Multi-energy power system energy storage coordinated dispatching method, device and medium

Abstract

The present application relates to the technical field of power system operation and control, and discloses a multi-energy power system energy storage coordinated scheduling method, equipment and medium, comprising: establishing a dynamic characteristic model reflecting the aging and loss evolution of the energy storage unit, expressed as a dynamic function of historical operation cumulative quantity, current state of charge and external environment parameters; performing fusion analysis on historical meteorological sequences, load sequences and carbon price sequences, and outputting interval prediction results of new energy output, system load and carbon price in the future scheduling period; dynamically calculating a trigger index representing system uncertainty according to the interval prediction results, and executing optimized scheduling according to the trigger index; updating the energy storage unit state according to the actual operation data, and correcting the dynamic characteristic model. The present application can intelligently cope with uncertainty, thereby significantly improving new energy consumption capacity and reducing overall operation cost.

Description

Technical Field

[0001] This invention relates to the field of power system operation and control technology, and in particular to a method, equipment and medium for coordinated dispatch of energy storage in multi-energy power systems. Background Technology

[0002] A multi-energy power system is essentially a comprehensive system integrating multiple energy forms. Its power supply side includes various power generation units such as thermal power, hydropower, wind power, and photovoltaic power; the transmission side is mainly composed of ultra-high voltage and extra-high voltage transmission networks; the energy distribution side includes distribution networks, heating networks, gas networks, and hydrogen energy networks; the load side includes various types of energy consumption units such as electricity, heat, cooling, hydrogen, and natural gas; and it is equipped with various forms of energy storage units such as electricity storage, thermal storage, hydrogen storage, gas storage, cold storage, and water storage. The entire system is an integrated system of multi-energy supply, transmission, conversion, storage, and consumption, achieved through the combined action of many links and devices, including "source-grid-load-storage-conversion." Because energy forms are diverse and their time spans are relatively large, the system must achieve energy and power balance at every temporal and spatial level. Short-term balance determines the system's dynamic characteristics and stability, while medium- and long-term balance determines its economic efficiency and operational effectiveness.

[0003] In a future power structure where wind power, solar power, and other new energy sources account for 50% to 70% of installed capacity, to ensure good performance in terms of energy economy, energy quality, and energy reliability, and to meet the requirements of zero-carbon or even negative-carbon energy supply, it is essential to achieve deep synergy and efficient coordination between various energy storage methods such as electricity storage, thermal storage, and hydrogen storage, and various power sources such as wind power, solar power, hydropower, and thermal power. Only on the basis of full coordination among multi-energy storage and multiple types of power sources can we continuously improve the safety and stability, carbon emission reduction capabilities, energy economy, and energy quality of the new power system that is based on new energy sources and compatible with multiple energy forms. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention provides a method, equipment and medium for coordinated scheduling of energy storage in multi-energy power systems.

[0005] Therefore, the technical problem solved by this invention is that in traditional multi-energy power system dispatching, the scheduling plan does not match the actual dynamic capabilities of the equipment due to the static simplification of the energy storage model, the rigidity of the scheduling framework, and the single optimization objective.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a multi-energy power system energy storage coordinated scheduling method, comprising: establishing a dynamic characteristic model reflecting the aging and loss evolution of energy storage units, expressed as a dynamic function of historical operating accumulation, current state of charge and external environmental parameters; By integrating and analyzing historical meteorological sequences, load sequences, and carbon price sequences, the system outputs interval forecasts for renewable energy output, system load, and carbon price during future scheduling periods. Based on the interval prediction results, trigger indicators representing system uncertainty are dynamically calculated, and optimized scheduling is performed according to the trigger indicators. The objective function for optimized scheduling includes the dynamic loss cost of energy storage calculated in real time based on the dynamic characteristic model, and the dynamic characteristic model is used as the core constraint for optimized scheduling. The energy storage unit status is updated based on actual operating data, and the dynamic characteristic model is corrected.

[0007] As a preferred embodiment of the multi-energy power system energy storage coordinated scheduling method of the present invention, the establishment of a dynamic characteristic model reflecting the aging and loss evolution of energy storage units includes collecting historical operating data of energy storage units within a set operating cycle, wherein the historical operating data includes charging and discharging current sequence, terminal voltage sequence, ambient temperature sequence and initial state of charge. Using the historical operating data as input, an equivalent internal resistance dynamic model characterizing the evolution of the internal health state of the energy storage unit is established, wherein the equivalent internal resistance is modeled as a continuous function of the cumulative throughput of the energy storage unit, the current state of charge, and the current operating temperature.

[0008] As a preferred embodiment of the multi-energy power system energy storage coordinated scheduling method of the present invention, wherein: the output of the interval prediction results of new energy output, system load and carbon price in the future scheduling period includes constructing a sequence prediction model containing an encoder and a decoder; The encoder includes a parallel first sub-coding network, a second sub-coding network, and a third sub-coding network, which are used to process historical meteorological feature time series, historical system load time series, and carbon price series, respectively, for feature extraction. The decoder calculates the attention weights corresponding to each feature vector in the three sequences and generates corresponding context vectors. The corresponding context vectors are concatenated and fused and then output through a fully connected output layer to output the quantile prediction values ​​for future continuous scheduling periods. The quantile predictions include high quantile predictions, middle quantile predictions, and low quantile predictions, which together constitute the interval prediction results.

[0009] As a preferred embodiment of the multi-energy power system energy storage coordinated scheduling method of the present invention, the step of performing optimized scheduling based on the triggering index includes obtaining the interval prediction results for the current and near future periods at each decision time. Based on the preset high quantile and low quantile forecast values, the forecast uncertainty width of new energy output in the first period is calculated, and then normalized by dividing by the median quantile forecast value of the corresponding period to obtain the first uncertainty component. Calculate the instantaneous rate of change of the system load at the current moment, as the second uncertainty component; The first uncertainty component and the second uncertainty component are weighted and summed to obtain a comprehensive trigger index; The triggering indicator is compared with a preset triggering threshold. If the triggering indicator is greater than the triggering threshold, a rolling optimization scheduling is initiated after determination. If the triggering indicator is less than or equal to the triggering threshold, the existing scheduling plan is maintained.

[0010] As a preferred embodiment of the multi-energy power system energy storage coordinated scheduling method of the present invention, the step of performing optimized scheduling based on the trigger index further includes dynamically setting the look-ahead time window length of this rolling optimization according to the value of the trigger index; Preset the maximum and minimum optimized window lengths; Calculate the difference between the trigger indicator and the trigger threshold to determine the window reduction amount; The difference between the maximum optimized window length and the window reduction amount is used as the initial time window length; If the initial time window length is less than the minimum optimization window length, then the time window length for this rolling optimization is the minimum optimization window length; otherwise, the initial time window length is used.

