasticity and adjustment range of power load under different disturbance trends.

[0052] In the integrated energy adaptive dispatching method described in this invention, step S104 inherits the disturbance trend index set from the previous stage and combines it with the current power load data. The aim is to construct a dynamic adjustment capability vector that accurately reflects the real-time adjustment capability of various load units under different disturbance trends. The generation of this vector requires not only considering the direction and intensity of the disturbance trend itself, but also combining it with specific load characteristics to quantify its elasticity level and adjustable range at a second-level response scale.

[0053] First, the current operating status of each electrical load node should be extracted, including the current instantaneous power, active and reactive power components, load type (such as rigid load, flexible load, and movable load), and the configuration of the control devices connected to it (such as load switches, frequency converters, smart sockets, etc.). This status data needs to be matched one by one with the disturbance direction, disturbance intensity, and response-sensitive sections in the disturbance trend indicators to identify which load units may be directly affected under the expected disturbance path, and whether these effects are controllable.

[0054] Subsequently, the regulation response capability of each load unit needs to be evaluated. In practical implementation, the regulation model can be divided according to the type of load. For example, for flexible loads such as electric motors, power regulation can be achieved by adjusting the frequency, voltage, or duty cycle; while for loads with hysteresis characteristics such as HVAC, their regulation capability is constrained by inertia, and it is necessary to calculate the remaining release potential of their cooling or heating loads; for loads involving energy storage control, the coupling relationship with the energy storage charging and discharging strategy also needs to be considered. All these factors will form a set of characteristic parameters to describe the instantaneous regulation capability of the current load unit, including the maximum adjustable amplitude, response delay time, and upper limit of regulation rate.

[0055] To structure and quantify the aforementioned feature parameters into a numerical expression that can be input into the scheduling model, each load node should be constructed as a regulation capacity vector. Each component of the vector corresponds to a regulation capacity evaluation index in a different dimension. For example, the first dimension represents the maximum positive scalability tolerance of the load power, the second dimension represents the maximum negative scalability tolerance of the load power, the third dimension represents the expected response time, the fourth dimension represents the upper limit of the regulation duration, the fifth dimension represents the confidence level, and so on. This vector can be categorized and encoded according to load type, importance level, or geographical location, facilitating subsequent on-demand use by the scheduling model.

[0056] In specific scenarios, if certain loads lack immediate adjustment capabilities (such as life support loads or constant power loads), their corresponding adjustment capability vectors should be zero or marked as unadjustable to avoid misjudgment leading to control strategy failure. Furthermore, when multiple disturbance trends coexist, the system should weightedly fuse the adjustment capabilities under different trends based on disturbance intensity weights, outputting a globally optimized load adjustment capability vector set.

[0057] The resulting dynamic adjustment capability vector not only quantifies the current load state's response to future disturbance trends but also serves as an indispensable constraint input and objective variable in the subsequent multi-objective optimization model. The accurate construction of this vector directly determines whether the scheduling strategy can achieve coordinated control of the source-load-storage system on a second-level scale, acting as a crucial bridge between disturbance perception and execution decision-making.

[0058] To facilitate understanding of the specific workflow of step S104, an example is provided below. In a commercial building, the system identifies that the solar irradiance will rapidly decrease within the next 10 seconds, constituting a high-intensity negative disturbance. Furthermore, the building falls within one of the response-sensitive segments defined in the disturbance response mapping structure. The system then reads the current power data and load type information of various electrical devices within the building, including the central air conditioning system, elevators, lighting system, and corridor fans. Calculations reveal that the central air conditioning system is currently operating at 75% load, with a 5kW load reduction adjustment capability, a response time of 3 seconds, a minimum duration of 10 seconds, and a confidence level of 0.92. While the lighting system is also adjustable, it is a rigid lighting load, supporting only a 2kW load reduction, with a response delay of 6 seconds and a confidence level of 0.78. The elevator system, being in operation, is marked as a non-adjustable load, with its adjustment capability vector being zero.

[0059] Ultimately, the building's dynamic adjustment capacity vector can be represented as a five-dimensional vector group: (+0kW, -5kW, 3s, 10s, 0.92) (central air conditioning), (+0kW, -2kW, 6s, 8s, 0.78) (lighting), and (0kW, 0kW, -, -, 0.00) (elevators). These vectors will be fed into the scheduling optimization model to calculate the building's total downward adjustment response capacity, time constraints, and risk level under the current disturbance trend.

[0060] In the example above, each load unit generates a five-dimensional vector to quantify its instantaneous adjustability under specific disturbance trends. Each dimension of this vector has a clear physical meaning and scheduling constraint significance, explained as follows: The first dimension represents the maximum positive upscaling power tolerance (unit: kW), that is, how much more power the load can increase without affecting its functional safety or user comfort under the current operating state. For example, for some motor-type equipment that is not operating at full load, its power consumption can be increased in a short period of time by increasing the frequency or voltage.

[0061] The second dimension represents the maximum negative power reduction tolerance (unit: kW), which is the maximum power reduction the load can achieve under the current conditions. For example, an air conditioner can temporarily reduce its output power, and lighting can turn off some lights to achieve peak shaving response. The third dimension represents the shortest response time (unit: seconds), which is the shortest time delay between the issuance of the control command and the actual power adjustment made by the load. The shorter the response time, the better the immediacy of the load adjustment, making it suitable for second-level disturbance control.

[0062] The fourth dimension represents the maximum sustained response time (in seconds), which is the longest time the load can continuously maintain its regulated state without affecting its main functions. For example, lighting can reduce brightness for 10 seconds, and air conditioning can reduce load for 20 seconds; after these times, normal operation must be restored. The fifth dimension represents the regulation confidence level (a decimal between 0 and 1), used to quantify the reliability of the load's regulation capability. It is obtained by combining historical regulation behavior, equipment stability, and communication status. For example, equipment with frequent control failures or inconsistent regulation effects has a low confidence level, and the scheduling system will prioritize allocating load resources with higher confidence levels.

[0063] This five-dimensional vector comprehensively reflects the load node's adjustment directionality, intensity, speed, maintenance capability, and execution reliability in the face of disturbance trends. It is a key parameter for power balance constraints, response target matching, and resource prioritization in subsequent scheduling optimization models.

