Intelligent dispatching method for integrated virtual power plant

Abstract

The present invention discloses an intelligent scheduling method for a comprehensive virtual power plant, which relates to the technical field of intelligent scheduling of power plants. By deploying a high-concurrency acquisition module on the edge side, redundant filtering and unified timing calibration are performed on multi-node sensor data such as cold and hot energy stations and energy storage equipment to generate high-quality timing fusion sequences and screen schedulable resources. An error dynamic evaluation mechanism is used to achieve online model correction. Aiming at economic, reliability, and environmental protection goals, an improved particle swarm algorithm is used to generate a multi-stage scheduling strategy, and the optimal instructions are output in combination with equipment constraints. Deviations are evaluated through priority control and real-time monitoring indicators to trigger rolling corrections and backup strategies, achieve equipment-level fault risk quantification and self-healing control, quickly isolate anomalies, and activate redundant resources. Through data-prediction-scheduling-execution closed-loop collaboration, resource utilization, scheduling accuracy, and system fault tolerance are significantly improved, and the scheduling capability of heterogeneous resources is enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent dispatching of power plants, and in particular to an intelligent dispatching method for an integrated virtual power plant. Background Art

[0002] Against the backdrop of today's energy transformation and deepening reform of the power system, with the rapid development of distributed renewable energy (such as photovoltaics and wind power) and the maturing of energy storage and adjustable load technologies, the traditional power grid, primarily based on centralized power sources, is gradually evolving into a multi-source integrated energy network. In application scenarios such as industrial parks, large commercial complexes, and urban energy microgrids, multiple energy types (electricity, cooling, heating, etc.) and numerous decentralized devices (cooling and heating energy stations, distributed photovoltaics, energy storage devices, adjustable loads, etc.) coexist, each with distinct operating characteristics and dispatchability features. The concept of a virtual power plant (VPP), based on the Internet of Things (IoT), intelligent forecasting, and dispatch optimization technologies, aggregates data and enables coordinated management of these heterogeneous resources, enabling deep mining and unified dispatch of distributed energy resources. This allows for the enhanced utilization and economic benefits of clean energy while meeting energy demand and ensuring system security. Driven by the continuous improvement of external power market mechanisms and the diversification of internal electricity demand, virtual power plants are becoming increasingly prominent in emerging power systems and have become a key tool in building a power grid with a high proportion of clean energy.

[0003] In the Chinese invention patent application publication number CN117439171A, an intelligent scheduling method, system and medium based on a virtual power plant are disclosed, which solve the shortcomings of the existing technology. The method includes the steps of obtaining historical electricity consumption data on the user side and predicting the future electricity demand on the user side based on the historical electricity consumption data on the user side; the virtual power plant formulates a power generation plan based on the future electricity demand on the user side, and determines the power generation proportion of multiple power generation energy sources based on the future power generation demand; the virtual power plant supplies power to the user side according to the power generation plan, and at the same time, the virtual power plant monitors the total power generation of multiple power generation energy sources and the power demand on the user side. If the difference between the total power generation of multiple power generation energy sources and the power demand on the user side exceeds a threshold, the power generation proportion of multiple power generation energy sources is adjusted in real time to ensure the actual usage demand of the total power generation of multiple power generation energy sources.

[0004] However, in combination with the above existing technologies and actual application scenarios, the key technical issues that need to be solved in the large-scale coupling of multi-source heterogeneous devices and dynamic operating environments are:

[0005] How to balance real-time data fusion, high-precision prediction, and feasible and reliable scheduling strategy generation for each distributed resource in a highly concurrent, highly uncertain, multi-type energy network, and be able to perform rapid self-healing control in the event of equipment failure or sudden fluctuations. Specifically, when power fluctuations or anomalies occur in the operation of hot and cold energy stations, adjustable loads, energy storage, and photovoltaic systems due to weather changes, electricity price fluctuations, or equipment aging, if there is a lack of a unified data fusion mechanism and dynamic resource identification means, the system will find it difficult to grasp the actual operating conditions of the equipment in a timely manner, which will lead to a decrease in the accuracy of the prediction model and the inability to effectively execute the multi-objective scheduling strategy. At the same time, in the event of a fault, such as an energy storage failure, a grid transient, or a large output fluctuation of a distributed power source, if immediate diagnosis cannot be performed and a quick switch to a backup scheduling plan cannot be made, it is easy to cause a local supply gap or energy waste, and even induce a wider range of energy supply security risks.