[0011] As a preferred embodiment of the multi-energy power system energy storage coordinated scheduling method of the present invention, the step of using the dynamic characteristic model as the core constraint for optimized scheduling includes introducing the state update equation in the dynamic characteristic model as an equality constraint into the optimized scheduling. The state update equation defines the mathematical relationship between the state of the energy storage unit at the end of the current scheduling period, the state at the end of the previous scheduling period, and the charging and discharging power in the current period. Meanwhile, the maximum allowable charging power and maximum allowable discharging power of the energy storage unit in the current state, calculated based on the dynamic characteristic model, will be used as the power inequality constraint of the current energy storage unit in the current scheduling period.

[0012] As a preferred embodiment of the multi-energy power system energy storage coordinated scheduling method of the present invention, the correction of the dynamic characteristic model includes collecting the actual operating data of the energy storage unit after each scheduling period, including the actual charging and discharging power, temperature and measured or estimated values ​​of state of charge. Using the actual operating data, the recursive least squares algorithm is employed to identify and update the function parameters in the dynamic characteristic model online. The updated function parameters are applied to the dynamic characteristic model of the next scheduling cycle to achieve closed-loop adaptive correction of the model.

[0013] This invention provides a multi-energy power system energy storage coordination and dispatch system.

[0014] As a preferred embodiment of the multi-energy power system energy storage coordinated dispatching system of the present invention, wherein: the dispatching instruction generated by the optimized dispatching is obtained; The operation of the energy storage unit in the multi-energy power system is controlled based on the dispatch instructions; during the control process, the state of the energy storage unit is updated and the dynamic characteristic model is corrected based on the actual operating data.

[0015] The present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of a multi-energy power system energy storage coordinated scheduling method.

[0016] The present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of a multi-energy power system energy storage coordinated scheduling method.

[0017] The beneficial effects of this invention are as follows: an economic cost model of the internal losses of the energy storage device is constructed based on the real-time status, enabling the dispatcher to actively manage the lifespan of the energy storage; the use of multi-source information fusion for interval prediction and robust optimization greatly enhances the system's ability to cope with the uncertainties of new energy power generation and load, and improves the system's safety and economy. A triggering and elastic window mechanism based on dynamic assessment of uncertainty is designed, so that the scheduling rhythm is no longer a fixed cycle, but can respond intelligently according to different needs, thereby optimizing a more reasonable and effective way of allocating computing resources; and online calibration and regular updates of the prediction model are carried out with the help of actual operation data, so that the system can adapt to environmental changes in the long term. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1This is a schematic diagram of a multi-energy power system energy storage coordinated scheduling method provided in one embodiment of the present invention.

[0020] Figure 2 This is a flowchart of the interval prediction process for a multi-energy power system energy storage coordinated scheduling method provided in one embodiment of the present invention.

[0021] Figure 3 The flowchart shows the corrected dynamic characteristic model of a multi-energy power system energy storage coordinated scheduling method provided in one embodiment of the present invention. Detailed Implementation

[0022] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0023] Example 1, referring to Figure 1 This is the first embodiment of the present invention, which provides a method for coordinated scheduling of energy storage in a multi-energy power system, including: S1: Establish a dynamic characteristic model that reflects the aging and loss evolution of energy storage units, expressed as a dynamic function of historical accumulated amount, current state of charge and external environmental parameters.

[0024] S2: Perform fusion analysis on historical meteorological sequences, load sequences, and carbon price sequences to output interval prediction results for renewable energy output, system load, and carbon price during future scheduling periods.

[0025] S3: Dynamically calculate the trigger index representing the uncertainty of the system based on the interval prediction results, and perform optimized scheduling based on the trigger index.

[0026] S4: Update the energy storage unit status based on actual operating data and correct the dynamic characteristic model.

[0027] It should be noted that, compared with the currently widely adopted technical solutions, the core function of this invention lies in coordinating energy storage resources in the power system in a more refined, flexible and predictive manner.

[0028] Existing methods typically model energy storage devices as simple units with fixed parameters and make scheduling decisions on a fixed timetable. This makes it difficult to accurately reflect the performance degradation of batteries in actual operation and lacks effective means to cope with the uncertainty of wind and solar power output and load fluctuations.

[0029] This embodiment establishes a dynamic model that reflects factors such as battery aging and temperature, enabling the scheduling system to perceive the health status of the equipment. By using interval prediction instead of single numerical prediction, it provides a risk buffer for decision-making. By dynamically triggering optimization based on real-time uncertainty, it breaks the rigid time cycle limitation. By continuously using operational data to update the model, the system can autonomously evolve along with equipment aging and changes in the external environment, thus achieving a better balance between safety, economy and long-term equipment operation.

[0030] Example 2, refer to Figure 2 and Figure 3 As an embodiment of the present invention, based on the above embodiment, a method for coordinated scheduling of energy storage in a multi-energy power system is provided.

[0031] Furthermore, in this embodiment, step S1 establishes a dynamic characteristic model reflecting the aging and loss evolution of the energy storage unit, expressed as a dynamic function of historical accumulated operating amount, current state of charge, and external environmental parameters. Specific steps include S101-S103: S101: For the target electrochemical energy storage unit, continuously collect its historical operating data over a complete life cycle or a representative operating phase (no less than 6 months). Required data includes: The charging and discharging current sequence is generated by measuring the main circuit current using a current sensor, forming a time series. The current value is specified to be positive during discharging and negative during charging, and this is maintained throughout the entire data processing flow. The sampling frequency is no less than 1kHz to meet the requirements for internal resistance calculation. The terminal voltage sequence is formed by measuring the terminal voltage at the positive and negative terminals of the battery cluster, creating a time sequence synchronized with the current. Ambient temperature sequence: Multiple temperature sensors are arranged at representative locations within the electrochemical energy storage unit (such as air inlet, air outlet, and surface center) to record the ambient temperature change sequence. Initial state of charge (SOC): At the beginning of the data recording period, an accurate initial SOC value is read through the BMS (Battery Management System) and used as the absolute benchmark for the entire subsequent SOC timing estimation. All data is collected and preprocessed by a unified data acquisition unit, and microsecond-level timestamps are used to ensure that data points of different physical quantities such as current, voltage, and temperature are strictly aligned on the time axis.