[0064] Furthermore, the step of combining the set of disturbance trend indicators with the current power load data to generate a dynamic adjustment capability vector oriented towards power load includes: Based on the disturbance direction, disturbance intensity and response sensitive section contained in the disturbance trend index set, the current power value, maximum adjustable power, maximum adjustable power, minimum response delay, maximum adjustment duration and whether the current state is within the adjustable range of each load node are extracted from the current power load data to form an initial adjustment parameter set. The initial adjustment parameter set is mapped to the disturbance trend index set. Based on the consistency between the disturbance direction and the adjustable direction of each load node, it is determined whether the load node can participate in the current adjustment. It is also determined whether the load node belongs to the response sensitive segment included in the disturbance trend index set. Load nodes that do not meet the above two conditions are removed to obtain the set of effective disturbance response nodes. For each load node in the set of effective nodes for disturbance response, its current response lag time is calculated, which is the expected minimum start time required to execute the load response command. This time is then compared with the predicted start time of the disturbance trend in the set of disturbance trend indicators to obtain the response delay value. Based on this, a penalty factor is introduced to proportionally reduce its maximum adjustable power value. The greater the response delay, the lower the penalty factor, which is used to reduce the calculation weight of the adjustment amplitude. Combine the current operating cycle information of each load node, extract the recovery time it has experienced after the most recent adjustment behavior, and determine whether it can participate in adjustment again within the current disturbance window based on the recovery time. If the recovery time is insufficient, reduce its adjustment amplitude estimate and mark the adjustment status of the node as limited. Calculate the response accuracy of each load node in the past disturbances, including its adjustment magnitude and target deviation under the past disturbance commands, adjustment time error and execution completion rate, and give a response reliability score accordingly. The higher the response reliability score, the more suitable the load node is to participate in the adjustment allocation under the current disturbance trend. The five parameters of each effective disturbance response node in the current scheduling cycle—maximum scalable power, maximum scalable power, minimum response delay, maximum adjustment duration, and response reliability score—are combined into a structured description unit, summarized by load node number, to form a dynamic adjustment capability vector oriented towards electricity load.

[0065] To achieve dynamic response capability assessment of various electricity loads in an integrated energy system under different disturbance trends, a dynamic adjustment capability vector oriented towards electricity loads needs to be established to support load priority ranking and adjustment amplitude allocation in subsequent adaptive scheduling processes. This vector, based on disturbance trend prediction results and combined with the real-time operating status and historical adjustment behavior of each load node, needs to accurately characterize the upper and lower limits of its adjustable capability, the timeliness of its adjustment response, and the reliability of its adjustment within the current cycle. The following will elaborate on the construction method of the dynamic adjustment capability vector from six aspects: data extraction, effective node identification, response delay correction, recovery cycle assessment, historical accuracy scoring, and vector construction.

[0066] First, two prerequisite inputs are required before generating the dynamic adjustment capability vector: a set of disturbance trend indicators and current electricity load power data. The disturbance trend indicator set is calculated based on the disturbance response mapping structure and trend prediction algorithm from the previous stage. It is typically represented in structured data format, including the disturbance direction (e.g., increased wind speed, decreased irradiance), disturbance intensity (usually expressed as a change per unit time, such as m / s² or W / m² / s), the start time of the disturbance trend, the estimated duration, and the response-sensitive section number corresponding to the affected area. The current electricity load power data includes parameters such as the real-time power value of each adjustable load node, set upper and lower power limits, whether it is under control, the allowable adjustment duration, and the control response delay time. This data is uploaded in real-time by the distribution automation system, smart circuit breakers, EMS (Energy Management System), or smart electricity terminals.

[0067] After obtaining these two input data, the first step is to extract the initial set of adjustment parameters for each load node. For each load node, its real-time power value is read from the current electrical load power data. Maximum allowable power increase value Maximum allowable power reduction value Minimum control response delay time Maximum continuous adjustment duration The system reads the node's control status bits to determine if the node is currently in a controllable state (e.g., whether it is locked, whether it has entered protection mode, etc.). These six data items are combined into a parameter group, which serves as the initial regulation capability description unit for the node. Nodes lacking regulation capability or with non-compliant status can be directly marked as non-participating nodes and excluded from subsequent steps.

[0068] Subsequently, the initial set of adjustment parameters for each load node is mapped one-to-one with the set of disturbance trend indicators. First, it is determined whether the adjustment direction of the load node matches the disturbance direction. For example, if the disturbance direction is positive (the load needs to increase to absorb excess energy), only load nodes currently in a low-load state with room for adjustment are retained. Second, it is checked whether the load node physically belongs to the response-sensitive section indicated by the current disturbance trend indicator set. The division of response-sensitive sections can be defined based on the power grid topology, power supply path, feeder number, or the domain to which the distribution transformer belongs. If a load node is not physically located in a sensitive section, its response capability is considered insufficient or it affects the system's adjustment effect, and it is removed. After filtering based on the above two conditions, a set of effective disturbance response nodes that meet the current adjustment requirements is obtained, providing a basis for subsequent adjustment capability calculations.

[0069] For each load node in the set of effective nodes in the disturbance response, it is necessary to calculate the timeliness of its regulation response. First, the currently set minimum response delay for that node is read. Then, extract the predicted start time of the current disturbance trend from the disturbance trend indicator set. And calculate the difference between the two, that is, the response delay value. .in, This represents the system real-time time of the current scheduling cycle, i.e., the current moment when the adjustment capability assessment operation is performed.

[0070] If the response delay value is negative, it indicates that the load's response speed is sufficient to complete the action before the disturbance occurs; if it is positive, it indicates that there is a risk of response lag. Based on this value, a response delay penalty factor is introduced. It can be set to linear or exponential decay form, for example: ; in For empirical adjustment, an initial setting of 0.3–0.5 is recommended. This penalty factor is then used to reduce the maximum adjustable power value. ; ; This is used to reduce the weight of delayed response nodes in the scheduling order, preventing untimely responses from causing deviations from the adjustment target.

[0071] In the integrated energy adaptive scheduling method based on meteorological forecasting described in this invention, variables and These represent the maximum adjustable power and the maximum adjustable power after response delay correction for each load node under a specific disturbance trend prediction scenario.

[0072] These two quantities are the original maximum regulatory capacity (i.e. and The results after timeliness correction are mainly used as the basis for evaluating and allocating the executable adjustment capabilities in the scheduling model.