[0006] To this end, the present invention provides an intelligent scheduling method for a comprehensive virtual power plant. Summary of the Invention

[0007] (1) Technical problems solved

[0008] In response to the deficiencies of the prior art, the present invention provides an intelligent scheduling method for a comprehensive virtual power plant. By deploying a high-concurrency acquisition module on the edge side, de-redundant filtering and unified timing calibration are performed on multi-node sensor data such as cold and hot energy stations and energy storage equipment to generate high-quality timing fusion sequences and screen schedulable resources. The model is corrected online through a dynamic error evaluation mechanism. Aiming at economic, reliability and environmental protection goals, an improved particle swarm algorithm is used to generate a multi-stage scheduling strategy, and the optimal instructions are output in combination with equipment constraints. Deviations are evaluated through priority control and real-time monitoring indicators to trigger rolling corrections and backup strategies, achieve equipment-level fault risk quantification and self-healing control, quickly isolate anomalies and start redundant resources. Through data-prediction-scheduling-execution closed-loop collaboration, resource utilization, scheduling accuracy and system fault tolerance are significantly improved, thereby solving the technical problems recorded in the background technology.

[0009] Within the overall framework of the virtual power plant, a systematic solution covering data collection and fusion, predictive modeling, scheduling optimization and closed-loop collaboration, as well as fault diagnosis and self-healing control is constructed to ensure that multiple energy resources can be managed collaboratively under complex working conditions, taking into account economic, reliability and environmental protection goals, and improving the ability to respond promptly to sudden failures or operational abnormalities.

[0010] (2) Technical solution

[0011] To achieve the above objectives, the present invention is implemented through the following technical solutions: an intelligent scheduling method for a comprehensive virtual power plant, comprising:

[0012] When multiple sensor nodes report raw data When the sensor data of distributed power equipment is aggregated by deploying the high-concurrency acquisition module on the edge side, the pre-processed data is generated after the message queue and timestamp pre-processing. , unify the timing calibration and generate cleaned data , through dynamic weights and attenuation factor The exponential weighted fusion algorithm generates time series fusion sequences , combined with equipment conditions and environmental data Constructing comprehensive feature vectors , using the discriminant function and the judgment threshold Generate resource status ;

[0013] When obtaining the time series fusion sequence and resource status When the initial eigenvector is transformed by wavelet Perform multi-scale decomposition to generate prediction input vector , generate load forecast after parallel training , power generation forecast and energy storage forecasting , using the error function Dynamically evaluate the prediction deviation and trigger online parameter correction when the error exceeds the error threshold;

[0014] When the prediction results and resource status When input, the objective function is optimized through multi-objective Integrate constraint objectives, use improved particle swarm optimization to perform multi-stage optimization on the scheduling decision vector X, build power balance hard constraints based on equipment rated parameters, and use adaptive weight adjustment mechanism to generate the optimal scheduling solution , and generate equipment execution plans, evaluate the effectiveness of the plans through economic-reliability-environmental indicators, and achieve multi-objective dynamic trade-offs and coupling with global-local scheduling strategies;

[0015] When the optimal scheduling plan When sent to the field equipment, the instructions are converted into equipment-level operation sequences through the field energy management system, and the priority allocation mechanism is used to synchronously send control instructions and collect monitoring indicator vectors. , build performance measurement function Evaluate execution deviations, when the deviations accumulate and exceed the performance threshold The rolling correction process is triggered when the local time window is optimized to generate the local optimal solution. ;

[0016] When monitoring indicator vector Fusion Sequence with Time Series When an exception occurs, build a fault judgment function , identify device-level failure modes, when the fault decision function Exceeding the fault threshold When the self-healing control is triggered, the backup scheduling plan is used to start the redundant equipment and adjust the load distribution strategy. After the repair, the safety threshold is Restore device scheduling permissions.

[0017] Preferably, sensors are deployed at the distributed power equipment, and each node is The original data reported is recorded as , original data With the corresponding timestamp Initial aggregation will be performed on the edge side;

[0018] For the original data Perform filtering and de-redundancy to obtain pre-processed data after preliminary screening ;Synchronous acquisition and unified timing alignment of different nodes;

[0019] Preferably, the preprocessed data Import into the central database and obtain cleaned data after preprocessing , weighted fusion of data from multiple nodes to generate a unified single-point fusion value at the moment :

[0020] Different moments Single point fusion value Arrange in sequence to form a time series fusion sequence ;

[0021] Preferably, the time series is fused into a sequence Integrate with equipment operating condition records and meteorological data to form a comprehensive feature vector :For the comprehensive feature vector Perform analysis to determine whether key resources are in a schedulable or abnormal state and define the judgment function :

[0022]

[0023] in is the judgment threshold, Used to comprehensively evaluate factors such as resource health and adjustability;

[0024] Each distributed power equipment is The schedulable status and operation mode are uniformly recorded as resource status Preferably, from the time series fusion sequence and resource status In the example, key data is selected and the initial feature vector is constructed by combining external input. , the initial eigenvector Perform wavelet transform on some time series components in the dataset to extract high-frequency disturbances and low-frequency trends, and decompose the original time series into several multi-scale components. Combine the multi-scale components obtained by wavelet transform with other key features to form a prediction input vector. ;

[0025] Preferably, after obtaining the predicted input vector Finally, short-term prediction is achieved based on deep neural networks, and ultra-short-term prediction is achieved using lightweight time series models, outputting key prediction quantities within different prediction time periods respectively;

[0026] The neural network model is trained and cross-validated using a rolling window approach, and the scheduling prediction model is obtained by dynamically adjusting the learnable parameters.