[0032] The original sampled data with timestamps is organized into a standardized data sequence arranged in chronological order; Immediately after generating the sequence, the first round of data cleaning is performed using a sliding window-based statistical filtering algorithm. The sliding window length is set to 100 sampling points (corresponding to a time span of 0.1 seconds for a 1kHz sampling rate), and the window step size is 1 sampling point, enabling point-by-point sliding. Instantaneous outliers in the current and voltage sequences are identified and marked. For marked outliers, linear interpolation of the preceding and following valid data is used to replace them, ensuring the continuity of the sequence.

[0033] Using the cleaned high-frequency current sequence and initial state of charge, a preliminary, high-time-resolution SOC estimation sequence is generated through the ampere-hour integration method, specifically including: The accurate SOC value at the start of the data recording is used as the initial value of the sequence; the high-frequency sampling current value at the current moment is obtained, where discharge is positive and charging is negative.

[0034] Calculate the change in battery charge within the current tiny time interval, subtract the current charge change from the previous SOC value to obtain the current SOC estimate; check whether the updated SOC value exceeds the physical range of [0,1]. If it does, force it to be set to the boundary value of 0 or 1.

[0035] The time pointer is advanced to the next moment, and the steps are repeated until all current data points have been processed, finally resulting in a high-time-resolution SOC estimation sequence that is perfectly aligned with the timestamps of the original current sequence.

[0036] By utilizing high-frequency voltage and current sequences, the instantaneous state of charge (SOC) is estimated during periods when the current is close to zero, which is then used for subsequent fusion calibration. The original current sequence is transformed into a sequence of state variables crucial for model training.

[0037] S102: From the preprocessed long sequence data, slide to extract multiple fixed-length subsequences. For each subsequence, perform downsampling and feature extraction at scheduling intervals. Construct input features: For each 15-minute window, the SOC value corresponding to the start time is used as the SOC feature input for that window, denoted as... ; The weighted average of the relevant temperature sensor readings within the calculation window is used as The normalized temperature is obtained by normalization. ; Calculate the cumulative charge / discharge ampere-hours from the start of the current segment to the end of the current window, as... .

[0038] To construct the output target, for the same window, select several short periods of relatively stable current (such as constant current charging and discharging segments). Using the voltage changes and current values ​​at the beginning and end of the selected periods, calculate the average ohmic internal resistance of the current period based on a simplified battery model (such as the Rint model), and use it as the reference true value of the internal resistance of the window. Finally, all windows The input feature matrix and the corresponding internal resistance reference true value are used as the output target vector, together forming the offline training dataset.

[0039] A nonlinear least squares optimization algorithm with boundary constraints is employed to find the parameter vector that globally minimizes the root mean square error between the model's predicted internal resistance and the reference internal resistance. Evaluation is performed using a separate data sequence that was not used in the training (i.e., the validation set). Performance; The prediction error of the calculated model on the validation set is used to confirm its generalization ability, and once this is confirmed, it is considered as the final offline parameter that can be used for online deployment.

[0040] S103: Establish an equivalent internal resistance dynamic model characterizing the evolution of the internal health state of the energy storage unit. Specifically, the equivalent internal resistance is modeled as a continuous function of the cumulative throughput of the energy storage unit, the current state of charge, and the current operating temperature. in, This represents the equivalent series resistance in the k-th time interval of the model output. This represents the state of charge of the energy storage unit at the beginning of the k-th time period. This represents the normalized temperature at time k. It represents the cumulative cyclic throughput ampere-hours starting from a certain reference state point, and is updated by calculating the sum of the absolute charge and discharge ampere-hours of the current period and the cumulative value of the previous period; , Indicates the reference internal resistance. This represents the coefficient of influence of SOC on internal resistance. and Together, they represent the nonlinear relationship between temperature and internal resistance. This represents the coefficient of influence of cyclic aging on the increase of internal resistance.

[0041] It should be noted that this embodiment sets up a typical regional power system, including thermal power, wind power, photovoltaic power, and a lithium-ion battery energy storage power station. The system dispatch cycle is 15 minutes (i.e., Δt = 0.25 hours). The energy storage power station has a rated capacity of 10MWh, a rated DC voltage of 1000V, and a rated power of 5MW. Through offline identification using historical data, the dynamic internal resistance model parameters of this energy storage unit are obtained as follows: =0.001, =-0.0002, =0.005, =0.1, =0.0005 (unit: ohms).

[0042] Before the start of a certain scheduling period k, the state of the energy storage unit is as follows: State of charge at the end of the previous period The current time period weighted average temperature ℃, normalized temperature Cumulative cycle ampere-hours Substituting into the equivalent internal resistance dynamic model for calculation, we obtain... That is, the dynamic internal resistance at the beginning of this period is about 9.89mΩ, which is within the range of typical lithium-ion battery internal resistance, and it increases reasonably with the degree of aging.

[0043] Furthermore, in this embodiment, step S2 performs a fusion analysis of historical meteorological sequences, load sequences, and carbon price sequences to output interval prediction results for renewable energy output, system load, and carbon price during the future scheduling period. Specific steps include S201-S203: S201: Construct a sequence prediction model that includes an encoder and a decoder. For each prediction execution time, the system prepares three sets of historical sequences of fixed length, including meteorological sequences, system load sequences, and carbon price sequences. A meteorological series is a sequence of multidimensional meteorological observation data over L consecutive time periods. Typical characteristic dimensions include total horizontal irradiance, direct normal irradiance, ambient temperature, relative humidity, wind speed, and wind direction. The data must be normalized before input.

[0044] The system load sequence, along with the historical values ​​of the total active power load of the system within the same time period as the meteorological sequence, constitutes a one-dimensional sequence, which is then normalized.

[0045] The carbon price series, as an independent time series input, needs to maintain the same time resolution and historical window length as the meteorological and load series. The raw carbon price data (such as the average closing price every 15 minutes or hour) constitutes a one-dimensional series, which needs to be normalized in the same way as the meteorological and load data before being input into the model, in order to eliminate the influence of dimensions and accelerate model convergence.

[0046] S2011: Employs a three-encoder structure to extract deep features from meteorological, load, and carbon valence sequences respectively.