[0073] Although the dispatching system can identify a load node that theoretically possesses a regulation capacity of 2 kW or 3 kW, if that node requires 30 seconds of preparation time before responding to a dispatch command, and the current disturbance is expected to arrive in 20 seconds, then this regulation capacity will be difficult to take effect within the current cycle. Therefore, the original... and It is insufficient to be used directly as a scheduling input; it must be converted in conjunction with the actual response time to obtain the capability of being "adjustable and timely".

[0074] After completing the delay correction, it is also necessary to evaluate the recovery time of the load node after its most recent adjustment action. This recovery time can be determined by recording the time point when it most recently completed the adjustment command. , with the current moment Time difference between Received. If the recovery time is less than the minimum recovery cycle set by the device. If the node is in a fatigued or limited state, it is not advisable to adjust it again. In this case, its maximum adjustable power value should be further reduced proportionally, or the adjustment state of the node should be directly marked as "limited" and the state should be encoded in the dynamic adjustment capability vector (e.g., 0 indicates completely unadjustable, 1 indicates fully adjustable, and 0.5 indicates limited adjustable).

[0075] In addition to analyzing the current cycle characteristics, the response accuracy of the load node in historical disturbances should also be considered to construct a regulation reliability assessment system. The following indicators can be extracted from historical scheduling data regarding its regulation performance over multiple past disturbances: the average deviation rate between its regulation target value and the actual achieved value (indicating its regulation magnitude achievement rate), the average time error from instruction issuance to regulation effectiveness (indicating execution delay capability), and the regulation instruction success rate (indicating execution reliability). These indicators are respectively... , , This indicates that a comprehensive response credibility score can be constructed based on these parameters. For example, using a weighted summation model: ; in , , For the weighting coefficients, satisfying The value can be set according to the system goals, with initial recommendations of 0.4, 0.3, and 0.3.

[0076] Finally, the five key pieces of information for each valid node in the disturbance response are summarized: maximum scalable power (corrected for response delay), maximum scalable power (corrected for response delay), minimum response delay (original value), maximum regulation duration (original value), and response reliability score (calculated based on historical performance). These are combined into a five-dimensional structured data unit. All valid nodes' data units are then assembled into a set of structured vector data according to their number or positional order, representing a dynamic regulation capability vector oriented towards the power load.

[0077] This vector will ultimately serve as input to subsequent multi-objective scheduling optimization models, supporting key decision-making processes such as load regulation prioritization, regulation intensity configuration, and resource allocation within time windows. The construction of the dynamic regulation capability vector uses high-frequency disturbance trend prediction results as input, combined with real-time power data and historical behavior data, possessing responsiveness, interpretability, and practical control adaptability, ensuring its stable and reliable engineering application value under complex power grid disturbance conditions.

[0078] Step S105: Construct a multi-objective scheduling optimization model based on the dynamic adjustment capability vector and the state of charge data of the energy storage device, and output a set of collaborative control strategies that meet the requirements of grid operation stability, renewable energy output utilization rate and energy storage device safety threshold.

[0079] In the integrated energy adaptive scheduling method based on meteorological forecasting described in this invention, step S105 is the core link to realize multi-objective linkage control of the energy system. Its main task is to construct a scheduling optimization model that adapts to the current disturbance trend based on the dynamic adjustment capability vector and the state of charge data of the energy storage device, thereby outputting a set of collaborative control strategies that meet multiple operating constraints.

[0080] The key to this step is first to clarify the objective function and constraints that the scheduling optimization model should consider. The objective function must simultaneously take into account the safety, economy, and energy sustainability of grid operation, and specifically includes, but is not limited to, the following three typical optimization objectives: First, grid operation stability, which can usually be quantified by frequency fluctuation, node voltage deviation, or power balance deviation; second, the utilization rate of renewable energy output, which requires absorbing as much available electricity as possible from local intermittent energy sources such as wind power and photovoltaics, and avoiding wind and solar curtailment; and third, the safe operating boundary of energy storage devices, which requires controlling their state of charge (SOC) within the set upper and lower limits to avoid overcharging or over-discharging and extend the cycle life of the energy storage system.

[0081] The input data for the scheduling model mainly consists of two parts: first, the dynamic adjustment capability vector generated in step S104, which represents the elastic adjustment capability of each power load unit under different disturbance trends. This vector is a multi-dimensional vector group containing indicators such as adjustable upper and lower limits of load, response time, and adjustment rate; second, the state of charge (SOC) data from the energy storage system, including technical parameters such as current SOC value, maximum charging and discharging power, and remaining capacity. These inputs together constitute the decision variables and constraint parameters of the scheduling model.

[0082] In terms of modeling methods, it is recommended to use a multi-objective constrained optimization framework to solve the scheduling problem. For example, a multi-objective function based on weighted linear combination or ε-constraint method can be constructed, with adjustable weight coefficients set among different objectives to achieve a flexible balance of operating strategies. Considering the uncertainty of the disturbance trend prediction results, robust optimization or fuzzy control strategies can be further introduced into the model, setting confidence intervals or elastic intervals for key variables, thereby improving the adaptability of the strategy under actual disturbances.

[0083] The scheduling model solution process should have high real-time performance; therefore, optimization algorithms with moderate computational overhead and fast convergence speed should be selected, such as improved particle swarm optimization, fast iterative quadratic programming (QP), or mixed integer programming methods accelerated by graph structures. In actual deployment, the solution cycle of the model should be controlled within seconds or less to ensure that the output control strategy has sufficient responsiveness.

[0084] The final set of coordinated control strategies should cover all controlled objects in the integrated energy system, including but not limited to the source side (such as the output setpoints of wind and solar inverters), the load side (such as start / stop or power adjustment commands for adjustable loads like air conditioning, motors, and lighting), and the storage side (such as the charging and discharging power setpoints of battery packs). The strategy set should be labeled in a structured format with the strategy's duration, objective, expected response strength, and corresponding priority. The strategy set must not only be optimal within the current scheduling cycle but also scalable to support a smooth transition to the next cycle and avoid frequent fluctuations.

[0085] In summary, step S105 establishes a rigorous multi-objective scheduling optimization model, combines actual controllability with system operating status, and outputs a set of executable, implementable, and adaptable control strategies, providing key support for achieving second-level dynamic collaborative control of integrated energy systems.