[0027] Preferably, when the scheduling prediction model is put into online operation, the actual measurement values are obtained at a high frequency. , and the model output Make comparisons; customize the error function The prediction accuracy is dynamically measured as follows:

[0028]

[0029] Where: is the target index, , is the number of prediction objects that need to be evaluated simultaneously; ,in: Indicates the The predicted value of a target, is the corresponding actual observation value; At the start of the current evaluation The relative time after, the value range ; For the corresponding The derivative error weight of each predicted target; Indicates the duration of the assessment; For the Time decay factor for each target;

[0030] According to the error function If the error continues to be higher than the preset error threshold, the online correction program will be triggered;

[0031] The latest prediction input vector collected in real time and target prediction value Incrementally incorporate the model training set, update the model parameters through small-scale iterative training, and dynamically adjust the weight matrix , bias vector and dynamic weight coefficients Wait, complete the online calibration;

[0032] Preferably, based on the short-term / ultra-short-term forecast results of key forecast quantities, combined with resource status , construct the multi-objective optimization objective function Based on the hard constraints of the scheduling decision vector X within the equipment's rated operating range, and with reference to the prediction range and time resolution, each discrete scheduling period should ensure power balance and safety margin.

[0033] Preferably, the particle swarm position represents the scheduling decision vector X, and the fitness is defined as the multi-objective optimization objective function The opposite number or other methods are used to achieve minimization; the system state in the scheduling period is used as the state space, the scheduling decision vector X is used as the action space, and the multi-objective optimization objective function is used. The corresponding rewards or penalties are used to learn the strategy; the algorithm initialization parameters are adjusted based on the device topology information, prediction duration, and time granularity;

[0034] The multi-objective optimization solution process is achieved by monitoring the multi-objective optimization objective function The convergence is determined by the iterative changes of the feasible solution and the satisfaction of the feasible solution constraints. If it cannot be fully converged within the preset number of iterations or time, the current optimal feasible solution is taken and the optimal scheduling plan is finally output. ;

[0035] The optimal scheduling solution will be output Map to the actual device level, format the scheduling strategy and output the optimal scheduling solution Decompose each performance in the objective function;

[0036] Preferably, according to the optimal scheduling plan , respectively extract the operating power, start-stop status and charge-discharge plan of distributed power equipment in each time period, and convert them into specific execution instructions for the equipment; the on-site energy management system matches the execution points one by one according to the corresponding resource identifiers;

[0037] In conjunction with the virtual power plant cloud platform, a priority order is established for the operating instructions of each type of equipment. The instructions are issued synchronously at the same time or within the same scheduling cycle. After receiving and executing the instructions, the on-site energy management system transmits the actual operation information of the equipment back to the cloud platform, which together with the data obtained by the sensors forms a comprehensive operation monitoring source.

[0038] Preferably, the monitoring indicator vector is defined according to the prediction results , to compare the optimal scheduling scheme The deviation between the execution effect and the current running state, the performance measurement function is introduced , the formula is as follows:

[0039]

[0040] Where: , indicating that at time When , the current execution deviation or the difference measurement function between the execution result and the scheduling plan;

[0041] The scheduling instructions to be executed, For real-time monitoring of indicator vectors; , a function for measuring scheduling deviation;

[0042] middle, When , the deviation is amplified superlinearly, making the large error account for a higher proportion in the overall evaluation: When , it returns to the linear situation; middle, is the weight factor of the derivative term, and All are control parameters; middle, , represents the time decay factor; To evaluate the window length;

[0043] If the performance metric function At the preset performance threshold If a nonlinear jump in the energy storage / load equipment is detected, an early warning signal will be sent to the cloud platform, and a rolling correction process can be triggered at the same time;

[0044] Preferably, when the performance metric function The performance threshold is exceeded for several consecutive monitoring periods. , automatically triggering the scheduling re-evaluation process, integrating the latest data collected in real time, and forming the current state vector ;

[0045] Use improved particle swarm optimization or deep reinforcement learning to quickly reoptimize the scheduling plan within the local time window to a new local optimal solution Compared with the original optimal scheduling solution After comparison, if the recent operation deviation or cost / emission can be reduced, the original scheduling instructions can be partially replaced, and the original plan can be used for the rest of the period. The local optimal solution after rolling correction Re-enter the instruction issuance process and accept real-time monitoring and evaluation;

[0046] Preferably, combined with time fusion sequence and monitoring indicator vector Later it is recorded as comprehensive feedback data , define the following fault judgment function :