[0047] Among them, the first sub-coding network is the meteorological coding branch. The meteorological feature sequence first extracts local spatiotemporal features through a one-dimensional convolutional neural network layer, and then inputs it into the first bidirectional long short-term memory network to capture the long-period, bidirectional temporal dependency relationship of meteorological conditions and output the meteorological hidden state sequence. The second sub-encoding network is the load encoding branch. The system load sequence is directly input into the second independent Bi-LSTM (bidirectional long short-term memory) network to learn the daily, weekly, and yearly time-series patterns of load changes and output the load hidden state sequence. The third sub-encoding network is the carbon price encoding branch. The carbon price sequence is input into the third Bi-LSTM network to learn the temporal patterns and periodic characteristics of carbon trading price fluctuations, and outputs the carbon price hidden state sequence.

[0048] S2012: The decoder uses the deep features obtained from the encoder to generate multivariate future prediction sequences. The initial hidden state of the decoder is obtained by fusing the final states of the load encoding branch and the carbon price encoding branch through a fully connected layer, and also contains basic trend information of historical load and carbon price.

[0049] At each prediction execution time of the decoder, a three-stage information retrieval and fusion process is performed, specifically including meteorological condition attention retrieval, load self-attention retrieval, and carbon price trend attention retrieval. The decoder calculates the similarity between the current hidden state of the meteorological sequence and all historical hidden states output by the meteorological coding branch to obtain another set of weights. The weights are used to sum the meteorological hidden states to generate a meteorological driving context vector. Specifically, the weights are calculated by using the current hidden state of the decoder as the query vector and the hidden states of all historical moments output by the meteorological coding branch as the key vectors. A similarity score representing the degree of matching between the current decoded state and each historical meteorological state is obtained by calculating the dot product of the query vector and each key vector. These scores are then normalized using a softmax function so that the sum of all scores equals 1, thus yielding the final weights.

[0050] The system can dynamically select the most relevant weather pattern corresponding to the current forecast time from a large amount of historical meteorological data, so that the forecast of meteorological sensitive variables such as wind and solar power output has a physical basis and is accurate.

[0051] The decoder calculates the similarity between the current hidden state of the system load sequence and all historical hidden states output by the load coding branch, obtains the weights, and then sums them up to generate a load trend context vector. This process involves identifying historical load change patterns similar to the current forecast scenario, i.e., similar patterns of weekdays, holidays, or event days, to improve the accuracy of load forecasting.

[0052] The decoder calculates the similarity between the current hidden state of the carbon price sequence and all historical hidden states output by the carbon price encoding branch, and obtains a third set of weights. The carbon price hidden states are then weighted and summed to generate a carbon price trend context vector.

[0053] Finally, the generated three context vectors—weather-driven, load-trend, and carbon-price-trend—are concatenated to obtain a complete fused context vector. The decoder uses the prediction output from the previous step and the fused vector as input to update its internal state.

[0054] The fusion mechanism considers the influence of three main factors—meteorological drivers, load inertia, and carbon price market signals—simultaneously in its predictions at each time point, enabling deep integration of multi-source information. The resulting prediction sequences for new energy sources, load, and carbon prices are logically consistent and coordinated. The updated hidden state is fed into the fully connected output layer, directly obtaining the quantile prediction values ​​for each target variable. This provides downstream scheduling decisions with quantitative uncertainty and predictive inputs that consider the correlation of multiple factors.

[0055] S2013: The fully connected output layer directly outputs the quantile predictions of multiple target variables.

[0056] For wind power output and photovoltaic power output, this layer outputs three scalar values, corresponding to the decimal place (0.1), median place (0.5), and ninetieth place (0.9), respectively. For the total system load, these three percentile values ​​are also output; For regional carbon trading prices, the output layer outputs the predicted quantile value (0.5).

[0057] The model provides the scheduling synchronization with a description of the uncertainty range of future renewable energy and load, as well as a benchmark expectation of carbon price, during a single forward propagation.

[0058] S202: Take historical datasets from 3 to 5 years. Align the timestamps of meteorological, load, renewable energy output, and carbon price data to the same time base, and uniformly resample or interpolate them at 15-minute intervals. If the carbon price data is not continuously traded, then the prices during non-trading periods need to be filled forward.

[0059] The dataset was divided strictly according to chronological order to avoid future information leaks. The earliest 70% of the data was used as the training set, the middle 15% as the validation set, and the last 15% as the test set.

[0060] The model is trained using a quantile loss function. For the three variables that require interval prediction—wind power, solar power, and load—the losses at the three quantiles τ=0.1, 0.5, and 0.9 are calculated respectively. For the carbon price variable, only the loss at the quantile (τ=0.5) is calculated. The total training loss of the model is the sum of the loss terms of all variables at all specified quantiles.

[0061] The adaptive moment estimation algorithm was used, with an initial learning rate of 0.001. A learning rate decay strategy based on the plateau of the validation set loss was employed, and dropout regularization was applied to the Bi-LSTM layer and the fully connected layer. The training process continued until the validation set loss no longer improved, triggering an early stopping mechanism.

[0062] S203: Package the trained model structure, weights, and normalized parameters, and deploy them as a prediction service that can be invoked remotely.

[0063] The forecasting service needs to periodically retrieve the latest historical sequence window of length L from the real-time database, including meteorological, load, and carbon price data, and perform normalization processing on the data in the same way as during the training phase. The system performs rolling forecasts, generating forecasts for the next 24 hours at 15-minute intervals, including the low, medium, and high percentiles of wind power output, photovoltaic power output, total system load, and the median percentile of carbon trading prices.

[0064] The above prediction results are written into a shared database for subsequent steps to read the interval prediction results of new energy sources and loads to calculate trigger indicators and to simultaneously read the median prediction values ​​of new energy sources and loads as a benchmark scenario. The median prediction value of carbon prices is also read to accurately calculate the carbon trading costs in the scheduling model.

[0065] The aforementioned shared database is implemented using time-series relational database tables to ensure data traceability.

[0066] Furthermore, in this embodiment of the application, step S3 dynamically calculates the triggering index characterizing the system uncertainty based on the interval prediction result, and performs optimized scheduling according to the triggering index. Specific steps include S301-S303: S301: Query the shared database to identify and lock the complete forecast batch that was most recently successfully published by the forecast service and whose forecast start time is no earlier than the current time. Each forecast batch should contain a unique publication identifier and forecast effective start time to ensure that the forecast data on which subsequent calculations depend has the highest timeliness and consistency, providing a reliable source of information for real-time risk assessment.

[0067] Based on the current system time, calculate the start time of the corresponding first complete scheduling period in the future, and extract key values ​​from the locked prediction batch data, including: The low quantile, middle quantile, and high quantile forecasts for new energy sources corresponding to the current target time period; The quantile of the system load forecast corresponding to the current target time period and the quantile of the system load forecast corresponding to the previous time period.