[0086] To further facilitate understanding of the actual operation process of step S105, the following examples illustrate how to construct a multi-objective scheduling optimization model and how to generate a set of cooperative control strategies. For instance, in a microgrid scenario, the system knows that the current wind speed will continue to rise in the next 15 seconds, predicting that the wind power output will increase from the current 80kW to 105kW. Simultaneously, the solar irradiance in the area decreases, resulting in a slight decrease in photovoltaic output. The energy storage device currently has a SOC of 78%, a maximum allowable discharge power of 50kW, and a maximum charging power of 40kW. Some flexible loads on the user side (such as central air conditioning and lighting systems) have a load reduction capability of 10kW generated by the dynamic adjustment capability vector in step S104, with response times of 2 seconds and 5 seconds respectively, an adjustment duration of no less than 15 seconds, and confidence levels of 0.92 and 0.85 respectively.

[0087] Against this backdrop, the system first constructs a scheduling optimization model with three objectives: the first objective is to minimize the power imbalance rate, i.e., to maximize the absorption of new wind power output rather than curtailment; the second objective is to limit the energy storage SOC to no more than 85% to prevent overcharging of wind power and resulting in battery life degradation; and the third objective is to minimize the intensity of load intervention, thereby improving user comfort and system economy. In the model, the dynamic adjustment capability vector provides the upper and lower limits of the load adjustable range, and the energy storage state provides the operational boundaries for charging and discharging.

[0088] Specifically, the first item: power balance error penalty (predicting balance target): ; : Renewable power generation (such as wind power, photovoltaic power) in second t; : Output power of the energy storage system at second t (negative for charging, positive for discharging); : Total system load power at second t; The total number of seconds contained in the scheduling cycle; Sublinear or quasi-quadratic power-enhanced weights (unlike absolute values, they enhance the penalty for large deviations), this term introduces the power exponent. Unlike conventional L1 or L2 norms, it improves the response sensitivity to large deviations and is used to emphasize second-level dynamic power balance.

[0089] Second item: Penalty for energy storage state of charge deviation (safety target): ; : No. The state of charge of an energy storage unit at time t. : No. The target reference SOC value for each energy storage unit; : respectively represent the first The maximum and minimum SOC values ​​allowed for each energy storage unit; : Total number of energy storage units participating in the scheduling; This item uses normalized squared deviation to measure SOC stability, avoids the concentrated use of a single battery under multi-energy storage collaborative scheduling, and strengthens the "median SOC regression" strategy, which is beneficial to long-term lifespan control.

[0090] Third item: Load adjustment cost function (user disturbance perception target): ; : The adjustment power of the j-th adjustable load at time t (positive for increasing load, negative for decreasing load); : Load adjustment weight, representing the importance of the j-th adjustable load to user comfort or production tasks during adjustment; : Expected response delay (seconds) for the j-th adjustable load; Small positive numbers, to prevent division by zero; Adjustable load quantity; The adjustment range term is multiplied by the "fast response penalty factor", i.e. This reflects that the faster the response, the higher the cost and the more sensitive the equipment, which helps with intelligent sorting and scheduling on the load side.

[0091] The optimization model adopts a linear weighted objective function, in which the weight of the power balance error term is set to 0.5, the weight of the energy storage SOC offset penalty term is 0.3, the weight of the load adjustment cost term is 0.2, and constraints are set such as the load adjustment amplitude must not exceed the lower limit in its maximum capacity vector, the energy storage discharge power must not exceed the physical allowable range, and the response time must meet the action preparation time before the predicted disturbance begins.

[0092] In terms of solution methods, a multi-objective solution framework based on particle swarm optimization is adopted. The robust range of the perturbation variable is set by combining the confidence interval of the perturbation trend, thus avoiding the interference of wind speed prediction errors on the accuracy of the strategy. The model iteration time is set to complete one convergence within 500ms, adapting to the second-level perturbation response requirement.

[0093] Specifically, the solution to the scheduling optimization problem first encodes all decision variables, including the regulation power setpoints of each load node, the charging and discharging power setpoints of the energy storage unit, and the inverter output setpoints, and sets upper and lower limits for each variable to strictly constrain its changes within the range of the dynamic regulation capability vector and the physical boundary.

[0094] In the particle swarm optimization algorithm, each "particle" represents a complete set of scheduling strategies. Its position vector is the current variable setting value, and its velocity vector controls the search direction and step size. During the initialization phase, the particle swarm is uniformly and randomly distributed within the allowed feasible region. Subsequently, the algorithm iterates through rounds, calculating the objective function value for each particle and evaluating its fitness by combining the weighted total cost of the three sub-objectives.

[0095] To improve the convergence efficiency and global search capability of the algorithm, this invention introduces a perturbation trend confidence interval constraint strategy into the basic particle swarm optimization algorithm: when the uncertainty of perturbation trend prediction is high, the key perturbation variables are simulated within their possible intervals, and the robustness of each particle within the confidence interval is scored, prioritizing the retention of particles with stable performance under multi-perturbation scenarios. In each iteration, the particle updates its current velocity and position based on its individual historical best position and global best position, and dynamically adjusts the weight coefficients between local and global settings to achieve adaptive correction of the search direction.

[0096] The entire solution process is set with a maximum number of iterations or a time limit (e.g., no more than 500ms), and terminates early when the convergence threshold condition is met (e.g., the change in the optimal solution is less than a set value for 5 consecutive rounds). The final output is the position of the particle with the best fitness, which serves as the optimal cooperative control strategy solution for the scheduling model in the current cycle. This solution is then converted into a set of control commands by the command generation module, completing the closed-loop execution with the device control end.

[0097] The final output set of collaborative control strategies is a structured list of data, as shown in the example below:

Device ID: Load Node #3, Adjustment Action Type: Load Reduction, Action Intensity: -6kW, Start-up Time: 2s, Duration: 20s, Strategy Confidence: 0.92, Priority: High

Device ID: Energy Storage Device #1, Adjustment Action Type: Charging, Action Intensity: +25kW, Start-up Time: Immediate, Duration: 10s, Strategy Confidence: 0.95, Priority: Medium

Device ID: Wind Power Inverter #A, Adjustment Action Type: Output Setting, Action Intensity: 105kW (Target Setting Value), Start-up Time: 1s, Duration: 15s, Strategy Confidence: 0.98, Priority: High

[0098] Through this process, the present invention completes the closed-loop preparation of the scheduling strategy before the arrival of disturbances, enabling the system to dynamically implement multi-objective collaborative control based on actual capabilities and environmental predictions, ensuring that wind power is not abandoned, energy storage does not exceed limits, and the power grid is not unbalanced, fully demonstrating the system's adaptive scheduling capability and practical engineering value.