[0047]

[0048] in: Indicates at time A measure of the risk of failure; is the possible failure mode or the number of leaf nodes, represents the failure mode index; For failure mode The relevant weight coefficient, is the recognition function of the time series deep learning model for the input data, Indicates the failure mode The set of trainable parameters of Indicates at time Failure Mode The trigger duration or severity factor, represents the fault attenuation factor;

[0049] When the fault judgment function Exceeding the preset fault threshold When a fault occurs, it is determined that there is a high fault risk, the corresponding device is considered a faulty device, and a fault diagnosis instruction is issued;

[0050] Preferably, when the fault judgment function Exceeding the preset fault threshold When a rapid upward trend occurs, the self-healing control process will be immediately entered;

[0051] Re-evaluate the faulty equipment or associated loads and issue instructions to reduce or prohibit dispatching of the faulty equipment, or increase the corresponding safety factor; suspend or terminate dispatching instructions for the faulty equipment and invoke the following backup dispatch plan;

[0052] After receiving the fault diagnosis instruction, it immediately switches to the backup strategy to quickly adapt to the changes in load and energy supply structure after the fault. It can perform secondary optimization on non-faulty equipment in the local time window and generate a temporary scheduling plan;

[0053] After the self-healing process is initiated, perform maintenance or isolate the faulty equipment for a period of time and gradually observe whether the risk value given by the fault tree model drops significantly:

[0054] like ,in: The safety threshold indicates that the equipment failure has been eliminated or the risk is controllable, and the normal scheduling plan can be returned to.

[0055] If the equipment is repaired, it can be added to the scheduling scope again; if the repair is not completed, it will remain isolated to avoid affecting the overall system operation.

[0056] (3) Beneficial effects

[0057] The present invention provides an intelligent scheduling method for a comprehensive virtual power plant, which has the following beneficial effects:

[0058] By deploying high-concurrency collection and distributed computing mechanisms on the edge side, the sensor data of various types of equipment such as energy storage, hot and cold energy stations, adjustable load equipment and photovoltaic systems are processed in a unified manner; on the one hand, it reduces the data island phenomenon and reduces the load on the central server; on the other hand, through edge pre-aggregation and dynamic identification, the changing patterns of equipment working modes are mined to provide accurate and real-time operation portraits for subsequent predictions and scheduling.

[0059] By leveraging big data analysis and deep learning algorithms, short-term and ultra-short-term forecasts are made for load demand and distributed power equipment. At the same time, correlation analysis is conducted between the energy storage charge status and external factors such as price and weather. Not only is the coupling relationship between cold and hot energy stations, distributed power sources, and energy storage considered in feature extraction, but model parameters are also dynamically updated through an online correction mechanism, significantly reducing prediction deviations. Therefore, actual operating conditions can still be tracked well during peak and valley periods or when the weather changes suddenly.

[0060] Multiple requirements such as economy, reliability and environmental protection are integrated into a unified objective function, and with energy storage capacity constraints and equipment safety margin restrictions as constraints, an improved particle swarm or deep reinforcement learning algorithm is comprehensively adopted to generate a multi-dimensional scheduling solution. This scheduling solution can fine-tune the energy storage and load distribution at different time scales to ensure maximum benefits under fluctuations in electricity prices, loads and photovoltaics, reduce scheduling risks and further improve the utilization rate of clean energy.

[0061] By monitoring the real-time changes of key operating indicators (such as energy storage SOC, load tracking error, etc.) and evaluation functions, an adaptive dynamic scheduling system is constructed. Once execution deviation or supply and demand imbalance is detected, local re-optimization is triggered and instructions are updated and sent to each resource end in a timely manner. This process effectively reduces the impact of prediction errors or external interference, ensuring that the virtual power plant is in an efficient and controllable state in daily operation.

[0062] By combining fault trees with time-series deep learning, we can measure failure risk and trigger backup dispatch modes or rapidly isolate faulty equipment when risk exceeds a threshold. This allows us to balance revenue and environmental goals in standard scenarios, while also enabling smooth transitions through temporary solutions during faults or abnormal conditions, preventing single-point failures from causing widespread energy supply disruptions.