[0068] Simultaneously, the extracted values ​​are validated before calculation. This includes checking for null values ​​and verifying whether they meet the predicted low quantile values. Median predicted value and median predicted value The relationship between high quantile predicted values; check whether the middle quantile predicted value of new energy output is greater than a very small positive number to avoid division by zero error.

[0069] If the verification fails, an error log is recorded, and the valid indicator value from the previous period is used as a replacement to ensure that the evaluation process can continue to operate stably in the event of data anomalies and maintain the overall availability of the scheduling system.

[0070] S302: First, calculate the width of the new energy output prediction interval, that is, the high quantile prediction value minus the low quantile prediction value; The next step is to take the median predicted value of new energy output as the benchmark. To prevent the median predicted value from being too small and causing abnormal amplification of components, a lower limit value is set. The actual benchmark value used is the larger of the median predicted value and the preset lower limit value. Finally, dividing the prediction interval width by this benchmark value yields the uncertainty component of the new energy prediction. This value is a dimensionless proportion that directly measures the degree of ambiguity in the prediction information.

[0071] The instantaneous load change rate component is calculated as the difference between the quantile of the load forecast for the target period and the quantile of the load forecast for the previous period, and the absolute value is taken; the ratio of the above absolute value to the load data sampling time interval is the rate of change of load power over time.

[0072] S3021: To make the instantaneous load change rate component comparable to the renewable energy component in terms of magnitude, a normalization operation is performed, ultimately yielding the trigger index for system uncertainty: in, These are weighting coefficients. These represent the high quantile, low quantile, and middle quantile predicted values ​​of new energy output for the k-th time period, respectively. , These are the predicted median quantile values ​​of the load for the k-th and (k-1)-th time periods, respectively. This is the sampling time interval for the load data.

[0073] It should be noted that this section follows the typical regional power system in step S103, and the weighting coefficients are used in the trigger index calculation. =0.7, =0.3, the normalized reference value for the load change rate is taken as the typical load change rate of the system being 10MW / h. In the same scheduling period k, the latest interval forecast data is obtained from step S200: Total output of new energy sources (wind power + solar power): Low quantile =20 MW, median =30MW, high quantile =40MW; System load: Median of the previous period =100MW, median quantile for this period =105MW.

[0074] First, calculate the uncertainty component of new energy forecasting: Secondly, to calculate the instantaneous load change rate component, it needs to be normalized first: actual load change rate. Divide by the reference rate of change and multiply by the weighting factor to obtain The comprehensive triggering indicators are: This value is used for the judgment in the next step.

[0075] S3022: The trigger judgment logic is based on the calculated comprehensive uncertainty trigger index. Based on preset threshold rules, it determines whether to issue an execution instruction to the optimization scheduling system.

[0076] Define the main trigger threshold The main trigger threshold range was determined through large-scale simulation backtracking of long-term historical operating data. At the same time, redefine the warning threshold within the range of the main trigger threshold. (0.10) and emergency trigger threshold (0.20), used for graded early warning and triggering.

[0077] At the end of each assessment period, the calculated comprehensive uncertainty trigger index will be used. With the main trigger threshold Compare; when ( If the value is 0.05, then proceed to the next step of image stabilization judgment; Otherwise, the system is determined to remain in its current state, optimization is not triggered, and the system continues to execute the current scheduling plan and waits for the next evaluation cycle.

[0078] To avoid frequent optimization triggers due to data noise or instantaneous jumps in predicted values, the system introduces an anti-jitter mechanism. One possible implementation is that the system maintains a fixed-length sliding time window, when... ( When the value is 0.05, this assessment will be recorded as an event exceeding the standard. Determine whether the number of out-of-limit events within the current sliding window (including the current evaluation period) has reached the preset minimum trigger count (1 time). If it has, it is considered a valid trigger; otherwise, it is not triggered.

[0079] Another alternative implementation is that when ( When the value is 0.05, the system does not trigger immediately but starts a delay timer. During the delay period, the evaluation continues according to the normal cycle. Value. If before the timer expires, If the value remains above the threshold, optimization will be triggered when the timer times out; otherwise, optimization will be triggered before the timer times out. If the value falls below the threshold, the current timer will be canceled and no trigger will be given.

[0080] It should be noted that, if The system is in a stable state and no commands are triggered; like The system enters an early warning state. In this state, the system records detailed logs and notifies the dispatcher, but optimization is not immediately triggered. If the early warning state persists for more than three consecutive evaluation cycles, it is upgraded to a triggered state. like The system enters an emergency state and immediately triggers optimization (the anti-shake judgment can be skipped).

[0081] S3023: Furthermore, when the triggering conditions are met, the system will generate a structured optimized triggering instruction event, in which the triggering conditions have been met through debouncing judgment and hierarchical triggering judgment.

[0082] The event includes the message type, the timestamp of the triggered message, the value of the triggered indicator, the associated prediction batch ID, the start time of the recommended adjustment operation, and other contextual information, namely the current load of the system and the output prediction of the wind turbine.

[0083] The generated optimized trigger command event is sent through a message queue to ensure that the event is sent successfully at least once to avoid missed triggers; if the triggering conditions are not met, the system will not generate a trigger event.

[0084] The system uses a message queue to send the generated optimized trigger command events, ensuring that the event is successfully sent at least once to prevent missed triggers; if the triggering conditions are not met, the system will not generate a trigger event.

[0085] The system will continue to execute the currently issued scheduling plan until the next evaluation cycle arrives. The result of each evaluation (whether it is triggered) will be saved to the database, and the timestamp, Ψ(k) value, values ​​of each component, prediction batch ID, trigger decision result, etc. will be recorded.

[0086] S303: When the system determines that a new round of rolling optimization needs to be started, it will immediately begin the process of dynamically setting the time window for elastic optimization.

[0087] The system performs calculations according to pre-set boundary parameters and control parameters. The boundary parameters include the maximum optimization window length and the minimum optimization window length, which represent the upper and lower limits of the optimization planning time, respectively.

[0088] The maximum optimization window length determines the range of long-term coordinated optimization that can be carried out when uncertainty is low, and it is usually matched with the remaining period of the day-ahead plan or the main adjustment cycle; the minimum optimization window length ensures that the optimization model has a sufficient time span to arrange dynamic processes such as the start-up, shutdown and ramp-up of key units under any triggering conditions, ensuring the basic feasibility of the solution. The control parameter is the window shrinkage sensitivity coefficient, which is defined as how many time periods the window length should be reduced when the comprehensive uncertainty trigger index exceeds the threshold. The coefficient needs to be set through simulation debugging. In typical high uncertainty scenarios, the window can quickly shrink to near the minimum allowable optimized window length.