[0099] Furthermore, the step of constructing a multi-objective scheduling optimization model based on the dynamic adjustment capability vector and the state-of-charge data of the energy storage device, and outputting a set of collaborative control strategies that satisfy the requirements of grid operation stability, renewable energy output utilization rate, and energy storage device safety threshold, includes: The maximum up-capacity, maximum down-capacity, minimum response delay, maximum adjustment duration, and response reliability score in the dynamic adjustment capability vector are used as the adjustment capability input parameters for each load node. The current state of charge value, allowable charging and discharging power range, and remaining capacity threshold in the energy storage device's state of charge data are used as the constraint inputs on the energy storage side. These are combined to form the parameter set required for scheduling optimization. Based on the disturbance direction and intensity contained in the disturbance trend index set, the load node adjustment direction and adjustment amplitude in the parameter set are matched to initially screen out a set of effective adjustment objects that can complete the response action before the start time of the disturbance trend prediction and whose adjustment direction is consistent with the disturbance trend. For the set of effective adjustment objects, an objective function for the scheduling optimization model is constructed. The objective function includes: minimizing the difference between the sum of power generation output, load power and energy storage charging and discharging power and the current grid load demand to ensure grid operation stability; maximizing the utilization rate of available power output from renewable energy devices while satisfying the boundary conditions of the adjustment capacity of all adjustable points; and setting a penalty term for the state of charge control target of energy storage devices to keep them within the allowable state of charge safety threshold range after the scheduling cycle ends. Based on the above objective function, a confidence interval parameter composed of the prediction error range in the disturbance trend index set is introduced to dynamically adjust the weight of each sub-objective. A smoothness constraint is set on the load-side adjustment frequency to limit the number of times the same load node can start and stop adjustment within a continuous scheduling cycle, thereby avoiding equipment fluctuations caused by frequent adjustments. During the process of solving the objective function and constraints, the response confidence score recorded in the dynamic adjustment capability vector for each load node is dynamically referenced, and all adjustment targets are weighted and sorted according to the response confidence, thereby improving the adjustment execution success rate and control response accuracy in scenarios with high disturbance intensity. The scheduling parameters, including the target power regulation amount, regulation duration, and start time, output from the scheduling optimization model are mapped one-to-one with the node numbers in the set of effective regulation objects to generate a collaborative control strategy structure unit. The charging and discharging regulation commands of the energy storage device and the renewable energy output allocation results are then integrated to construct the collaborative control strategy set that meets the requirements of grid operation stability, renewable energy output utilization rate, and energy storage device safety threshold.

[0100] To achieve stable grid operation under disturbance trends, full absorption of renewable energy output, and safe management of energy storage devices, a multi-objective scheduling optimization model must be constructed based on the system's current response capabilities. This optimization model must not only comprehensively consider the dynamic adjustment capabilities of the three types of objects—source, load, and storage—but also take into account the uncertainty of disturbance trends, energy storage safety constraints, and the feasibility of load response, gradually generating a set of executable collaborative control strategies.

[0101] First, the basic data required for the scheduling optimization model needs to be input from two dimensions. On the one hand, it is necessary to read and organize the dynamic adjustment capability vectors generated by each load node in the previous step, which includes five key indicators: maximum up-range power, maximum down-range power, minimum response delay, maximum adjustment duration, and response reliability score. These data represent the maximum upward and downward adjustment capability boundaries of the load node (in kilowatts), the minimum time required for its adjustment response (usually in seconds), the longest allowable adjustment duration (e.g., not exceeding a certain thermal inertia or process cycle), and the adjustment stability score calculated based on historical performance (usually normalized to a real number between 0 and 1).

[0102] On the other hand, it is necessary to introduce the state of charge (SOC) data of energy storage devices as another input source for the scheduling model. For each energy storage unit, the current SOC, the maximum allowable charge / discharge power (in kilowatts), and the minimum allowable remaining capacity threshold (usually also a percentage) need to be obtained. These data together constitute the operating boundary conditions related to energy storage devices during the scheduling process, in order to prevent problems such as overcharging, over-discharging, or efficiency degradation caused by system scheduling commands.

[0103] The two sets of data above are combined into a parameter set, which serves as the input variable set for the scheduling optimization model. Next, to ensure that the scheduling instructions align with the directionality of the disturbance trend, a preliminary screening of the adjustment targets is required. First, the disturbance direction (e.g., indicating whether the load is expected to decrease or increase) and disturbance intensity (reflecting the urgency and scale of adjustment) for the current period are read from the disturbance trend indicator set. Then, each load node in the dynamic adjustment capability vector is traversed to determine whether its current upscaling or downscaling power aligns with the disturbance direction. For example, when the disturbance trend is upward, only nodes with current upscaling power space are retained. Secondly, it is also necessary to determine whether the minimum response delay of each node is less than the time interval between the current moment and the expected occurrence of the disturbance. If the delay limit is exceeded, the node is considered unable to respond effectively before the disturbance occurs and must be removed from the adjustment target list. After screening based on the above two conditions, a set of effective adjustment targets suitable for scheduling optimization in this period is obtained.

[0104] When constructing the optimization objective function, it is necessary to consider the three core operational objectives of the power grid: power balance, clean energy utilization, and energy storage security. First, the objective function term for grid operational stability is set, which minimizes the difference between the sum of the total power of all generating units, adjustable loads, and energy storage devices and the total system demand within the current cycle. For example, for each scheduling cycle... Its basic form is: When constructing the optimization objective function, it is necessary to consider the three core operational objectives of the power grid: power balance, clean energy utilization, and energy storage security. First, the objective function term for grid operational stability is set, which minimizes the difference between the sum of the total power of all generating units, adjustable loads, and energy storage devices and the total system demand within the current cycle. For example, for each scheduling cycle... Its basic form is: ; in, : Refers to the power generation capacity of a controllable power generation device in the current cycle, measured in kilowatts; This refers to the total adjusted load power, including both positive and negative loads. : Refers to the net output of the energy storage device (positive for discharging, negative for charging); This refers to the total load demand power currently measured by the system.