[0063] In summary, information accuracy is improved through multi-source data fusion and dynamic identification, perception of load and distributed power fluctuations is enhanced through short-term and ultra-short-term forecasts, flexible balance is made between economic and environmental demands through multi-objective optimization scheduling, real-time balance is maintained through closed-loop collaborative optimization, and system safety redundancy and rapid recovery are guaranteed with the help of fault diagnosis and self-healing mechanisms, which greatly improves the overall practicality and efficiency of virtual power plants. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 It is a flow chart of the intelligent scheduling method of the comprehensive virtual power plant of the present invention. DETAILED DESCRIPTION

[0065] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0066] See also Figure 1 The present invention provides an intelligent scheduling method for a comprehensive virtual power plant, including:

[0067] Step 1: When multi-source sensor nodes report raw data When the sensor data of distributed power equipment is aggregated by deploying the high-concurrency acquisition module on the edge side, the pre-processed data is generated after the message queue and timestamp pre-processing. , unify the timing calibration and generate cleaned data , through dynamic weights and attenuation factor The exponential weighted fusion algorithm generates time series fusion sequences , combined with equipment conditions and environmental data Constructing comprehensive feature vectors , using the discriminant function and the judgment threshold Generate resource status ;

[0068] The step 1 includes the following:

[0069] Step 101: High-concurrency acquisition and edge preprocessing

[0070] Sensors are deployed at hot and cold energy stations, adjustable load equipment, energy storage and photovoltaic systems, etc. Each node is The original data reported is recorded as , and its corresponding timestamp is , here Indicates the sensor node number, raw data With the corresponding timestamp Initial aggregation will be performed on the edge side;

[0071] For the original data Perform filtering and redundancy removal. When the same node reports multiple times in a very short period of time and the data has no obvious difference, it can be identified as duplicate data and discarded; obtain pre-processed data after preliminary screening ; Use message queue or high concurrent data bus to achieve synchronous acquisition and unified timing alignment of different nodes; timestamp After edge processing, it can be aligned to a unified reference time axis;

[0072] When in use, parallel data collection and edge preprocessing can significantly reduce the load on the central server and shorten network latency. Compared with the traditional centralized collection mode, the introduction of edge filtering and de-redundancy mechanisms ensures that massive sensor data has been preliminarily cleaned and formatted before uploading.

[0073] Step 102: Data cleaning and fusion algorithm implementation

[0074] Preprocess the data Import data into the central database, mark and process the obviously abnormal or missing data, and obtain the cleaned data after preprocessing If a node is offline or the data jump is too large, it will be recorded separately for subsequent operation and maintenance.

[0075] Use the following formula to perform weighted fusion on the data of multiple nodes to generate a unified single-point fusion value at the moment :

[0076]

[0077] in: is the number of nodes participating in the current fusion; For nodes The dynamic weight of In-range adjustments are made based on factors such as node historical accuracy or device status; Cleaned data for this node; The node attenuation factor is between 0 and 1 and is used to dynamically correct the data contribution according to the node aging or interference level; It is an exponential function used to smoothly amplify or attenuate the data impact of different nodes;

[0078] Different moments Single point fusion value Arrange in sequence to form a time series fusion sequence , recorded as ;

[0079] When in use, the custom weighted exponential decay function can dynamically reduce the weight of nodes with high noise or low credibility, further improve the accuracy of data fusion, and unify the output time series fusion sequence. It provides high-quality multi-source data indicators for subsequent resource identification and predictive analysis.

[0080] Step 103: Dynamic resource identification and state modeling

[0081] Fusion of time series Integrate with equipment operating condition records (such as energy storage SOC, cooling / heating load on status) and meteorological data to form a comprehensive feature vector :

[0082]

[0083] is the single point fusion value; Indicates equipment status (such as energy storage charge level, load start and stop flags, etc.); Represents environmental data (such as outdoor temperature, light intensity, etc.).

[0084] Using a set discriminant function or lightweight machine learning model The comprehensive feature vector Perform analysis to determine whether key resources are in a schedulable or abnormal state and define the judgment function :

[0085]

[0086] in is the judgment threshold, Used to comprehensively evaluate factors such as resource health and adjustability;

[0087] All distributed power equipment (such as energy storage devices, adjustable load equipment, cold and heat source stations and distributed power sources, etc.) The schedulable status and operation mode are uniformly recorded as resource status , forming a resource configuration file;

[0088] like , it means that the resource has the potential to respond to external scheduling. It is considered to be currently unadjustable or requires additional maintenance attention.

[0089] When using, comprehensive use of time series fusion sequence , device status and environmental data The ability to characterize multi-dimensional coupling characteristics has been improved, effectively identifying high-value scheduling resources and filtering out faulty or unstable resources. Compared with the traditional method that relies on a single threshold, this dynamic identification mechanism can adaptively adjust the judgment threshold and feature weight under different working conditions. The multi-source fusion results are deeply coupled with multi-dimensional operation information to construct a more flexible comprehensive feature vector. , through a configurable discriminant function It achieves refined identification and status marking, and is more adaptable.