[0089] The dynamic calculation of the window length involves the following steps, each designed to accurately translate risk assessment into resource allocation: A. Obtain the most recently calculated effective comprehensive uncertainty trigger index value as the quantitative input for the current risk assessment; B. Calculate the difference between this indicator and the main trigger threshold (the excess amount). The purpose is to indicate how much the current risk exceeds the normal tolerance level. This is the basis for making the decision on the adjustment window. C. Multiply the excess by the window reduction sensitivity coefficient, then round up to obtain the theoretically reduced number of time periods. This transforms the severity of the risk into a specific amount of time reduction, establishing a direct link between risk and decision-making resources. D. Subtract the theoretical reduction amount from the maximum allowable optimization window length to obtain the initial window length, thus initially realizing the reverse adjustment of the field of view based on the risk level.

[0090] E. Implement boundary protection for the initial window length: if it is less than the minimum allowable value, use the minimum value; if it is greater than the maximum allowable value, use the maximum value; otherwise, use the specified value.

[0091] This step ensures that the final window length remains within an acceptable range for the system, preventing invalid or overloaded optimizations due to extreme calculation values.

[0092] F. The finalized optimization window length is injected as an important parameter into the optimization trigger command event.

[0093] Furthermore, in this embodiment of the application, step S4 updates the energy storage unit state based on actual operating data and corrects the dynamic characteristic model, specifically including steps S401-S404: S401: Upon receiving a trigger event, first parse the event content to obtain the start time and window length of this optimization; then, based on the current system topology and the status of adjustable resources, define the decision variables for each scheduling period within the optimization window.

[0094] The decision variables include: the active power output setpoints of all online conventional thermal power units; the charging and discharging power of each electrochemical energy storage power station (as two independent non-negative variables), and the associated binary variables representing their charging and discharging states; the planned curtailment power values ​​of each wind power and photovoltaic power station; and, if an AC power flow model is used, the voltage amplitude and phase angle variables of each node; all decision variables must have their time index and physical unit identifier clearly defined.

[0095] S402: Construct the objective function for optimizing the scheduling.

[0096] S4021: The optimization aims to minimize the total expected operating cost of the system within the rolling window. The objective function consists of a linear superposition of three costs: conventional power generation fuel cost, carbon trading cost, and dynamic energy storage loss cost. The conventional power generation fuel cost includes calculating the fuel consumption cost of all thermal power units in each time period within the optimization window; the system sums this cost for all units in all time periods.

[0097] The carbon trading cost includes the fee required to calculate the carbon emissions generated by thermal power units. For each unit in each time period, the carbon emissions are first calculated based on its output and carbon emission intensity coefficient for the current time period. This carbon emission is then multiplied by the predicted quantile value of the carbon price for the corresponding time period to obtain the carbon cost for that unit in that time period. The carbon costs for all units in all time periods are then summed.

[0098] Among them, the dynamic loss cost of energy storage includes the conversion of the operational health losses of energy storage equipment into economic costs.

[0099] For each electrochemical energy storage unit j in each time period k, its loss cost is calculated by the formula: In the formula, For the dynamic loss cost of energy storage, This is the cost conversion factor. , These are the charging current and discharging current of energy storage unit j in time period k, respectively, which are obtained by dividing the corresponding charging / discharging power by the rated DC voltage of the energy storage system. The length of the scheduling period; In the calculation of loss costs, This represents the dynamic equivalent internal resistance value of energy storage unit j during time period k. It is used as the core parameter. During optimization, the dynamic characteristic model service is called, and the state of charge of energy storage unit j at the end of the previous time period, the average temperature prediction value of the current time period, and the cumulative cycle ampere-hours are input to obtain the calculated dynamic equivalent internal resistance value in real time.

[0100] It should be noted that this section refers to the typical regional power system in step S103, and the energy storage loss cost conversion factor... =0.01 yuan / Wh (i.e., 10 yuan / kWh). In the optimized scheduling triggered by the above, assume that the planned discharge power of energy storage unit j in this time period t is 2MW, and the charging power is zero. Then the discharge current The charging current is 2000A. The value is 0, and the resistance during this period is calculated previously. =0.0098945Ω, time period length 0.25h; calculate instantaneous power loss: Energy loss during the period: The final cost is: The energy storage loss cost due to discharge during this period is approximately 99 yuan. This cost will be included in the optimization objective function to guide scheduling decisions and avoid high-power operation under high internal resistance conditions.

[0101] S4022: The optimization problem must satisfy both physical and operational constraints, which together constitute the feasible region of the solution. Specifically, these constraints include the following: A. System balance and network constraints include node power balance constraints, i.e., the net injected power of any node at any time period must be equal to the outflow power, and network power flow constraints; at the same time, the active power flow of all transmission lines and transformers must be within their safe transmission limits. B. Conventional unit operation constraints include: the output of each thermal power unit must be between the minimum and maximum output allowed by technology; the output variation between adjacent time periods must not exceed the unit's inherent ramping capacity limit; and the start-up and shutdown of the unit must meet the requirements of minimum continuous operating time and minimum continuous shutdown time. C. The constraints on the operation of new energy sources include that the planned power curtailment of each new energy power station in any given time period must be non-negative and must not exceed the median value of the predicted new energy output in that time period, that is, the power curtailment is capped at the theoretical maximum power generation capacity. D. The establishment of dynamic operation constraints for energy storage depends on the dynamic characteristic model constructed in step S1. Specifically, it includes: (a) The dynamic internal resistance equality constraint includes the dynamic internal resistance calculation model as an equality constraint introduced into the optimization scheduling; (b) The state update equation constraint includes the state update equation as an equality constraint. The state update equation describes the mathematical relationship between the state of charge of the energy storage unit at the end of the current scheduling period, the state of charge at the end of the previous scheduling period, and the charging and discharging power of the current period. To further explain, the state of charge at the end of the current period is equal to the state of charge at the end of the previous period, plus the difference between the net charge and net discharge during this period, and then divided by the rated capacity. This constraint ensures the continuity of the energy storage state.

[0102] The net charge received during this period is the product of charging power, efficiency, and time; the net discharge during this period is the product of discharging power, efficiency, and time.