[0105] The second objective function term is to maximize the utilization of renewable energy output, requiring the absorption of as much available output from intermittent power sources such as solar and wind power as possible. Its optimization objective is to maximize the current predicted output of renewable energy. The proportion of data actually absorbed by the scheduling system. This part does not require additional modeling; simply add a positive objective term to the objective function or add a penalty coefficient to the unutilized portion.

[0106] The third objective function term is energy storage safety control, ensuring that the energy storage SOC remains within a specified safety threshold at the end of the current cycle. For example, the safety threshold range is set as follows: If the expected state exceeds this range, a penalty term is introduced to penalize the scheduling target.

[0107] In the integrated energy adaptive dispatch method described in this invention, to maximize the utilization of renewable energy output, the system first obtains the available output data of various renewable energy devices within each dispatch cycle through the connected wind and solar power forecasting systems before the start of each dispatch cycle. This data is typically predicted based on current meteorological information and equipment characteristic models, accurately estimating the theoretically maximum power output of wind and solar power devices during the current period. Upon receiving this available output data, the dispatch optimization model uses it as a reference upper limit for green energy absorption capacity and sets optimization directions in the dispatch objectives to minimize the unabsorbed portion. During the dispatch scheme evaluation process, the system assigns a reduced weight to the wasted renewable energy power in each feasible strategy or penalizes it as a negative indicator, thereby guiding the dispatch model to automatically prioritize schemes that can accommodate more renewable energy output. In this way, the maximum absorption of wind and solar power output can be achieved without affecting system safety and load operation, effectively reducing wind and solar curtailment and improving the utilization rate of clean energy.

[0108] Meanwhile, to ensure that the energy storage devices remain within the permissible state of charge range during frequent charge and discharge regulation, this invention introduces target constraints for energy storage safety control into the scheduling optimization model. Within each scheduling cycle, the system estimates the energy level of each energy storage unit at the end of the cycle based on the current state of charge (SOC) information of the energy storage devices and the planned charge / discharge power and duration for that cycle. If the prediction indicates that the energy level of a certain energy storage unit will exceed its design upper limit or fall below a set lower limit at the end of the cycle, the system will penalize this strategy during the optimization process, specifically by methods including but not limited to lowering its priority or eliminating its feasibility. The degree of this penalty can be dynamically adjusted according to the extent of the exceedance to ensure that the scheduling system has both flexible adjustment capabilities and does not cause damage to the energy storage devices due to overcharging or discharging.

[0109] Since the prediction results of disturbance trends have a certain degree of error, a confidence interval parameter for disturbance trend prediction needs to be introduced to enhance the robustness of the model in actual operation. This confidence interval can be obtained based on historical data training results or model residual estimation, and the weights of each sub-objective function can be dynamically adjusted during the scheduling optimization process. For example, when the confidence level of the disturbance trend is low, the weight of the response speed objective can be appropriately reduced to avoid false triggering of adjustment actions; conversely, when the confidence level is high, the constraint priority of rapid adjustment response can be increased.

[0110] Furthermore, to prevent frequent start-stop cycles of load nodes within consecutive periods, which could cause equipment overload or user discomfort, a smoothness constraint on adjustment behavior needs to be introduced. This constraint can be set so that each load node cannot adjust its start-stop frequency more than once within three consecutive scheduling cycles, or a minimum adjustment maintenance period can be set to ensure that once in adjustment mode, it must remain in adjustment mode for a certain period of time. This limitation can be controlled by introducing integer state variables or start-stop flags, and can be added to the optimization model as a constraint condition.

[0111] After constructing the entire objective function and constraints, the model solution phase begins. During the solution process, the response reliability score of each load node is used as the scheduling weight. That is, when optimizing target ranking or resource allocation, nodes with high response accuracy are given priority. For example, when power deviation requires coordinated adjustment by multiple nodes, the adjustment ratio is allocated based on response reliability; nodes with higher reliability receive a larger adjustment allocation, thereby improving the success rate of the scheduling strategy under disturbances.

[0112] The model can be solved using linear programming, mixed-integer programming, or other multi-objective constrained optimization methods to obtain the target power adjustment value, corresponding adjustment duration, and control initiation time for each regulated object. These scheduling results are then mapped one-to-one with the regulated object numbers to generate structured scheduling and control strategy units. Each control strategy unit contains the following fields: node number, power setpoint, command issuance time, duration, and priority or confidence level.

[0113] Finally, all control strategy units are combined to form a coordinated control strategy set. This set includes load regulation commands, energy storage charging and discharging control commands, and renewable energy output utilization commands. The data structure can be JSON, XML, or tabular, and it serves directly as input to the control center's scheduling command module. This drives the control equipment to complete second-level dynamic response operations, meeting the system objectives of grid operation stability, renewable energy output utilization rate, and energy storage device safety threshold control.

[0114] In summary, a complete multi-objective scheduling optimization method can be realized by extracting and filtering input parameters, constructing and setting constraints for optimization objectives, controlling the weights of response capability and confidence, and generating the final scheduling results in a structured manner.

[0115] Step S106: The set of collaborative control strategies is converted into a set of control commands and sent to the power generation device, energy storage device and adjustable load device respectively to perform second-level dynamic adaptive scheduling control of the integrated energy system.

[0116] In the integrated energy adaptive scheduling method described in this invention, step S106 is a key step in converting the optimization results into action commands that can be actually executed and sending them to each control unit of the energy system. The quality of its implementation directly determines whether the entire scheduling closed loop can be truly implemented and ensure that the source, load and storage devices respond to the scheduling strategy as expected.

[0117] This step first classifies and maps the control objectives into strategies and instruction structures based on the set of cooperative control strategies output in step S105. The set of control strategies is essentially a structured data packet containing elements such as the controlled object, control objective, control strength, control direction, duration of action, and priority order within the current period. In actual systems, the control instruction formats, communication interfaces, and protocols required by different controlled objects often differ; therefore, unified parsing and formatting conversion are necessary.

[0118] For power generation devices, such as wind turbines and photovoltaic inverters, control commands must include parameters such as active power setpoints, reactive power compensation setpoints, and voltage control mode switching flags. These commands will be issued through the photovoltaic controller or wind farm monitoring system and will maintain interface consistency with the grid automatic voltage control (AVC) system. The command format must comply with mainstream standards such as IEC 61850 or Modbus to ensure smooth integration into existing dispatching systems.