[0090] Step 2: When obtaining the time series fusion sequence and resource status When the initial eigenvector is transformed by wavelet Perform multi-scale decomposition to generate prediction input vector , generate load forecast after parallel training , power generation forecast and energy storage forecasting , using the error function Dynamically evaluate the prediction deviation and trigger online parameter correction when the error exceeds the error threshold;

[0091] The second step includes the following:

[0092] Step 201: Feature data assembly and multi-dimensional preprocessing

[0093] From time series fusion and resource status In the system, key data that can reflect the real-time operation characteristics of the system are selected (such as the integrated power value of distributed photovoltaics, the current state of charge of energy storage, the start and stop signals of cold and hot load equipment, etc.), and the initial feature vector is constructed by combining external inputs such as environmental / meteorological data and market electricity prices. ;

[0094] Among them, the time series fusion sequence Indicates at time Comprehensive measurement information after fusion processing in step 1 (such as weighted results of PV array output, load-side current, etc.); Used to indicate the schedulability and operation mode of each resource;

[0095] In order to capture the fluctuation characteristics of load and distributed generation at different time scales, the initial eigenvector Wavelet transform is performed on some time series components in to extract high-frequency disturbances and low-frequency trends, and the following wavelet analysis formula is defined:

[0096]

[0097] in: In continuous time The target component on (can come from the initial eigenvector Any time series data in ); is the complex conjugate function of the mother wavelet; and They represent the translation and scale parameters (both are positive values) respectively, which can be adjusted within a certain range according to the needs of the forecast period;

[0098] This wavelet analysis can decompose the original time series into several multi-scale components, making it easier to handle slow-changing trends and short-term fluctuations in subsequent scheduling prediction models.

[0099] The multi-scale components obtained by wavelet transform are combined with other key features (such as ambient temperature and humidity, energy storage charge state, etc.) to form the final prediction input vector , for the predicted input vector Perform uniform normalization (such as mapping intervals to or ), and retain the same feature order and labeling;

[0100] Fusion of time series and resource status Deep integration helps improve the accuracy of the scheduling prediction model in describing the real dynamics of the system.

[0101] With the help of time-frequency domain analysis methods such as wavelet transform, the fluctuation characteristics of load and distributed power supply at different time scales can be effectively separated, and the prediction input vector can be improved. Recognizability;

[0102] Step 202: Constructing short-term and ultra-short-term scheduling prediction models

[0103] After getting the predicted input vector Finally, short-term prediction is achieved based on deep neural network (DNN). At the same time, lightweight time series model (such as network based on gated recurrent unit GRU) is used to achieve ultra-short-term prediction, and the key prediction quantities within different prediction time are output respectively. (load), (Distributed Generation Power) and (Storage charge trend), etc.

[0104] in Indicates the prediction step size. For example, short-term prediction can be taken as discrete moments in the range of 1 to 24 hours, while ultra-short-term prediction can be taken as intervals of 15 minutes or shorter.

[0105] More specifically, obtain the time series fusion sequence and historical load data over several periods , distributed power supply data and external weather data , these features are combined into the prediction input vector And send it to the dispatch forecast model to generate short-term or ultra-short-term load and power forecast values , Wait for the result;

[0106] The following formula gives a general neural network prediction output formula (for example only, not traditional mean or simple error statistics):

[0107]

[0108] in: Indicates the predicted value of the target variable (load, power output, etc.); Indicates the number of parallel submodules or different channels in the network (e.g., aggregation of multiple hidden layer units); For the The dynamic weight coefficient output by each submodule (adaptive adjustment can be made according to the characteristics of the current moment, and the value range can be ), which can be regarded as a learnable parameter within the network;

[0109] and Respectively represent the submodules at time For the prediction input vector The weight matrix and bias vector of ;

[0110] Represents a nonlinear activation function, which is used to enhance the model's ability to fit complex relationships, such as ReLU, tanh, etc.

[0111] Output of multiple key forecast quantities for short-term and ultra-short-term models , , , can be obtained separately through a similar structure as above, and then weighted fusion is performed to obtain the joint prediction results of multiple targets or multiple time periods;

[0112] Using historical data and multi-source fusion information, the above neural network model is trained and cross-validated in a rolling window manner to reduce the risk of overfitting. During the training process, the weight matrix is dynamically adjusted. , bias vector and dynamic weight coefficients After learning the parameters, the scheduling prediction model is obtained, which enables the model to adaptively respond to the effects of equipment aging, weather changes, and random load fluctuations;

[0113] By integrating deep neural networks with time series learning units such as gated loops, collaborative modeling of short-term and ultra-short-term forecasts is achieved within the same framework, significantly improving overall forecast accuracy.

[0114] The adaptive weight and bias update mechanism enables the model to respond quickly to changes in the external environment or system status. Compared with traditional single algorithms or simple linear regression, this step adopts a multi-channel neural network + adaptive weight approach to provide stronger tracking capabilities for complex nonlinear working conditions (such as drastic changes in weather or sudden load increases). By parallel modeling of long periods (short-term) and high-frequency bands (ultra-short-term), this solution takes into account both the global and fine-grained requirements of operation.