[0103] (c) The dynamic power inequality constraint includes using the maximum allowable charging power and maximum allowable discharging power of the energy storage unit in the current state, calculated based on the dynamic characteristic model, as the inequality constraint for power decision during the current scheduling period; Furthermore, when optimizing the model construction, for each energy storage unit in the current time period, based on its state of charge at the end of the previous time period and the current dynamic internal resistance calculated by the dynamic characteristic model, its maximum sustainable charging current and discharging current within the allowable voltage range are solved in real time and converted into power limits.

[0104] These two real-time changing power limits serve as inequality constraints, strictly limiting the charging and discharging power of the energy storage unit to the corresponding upper limits during the current period.

[0105] (d) The state of charge boundary constraint includes the requirement that the state of charge of the energy storage unit must always remain within the safe upper and lower limits specified by the manufacturer throughout the entire optimization window.

[0106] S4023: A hierarchical strategy combining an outer evolutionary algorithm and an inner numerical optimization is used to solve the current optimal scheduling problem.

[0107] An improved NSGA-III algorithm is selected for the outer layer to handle integer combinations and multi-objectives. The algorithm maintains a fixed-size population by adaptively adjusting the crossover and mutation probabilities. Each individual represents a complete candidate scheduling scheme through a specific encoding scheme, and the decision variables for all time periods within the optimization window are divided into integer variables (such as unit start-up and shutdown, and energy storage charging and discharging status) and continuous variables (such as power setpoint). In this context, integer variables are encoded using binary strings, with each bit representing the state of a device within a given time period; continuous variables are not directly encoded, but are solved by combining integer variables carried by the outer layer and then optimizing the solution in the inner layer. The algorithm iteratively evolves the population through operations such as selection, crossover, and mutation. The crossover and mutation probabilities are dynamically adjusted based on the number of iterations and the measured value of population diversity. Through non-dominated sorting and reference point mechanisms, the algorithm makes the population approach the Pareto front of the two objectives of minimizing economic efficiency and carbon emissions.

[0108] The inner layer is solved by continuous nonlinear programming; for each individual in the outer population, that is, a set of determined start-stop state combinations, namely the start-stop state of each thermal power unit in each time period and the operating mode of each energy storage unit in each time period (in charging state, discharging state or idle state), the optimization problem is simplified to a nonlinear programming problem with the continuous power value of each unit and energy storage as variables. The original dual interior-point solver is invoked to solve the subproblem, yielding the optimal power allocation and the corresponding minimum sub-cost under this specific start-stop combination.

[0109] When the evolutionary algorithm reaches the termination condition, i.e., the fitness of the population does not improve significantly over multiple generations, it outputs a set of Pareto optimal solutions. Each solution corresponds to a complete scheduling scheme and its economic cost and carbon emissions.

[0110] S403: The system starts from the known current actual state of the energy storage unit, that is, the state at the end of time period k-1. For each future time period within the optimization window, the planned charging and discharging power of this time period is substituted into the state update equation to calculate the theoretical state of charge at the end of the current time period. By further utilizing the theoretical state of charge and other inputs required by the dynamic characteristic model, the theoretical internal resistance for the next time period is calculated through the dynamic internal resistance model. By calculating the theoretical state of charge and theoretical internal resistance, the initial values ​​for the next optimization cycle are obtained and temporarily stored as the predicted initial state values ​​for the corresponding energy storage unit in the next rolling optimization calculation. This process ensures that the optimization model always starts a new optimization from a coherent state consistent with its previous decision.

[0111] S404: Correcting the dynamic characteristic model. Specifically, this includes steps S4041-S4043: S4041: The system uses the actual average charging and discharging power and charging and discharging efficiency parameters of the scheduling period, and calculates the predicted value of the state of charge at the end of the period according to the state update equation constraint. The predicted value is compared with the measured value of the state of charge at the end of the period, and the absolute deviation between the two is calculated. When the deviation exceeds one percent of the rated capacity, the system will use the collected measured value as the correction value of the end-of-charge state for that period, overriding the original predicted value, to ensure the accuracy of the energy storage state. Meanwhile, the corrected state of charge value will serve as the starting reference for the state estimation of the next scheduling cycle.

[0112] S4042: The recursive least squares estimation algorithm is used to continuously identify and update the dynamic characteristic parameters online. For each new data sample, the following operations are performed sequentially: Using the current estimated values ​​of dynamic characteristic parameters and input features, calculate the predicted value of the internal resistance for the current time period; Calculate the error between the predicted value and the actual reference value; Calculate the gain vector for this parameter update based on the current covariance matrix, input eigenvector, and forgetting factor; Using the calculated gain vector and prediction error, the current model parameter vector is slightly modified to obtain the updated parameter vector. Using the same forgetting factor and gain vector, the covariance matrix is ​​updated to prepare for processing the next data sample. Simultaneously, the cumulative cycle throughput ampere-hours of the energy storage unit are updated based on the actual charging and discharging current and time during this period.

[0113] S4043: Synchronously write the updated dynamic characteristic model parameter vector to the shared database. All subsequent prediction calls for internal resistance will use this latest parameter.

[0114] The corrected state of charge value of the energy storage unit, the updated cumulative ampere-hours, and other latest status information are written into the shared database, which also serves as the initial state for the next round of triggering evaluation and optimization calculations.

[0115] After completing all the above update and archiving operations, the previous scheduling instruction generation and execution cycle officially ends. Simultaneously, based on the latest model parameters and device status, the system automatically enters the waiting or ready state for the next scheduling cycle.

[0116] Example 3 is the third embodiment of the present invention, which differs from the previous two embodiments in that: This embodiment also provides an electronic device, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the multi-energy power system energy storage coordinated scheduling method proposed in the above embodiment.

[0117] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, it implements the multi-energy power system energy storage coordinated scheduling method proposed in the above embodiments.