[0119] For energy storage devices, such as battery energy storage systems, control commands should include power settings for charging or discharging, maximum allowable current, voltage protection lower limit, and duration of action. The commands are received by the energy storage management system (EMS or BMS), which verifies in real time whether the state of charge (SOC) meets the command execution conditions to prevent execution beyond the SOC limit. If necessary, the commands should also include auxiliary control parameters such as temperature protection mechanisms to ensure safe operation of the equipment under disturbances.

[0120] For adjustable load devices, such as building air conditioning, industrial motors, lighting group control, or cold chain loads, control commands can include on / off commands, power setting ranges, start / stop priorities, and maximum response delays. To achieve flexible control, the system can also support a distributed response mode, which allows the local controller to make optimal decisions under local conditions based on the received target setpoints, such as using tiered load switching, frequency response, or comfort-first strategies.

[0121] All control commands should undergo real-time status checks and conflict detection by the dispatch center before being issued. For example, when multiple strategies coexist, if the same object receives both charging and peak shaving commands simultaneously, arbitration or a weighted fusion mechanism should be used based on preset priorities. Once a command is confirmed, it is sent from the dispatch center to the corresponding control terminal via a secure communication link (such as a 5G private network, industrial Ethernet, NB-IoT, etc.). The entire process should have low latency, high stability, and a mechanism for retransmission in case of disconnection to ensure second-level dispatch accuracy.

[0122] After the command is issued, the system should initiate an execution confirmation mechanism, requiring each terminal to report the status change results after the command is executed, including whether the response was timely, the actual response magnitude and the deviation from the expected value, etc., for subsequent scheduling accuracy correction and model adaptive adjustment. Simultaneously, this step should also have a logging function to archive and save each control command issuance and response, facilitating iterative optimization and operational analysis of the scheduling strategy.

[0123] Through the above mechanism, this step completes a closed loop from strategy set to control action, enabling the entire integrated energy system to achieve precise and rapid linkage response when facing second-level meteorological disturbance trends, thereby ensuring grid stability, improving energy efficiency and extending equipment life.

[0124] To facilitate understanding, the following concrete example illustrates how to transform a set of cooperative control strategies into a set of control instructions and complete the actual scheduling and execution process. Within a certain scheduling cycle, the system generates the following three cooperative control strategies:

Device ID: Energy Storage Device #2, Regulation Action Type: Discharge, Action Intensity: +20kW, Start-up Time: 0s, Duration: 15s, Priority: High

Equipment ID: Load Node #5 (Cold Storage Compressor), Adjustment Action Type: Load Reduction, Action Intensity: -4kW, Start-up Time: 2s, Duration: 10s, Priority: Medium

Device ID: Photovoltaic Inverter #A, Adjustment Action Type: Output Setting, Target Value: 85kW, Start-up Time: 0s, Duration: 20s, Priority: High

[0125] For energy storage device #2, the system generates an EMS control command with the following format: {Device ID: ES2, Type: Discharge, Power: 20kW, Current Limit: 40A, Duration: 15s, Startup Time: Immediate}; The instruction is issued through the BMS interface and includes SOC out-of-range protection and over-temperature protection flags.

[0126] For load node #5, the system identifies it as an industrial cold chain load with a local intelligent control unit, and the switching control strategy is as follows: {Node ID: LD5, Type: Load Reduction, Power Reduction: 4kW, Mode: Automatic Adjustment, Maximum Delay: 2s, Duration: 10s}; This instruction employs a local execution confirmation mechanism, allowing the controller to select the optimal adjustment method within a specified range based on the current compressor frequency status.

[0127] For photovoltaic inverter #A, the system generates the following configuration commands through the AVC platform: {Device ID: PV-A, Output setting: 85kW, Power factor target: 0.99, Voltage control mode: Constant voltage, Start-up: Immediate, Duration: 20s}; The instructions are transmitted to the inverter communication interface using the Modbus TCP protocol, and the response log is recorded by the central dispatch server.

[0128] After format conversion, the three control commands are uniformly added to the scheduling and distribution queue. After status conflict checks (confirming that photovoltaic power generation is not concurrently limited, energy storage is not in a fault state, and the load is not overloaded), they are distributed from the dispatch center to each terminal via the 5G private network. Each terminal will provide feedback on execution status, effective delay, execution error, and other information after execution. The system will then dynamically correct the model parameters or adjust the scheduling weight for the next cycle based on this information.

[0129] The second embodiment of the application provides an electronic device, the electronic device comprising: processor; The memory is used to store a program, which, when read and executed by the processor, executes the integrated energy adaptive scheduling method based on weather forecasting provided in the first embodiment of this application.

[0130] The third embodiment of this application provides a computer-readable storage medium storing a computer program thereon. When the program is executed by a processor, it executes a comprehensive energy adaptive scheduling method based on weather forecasting provided in the first embodiment of this application.

[0131] Although this application discloses preferred embodiments as described above, it is not intended to limit this application. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of this application. Therefore, the scope of protection of this application should be determined by the scope defined in the claims of this application.