[0115] Step 203: Real-time evaluation and online model correction

[0116] When the scheduling prediction model is put into operation, the actual measurement values are obtained at a higher frequency (such as every 5 or 15 minutes) (may include current real load, photovoltaic output, energy storage status, etc.), and model output Make comparisons;

[0117] Define the following custom error function , dynamically measure the prediction accuracy:

[0118]

[0119] Where: is the target index, , The number of forecast objects that need to be evaluated simultaneously (such as load, distributed generation power, energy storage charge level, etc.); ,in: Indicates the The predicted value of a target, is the corresponding actual observation value; At the start of the current evaluation The relative time after, the value range , can be determined according to business needs the size of the session (e.g., 15 minutes, 1 hour, or longer);

[0120] For the corresponding The derivative error weight (non-negative real number) of each prediction target is used to balance the relative importance of the error amplitude and the error change rate in the comprehensive evaluation; if , the error trend is not considered, and only the error value itself is measured. When it is large, it emphasizes the penalty / reward for the growth or decay speed of the prediction deviation in the evaluation;

[0121] For the The time decay factor of the target (non-negative real number), which controls the As time increases (i.e., time passes), the weights assigned to earlier errors and later errors in the evaluation are

[0122] when When the value is large, the The error penalty is stronger: if , then for the entire interval The errors within are accumulated indiscriminately; The evaluation duration (a positive real number) can be set based on the business scenario. For example, an evaluation duration of 5 to 15 minutes is used for ultra-short-term evaluation, while an evaluation duration of several hours is used for short-term evaluation. This corresponds to the time scale requirements for short-term and ultra-short-term predictions in step 2.

[0123] According to the error function If the error continues to be higher than the preset error threshold, the online correction program is triggered; the latest prediction input vector collected in real time is used and target prediction value Incrementally incorporate the model training set, update the model parameters through small-scale iterative training, and dynamically adjust the weight matrix , bias vector and dynamic weight coefficients Wait, complete the online calibration;

[0124] If a significant change in the system equipment status is detected (for example, a cooling unit is overloaded or some photovoltaic modules fail), the resource status will also be used to Interact with this step to reinitialize or accelerate the update of model weights;

[0125] The real-time evaluation results are regularly output as a forecast quality report, including the error function for the most recent period. Trend and comprehensive evaluation of short-term / ultra-short-term forecast accuracy, using a custom error function , more flexibly highlighting the importance of short-term prediction accuracy, meeting the virtual power plant's demand for rapid response. Through online correction, the model can adaptively track changes in equipment status and external conditions, avoiding the accuracy degradation of pure offline training models in non-stationary environments. Different from the common static measurement methods of mean square error or mean absolute error, the error function is introduced. The near-term prediction quality and the far-term prediction error are organically integrated and given different priorities.

[0126] Step 3: When the prediction results and resource status When input, the objective function is optimized through multi-objective Integrate constraint objectives, use improved particle swarm optimization to perform multi-stage optimization on the scheduling decision vector X, build power balance hard constraints based on equipment rated parameters, and use adaptive weight adjustment mechanism to generate the optimal scheduling solution , and generate equipment execution plans, evaluate the effectiveness of the plans through economic-reliability-environmental indicators, and achieve multi-objective dynamic trade-offs and coupling with global-local scheduling strategies;

[0127] The step three includes the following:

[0128] Step 301: Multi-objective function definition and constraint modeling

[0129] Based on short-term / ultra-short-term forecast results, such as load forecast , Distributed Power Generation Prediction And energy storage charge trend forecast , combined with the schedulable resource status, that is, the resource status ,

[0130] Constructing multi-objective optimization objective function , in order to measure the economic, reliability and environmental protection effects at the same time, the time domain is , then define:

[0131]

[0132] in: 、 are the intermediate values of the first and second objective functions respectively;

[0133] X represents the dispatch decision vector to be solved (e.g., the dispatch or output instructions for each energy storage, load device, cooling and heating energy station, photovoltaic, etc.); , , are the weight coefficients of the targets (all are non-negative real numbers, and at least one is greater than 0);

[0134] Can represent the current time The running cost function and the prediction results , And the market price, on-grid electricity price, etc. are combined to calculate; Used to measure the deviation of system power supply / energy supply reliability, such as the difference between load demand and actual supply capacity, or the penalty term caused by insufficient safety margin;

[0135] It measures emission levels or clean energy utilization; It can represent the energy offset of energy storage charging and discharging operations; It can reflect the peak-shaving gap or potential risk caused by the mismatch between the current demand and the forecast demand; , is an exponential decay factor related to energy storage scheduling and reliability, with values ranging from 0 to 10;

[0136] The rated operating range of the equipment, such as the maximum charge and discharge power of the energy storage, and the upper and lower capacity limits of the cooling and heating energy stations, is used as the hard constraint of the scheduling decision vector X. With reference to the prediction range and time resolution, each discrete scheduling period should ensure power balance and safety margin, such as:

[0137]

[0138] in: Indicates the power drawn from (or sent to) the grid. Indicates line and conversion equipment losses; It is the additional available power after adjustable load peak shifting or energy storage release; environmental protection or emission indicators can also be made into hard constraints or soft constraints, such as carbon emission limits, air pollutant emission limits, etc.