[0118] The storage medium proposed in this embodiment and the method for coordinated scheduling of energy storage in multi-energy power systems proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0119] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0120] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for coordinated dispatch of energy storage in multi-energy power systems, characterized in that: include, A dynamic characteristic model reflecting the aging and loss evolution of energy storage units is established, which is expressed as a dynamic function of historical accumulated amount, current state of charge and external environmental parameters. By integrating and analyzing historical meteorological sequences, load sequences, and carbon price sequences, the system outputs interval forecasts for renewable energy output, system load, and carbon price during future scheduling periods. The fusion analysis includes constructing a sequence prediction model containing an encoder and a decoder. The decoder uses the historical meteorological hidden state, load hidden state and carbon price hidden state output by the encoder to perform meteorological condition attention retrieval, load self-attention retrieval and carbon price trend attention retrieval at each prediction execution time, and generates meteorological driving context vector, load trend context vector and carbon price trend context vector respectively. The three context vectors are concatenated and used together with the prediction output as the input of the decoder to update the internal state of the decoder. Based on the interval prediction results, trigger indicators representing system uncertainty are dynamically calculated, and optimized scheduling is performed according to the trigger indicators. At each decision point, obtain the interval forecast results for the current and near-future periods; Based on the preset high quantile and low quantile forecast values, the forecast uncertainty width of new energy output in the first period is calculated, and then normalized by dividing by the median quantile forecast value of the corresponding period to obtain the first uncertainty component. Calculate the instantaneous rate of change of the system load at the current moment, as the second uncertainty component; The first uncertainty component and the second uncertainty component are weighted and summed to obtain a comprehensive trigger index; The triggering indicator is compared with a preset triggering threshold. If the triggering indicator is greater than the triggering threshold, a rolling optimization scheduling is initiated after determination. If the triggering indicator is less than or equal to the triggering threshold, the existing scheduling plan is maintained. The objective function for optimized scheduling includes the dynamic loss cost of energy storage calculated in real time based on the dynamic characteristic model, and the dynamic characteristic model is used as the core constraint for optimized scheduling. For each electrochemical energy storage unit j in each time period k, its loss cost is calculated by the formula: In the formula, For the dynamic loss cost of energy storage, This is the cost conversion factor. , These are the charging current and discharging current of energy storage unit j in time period k, respectively, which are obtained by dividing the corresponding charging / discharging power by the rated DC voltage of the energy storage system. The length of the scheduling period; In the calculation of loss costs, This represents the dynamic equivalent internal resistance value of energy storage unit j during time period k. It is used as the core parameter. During optimization, the dynamic characteristic model service is called. The state of charge of energy storage unit j at the end of the previous time period, the average temperature prediction value of the current time period, and the cumulative cycle ampere-hours are input to obtain the calculated dynamic equivalent internal resistance value in real time. The energy storage unit status is updated based on actual operating data, and the dynamic characteristic model is corrected.

2. The multi-energy power system energy storage coordinated dispatch method as described in claim 1, characterized in that: The establishment of a dynamic characteristic model reflecting the aging and loss evolution of the energy storage unit includes collecting historical operating data of the energy storage unit within a set operating cycle. The historical operating data includes charging and discharging current sequence, terminal voltage sequence, ambient temperature sequence, and initial state of charge. Using the historical operating data as input, an equivalent internal resistance dynamic model characterizing the evolution of the internal health state of the energy storage unit is established, wherein the equivalent internal resistance is modeled as a continuous function of the cumulative throughput of the energy storage unit, the current state of charge, and the current operating temperature. A dynamic model of equivalent internal resistance characterizing the evolution of the internal health state of an energy storage unit is established. Specifically, the equivalent internal resistance is modeled as a continuous function of the cumulative throughput of the energy storage unit, the current state of charge, and the current operating temperature. in, This represents the equivalent series resistance in the k-th time interval of the model output. This represents the state of charge of the energy storage unit at the beginning of the k-th time period. This represents the normalized temperature at time k. It represents the cumulative cyclic throughput ampere-hours starting from a certain reference state point, and is updated by calculating the sum of the absolute charge and discharge ampere-hours of the current period and the cumulative value of the previous period; , Indicates the reference internal resistance. This represents the coefficient of influence of SOC on internal resistance. and Together, they represent the nonlinear relationship between temperature and internal resistance. This represents the coefficient of influence of cyclic aging on the increase of internal resistance.

3. The multi-energy power system energy storage coordinated dispatch method as described in claim 2, characterized in that: The output of the interval prediction results of new energy output, system load and carbon price within the future scheduling period includes the construction of a sequence prediction model containing an encoder and a decoder; The encoder includes a parallel first sub-coding network, a second sub-coding network, and a third sub-coding network, which are used to process historical meteorological feature time series, historical system load time series, and carbon price series, respectively, for feature extraction. The attention weights corresponding to each feature vector in the above sequence are calculated in the decoder, and the corresponding context vectors are generated. The corresponding context vectors are concatenated and fused and then passed through a fully connected output layer to output the quantile prediction values ​​for future continuous scheduling periods. The quantile predictions include high quantile predictions, middle quantile predictions, and low quantile predictions, which together constitute the interval prediction results.

4. The multi-energy power system energy storage coordinated dispatch method as described in claim 3, characterized in that: The step of performing optimization scheduling based on the triggering indicator also includes dynamically setting the length of the look-ahead time window for this rolling optimization according to the value of the triggering indicator. Preset the maximum and minimum optimized window lengths; Calculate the difference between the trigger indicator and the trigger threshold to determine the window reduction amount; The difference between the maximum optimized window length and the window reduction amount is used as the initial time window length; If the initial time window length is less than the minimum optimization window length, then the time window length for this rolling optimization is the minimum optimization window length; otherwise, the initial time window length is used.

5. The multi-energy power system energy storage coordinated dispatch method as described in claim 4, characterized in that: The step of using the dynamic characteristic model as the core constraint for optimal scheduling includes introducing the state update equation in the dynamic characteristic model as an equality constraint into the optimal scheduling. The state update equation defines the mathematical relationship between the state of the energy storage unit at the end of the current scheduling period, the state at the end of the previous scheduling period, and the charging and discharging power in the current period. Meanwhile, the maximum allowable charging power and maximum allowable discharging power of the energy storage unit in the current state, calculated based on the dynamic characteristic model, will be used as the power inequality constraint of the current energy storage unit in the current scheduling period.

6. The multi-energy power system energy storage coordinated dispatch method as described in claim 5, characterized in that: The correction of the dynamic characteristic model includes collecting actual operating data of the energy storage unit after each scheduling period ends, including actual charging and discharging power and estimated state of charge. Using the actual operating data, the recursive least squares algorithm is employed to identify and update the function parameters in the dynamic characteristic model online. The updated function parameters are applied to the dynamic characteristic model of the next scheduling cycle to achieve closed-loop adaptive correction of the model.

7. The multi-energy power system energy storage coordinated dispatch method as described in claim 6, characterized in that, It also includes obtaining the scheduling instructions generated by the optimized scheduling; The operation of the energy storage unit in the multi-energy power system is controlled based on the dispatch instructions; during the control process, the state of the energy storage unit is updated and the dynamic characteristic model is corrected based on the actual operating data.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the multi-energy power system energy storage coordinated scheduling method as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the multi-energy power system energy storage coordinated scheduling method as described in any one of claims 1 to 7.

Images