Claims

1.A method for integrated energy adaptive scheduling based on weather prediction, characterized in that, The method comprises the following steps: Collecting second-level meteorological disturbance data in the target area, including wind speed, solar radiation, temperature and humidity, and synchronously collecting power consumption load power data, energy storage device state of charge data and power grid boundary flow data; Fusing the second-level meteorological disturbance data, power consumption load power data, energy storage device state of charge data and power grid boundary flow data, constructing a disturbance response feature set, and constructing a disturbance response mapping structure representing the time correlation relationship between disturbance variables and load variables based on the disturbance response feature set; Based on the disturbance response mapping structure, predict the change trend of the disturbance variable in the second-level period, and generate a disturbance trend index set containing the disturbance direction, disturbance intensity and response sensitive section; Combine the disturbance trend index set with the current power consumption load power data to generate a dynamic adjustment capability vector for the power consumption load, and the dynamic adjustment capability vector is used to represent the adjustment flexibility and adjustment amplitude of the power consumption load under different disturbance trends; According to the dynamic adjustment capability vector and the energy storage device state of charge data, a multi-objective scheduling optimization model is constructed, and a cooperative control strategy set meeting the requirements of power grid operation stability, renewable energy output utilization rate and energy storage device safety threshold is output; The cooperative control strategy set is converted into a control instruction set and respectively sent to the power generation device, the energy storage device and the adjustable load device to perform second-level dynamic adaptive scheduling control on the integrated energy system. 2.The intelligent power distribution load forecasting and adaptive scheduling method of claim 1, wherein, The fusion of the second-level meteorological disturbance data, power consumption load power data, energy storage device state of charge data and power grid boundary flow data, the construction of the disturbance response feature set, and the construction of the disturbance response mapping structure representing the time correlation relationship between disturbance variables and load variables comprise: A three-dimensional sliding tensor containing time index, disturbance variable index and load variable index is constructed in continuous second-level time window, and the sliding tensor is used to extract the joint evolution characteristics of each disturbance variable and each load variable in each time window to form a disturbance response feature set containing disturbance change rate, disturbance standard deviation, load response gradient and power change consistency factor; In constructing the disturbance response mapping structure representing the time correlation relationship between disturbance variables and load variables, based on the disturbance life cycle range of each disturbance variable in the disturbance response feature set, a causal determination method based on time delay cross correlation is used to construct the causal response path between the disturbance variable and the load variable, and the disturbance life cycle is taken as the constraint boundary; According to the causal response path, a dynamically updatable bidirectional weighted causal graph structure is constructed between each disturbance variable and load variable, the weight of each edge in the bidirectional weighted causal graph structure is calculated based on disturbance intensity index, disturbance stability factor and disturbance life cycle standardized length, and a signal-to-noise ratio adaptive adjustment mechanism is introduced to compress and correct the causal edge weight in the low credibility data window; The bidirectional weighted causal graph structure is taken as the disturbance response mapping structure. 3.The intelligent power distribution load forecasting and adaptive scheduling method of claim 1, wherein, The combination of the disturbance trend index set and the current power consumption load power data to generate a dynamic adjustment capability vector for the power consumption load comprises: extracting the current power value, the maximum up-regulation power, the maximum down-regulation power, the minimum response time delay, the maximum regulation duration and whether the current state is within the adjustable range of each load node from the current power consumption load power data based on the disturbance direction, the disturbance intensity and the response sensitive section contained in the disturbance trend index set, to form an initial adjustment parameter set; corresponding the initial adjustment parameter set with the disturbance trend index set, judging whether each load node can participate in the current adjustment according to the consistency of the disturbance direction and the adjustable direction of each load node, and judging whether the load node belongs to the response sensitive section contained in the disturbance trend index set, eliminating the load nodes that do not meet the above two conditions to obtain a disturbance response effective node set; for each load node in the disturbance response effective node set, calculating the current response lag time, i.e. the expected minimum start-up time required for load response instruction execution, and comparing it with the predicted disturbance trend start time in the disturbance trend index set to obtain a response delay value, and introducing a penalty factor to proportionally reduce the maximum adjustable power value of each load node according to the response delay value, the larger the response delay, the lower the penalty factor, which is used to reduce the calculation weight of the adjustment amplitude; combining the current operation cycle information of each load node, extracting the recovery time experienced after the last adjustment behavior, and judging whether it can participate in adjustment again within the current disturbance window based on the recovery time, if the recovery time is insufficient, reducing the adjustment amplitude estimation value and marking the adjustment state of the node as limited; calculating the response accuracy of each load node in the past disturbance, including the adjustment amplitude and target deviation, adjustment time error and execution completion rate under the past disturbance instruction, and giving a response credibility score according to the response accuracy, the higher the response credibility score, the more suitable the load node is for participating in adjustment allocation under the current disturbance trend; combining the maximum up-regulation power, the maximum down-regulation power, the minimum response time delay, the maximum regulation duration and the response credibility score of each disturbance response effective node in the current scheduling cycle as a structured description unit, and summarizing them according to the load node number to form a dynamic adjustment capability vector for power consumption load. 4.The intelligent power distribution load forecasting and adaptive scheduling method of claim 3, wherein, the multi-objective scheduling optimization model is constructed according to the dynamic adjustment capability vector and the state of charge data of the energy storage device, and a set of collaborative control strategies that meet the requirements of power grid operation stability, renewable energy output utilization rate and energy storage device safety threshold are output, including: the maximum up-regulation power, the maximum down-regulation power, the minimum response time delay, the maximum regulation duration and the response credibility score in the dynamic adjustment capability vector are used as the adjustment capability input parameters of each load node, and the current state of charge value, the allowed charging and discharging power range and the remaining capacity threshold in the state of charge data of the energy storage device are used as the constraint input of the energy storage side, to form a parameter set required for scheduling optimization; Based on the disturbance direction and intensity contained in the disturbance trend index set, the adjustment direction and amplitude of the load node in the parameter set are matched, and the effective adjustment object set that can complete the response action before the disturbance trend prediction starting time and is consistent with the disturbance trend in the adjustment direction is preliminarily screened out; For the effective adjustment object set, the objective function of the scheduling optimization model is constructed, which includes: minimizing the difference between the sum of the generation output, the load power and the energy storage charging and discharging power and the current time grid load demand to ensure the stability of the grid operation; at the same time, under the premise of meeting all the adjustable node adjustment ability boundary conditions, the available output utilization rate from the renewable energy device is maximized; and a penalty term is set for the state of charge control target of the energy storage device to keep it within the allowed state of charge safety threshold range after the scheduling period ends; On the basis of constructing the above objective function, the confidence interval parameter composed of the prediction error range in the disturbance trend index set is introduced to dynamically adjust the weight of each sub-target, and the smoothness constraint of the load side adjustment frequency is set to limit the adjustment start-stop times of the same load node in the consecutive scheduling period, avoiding the device fluctuation caused by frequent adjustment; During the solving process of the objective function and the constraint condition, the response confidence score recorded by each load node in the dynamic adjustment ability vector is dynamically referenced, and all adjustment targets are weighted and sorted according to the response confidence, improving the adjustment execution success rate and control response accuracy in the high-intensity disturbance scenario; The scheduling parameters including the target power adjustment amount, adjustment duration and start time output by the scheduling optimization model solution are one-to-one corresponding to the node numbers in the effective adjustment object set, generating the cooperative control strategy structure unit, and the charging and discharging adjustment instructions of the energy storage device and the renewable energy output allocation results are fused to construct the cooperative control strategy set that meets the requirements of grid operation stability, renewable energy output utilization rate and energy storage device safety threshold.

Images