[0139] When in use, by introducing exponential decay and integral forms, more emphasis is placed on the cumulative performance in the entire time domain when measuring the scheduling effect, and differentiated dynamic amplification / penalty is performed on frequent energy storage operations or load differences, while taking into account economy, reliability and environmental protection, and being able to flexibly balance the multiple objectives of the virtual power plant; compared to the use of traditional weighted linear models or simple cost functions, this step designs a comprehensive evaluation function of multiple objectives + exponential correction and combines the energy storage and load time series information obtained from the second step prediction to control the scheduling smoothness and stability.

[0140] Step 302: Implementation of multi-objective optimization algorithm

[0141] The particle swarm position represents the scheduling decision vector X, and the fitness is defined as the multi-objective optimization objective function The opposite number of or other methods are used to achieve minimization; the system state (including predicted load, energy storage margin, price information, etc.) during the dispatch period is used as the state space, the dispatch decision vector X is used as the action space, and the multi-objective optimization objective function is used. The algorithm learns strategies based on corresponding rewards or penalties; adjusts and sets algorithm initialization parameters such as particle swarm size, exploration step size, or the initial network structure of reinforcement learning based on device topology information (such as the number of hot and cold energy stations, storage capacity, and adjustable load types), prediction duration, and time granularity.

[0142] This enables the algorithm to conduct targeted optimization in a search space that matches the device scale and time granularity, thereby improving the efficiency and stability of solving complex energy management problems.

[0143] During the execution process, to take into account both short-term (such as 1-24 hour scheduling) and ultra-short-term (such as 15-minute rolling scheduling) requirements, the optimization can be divided into multiple stages:

[0144] Local stage: Make detailed corrections to key periods (peak load periods, photovoltaic fluctuation periods, etc.) at the minute level;

[0145] Macro stage: Generate a global scheduling strategy with hourly steps;

[0146] When the improved particle swarm algorithm finds that carbon emissions or safety margins are approaching the upper / lower limit during the iteration process, the weight coefficient can be adaptively increased. or proportion, to explicitly guide the solution towards a greener and more reliable direction;

[0147] The multi-objective optimization solution process is achieved by monitoring the multi-objective optimization objective function The convergence is determined by the iterative changes of the feasible solution and the satisfaction of the feasible solution constraints. If it cannot be fully converged within the preset number of iterations or time, the current optimal feasible solution is taken and the optimal scheduling plan is finally output. , including: the start and stop sequence of each load device, load distribution, the charging and discharging power of each energy storage in each time period, the priority utilization rate of clean energy (such as photovoltaics), and the interaction strategy with the power grid (grid connection, up / down regulation, etc.);

[0148] The improved particle swarm algorithm (PSO) achieves rapid convergence in high-dimensional search spaces and exhibits strong adaptability to complex nonlinearities and constraints. A multi-stage solution strategy effectively combines short-term and ultra-short-term demands, improving the feasibility and precision of scheduling strategies across different timescales. The introduction of a solution mechanism combining adaptive weights with a multi-stage approach balances economic and environmental objectives at a macro level while enabling precise adjustment of energy storage and load within local timeframes to address high-frequency fluctuations or unpredictability.

[0149] Step 303: Scheduling result analysis and next step connection

[0150] The optimal scheduling solution will be output Map to the actual device level and format the scheduling policy output, where:

[0151] Energy storage equipment: Ensure that the charge and discharge power and SOC drift in each sub-period match the predicted power consumption / generation situation; cooling and heating energy stations and adjustable loads: Define the start and stop timing and operating power to ensure that the district heating / cooling and process requirements are met; photovoltaic and other distributed power sources: Develop a priority grid connection or self-use strategy based on weather forecasts and electricity price fluctuations;

[0152] The optimal scheduling solution Decompose the performance of each item in the objective function, for example:

[0153] Economic cost indicators (such as total electricity purchase costs or revenue); reliability indicators (such as maximum load difference, number of default penalties); environmental protection indicators (such as cumulative carbon emissions or renewable energy utilization rate);

[0154] Execute the optimal scheduling plan , and dynamically adjust as real-time monitoring data and prediction errors change; if it is found during execution that the deviation between prediction and reality increases, it may trigger online optimization or local rescheduling to form a closed-loop collaborative mechanism.

[0155] When used, the multi-objective decomposition of scheduling results helps decision makers understand the degree of achievement in the economic, reliability, and environmental dimensions, providing an effective reference for subsequent closed-loop optimization. Furthermore, during the multi-objective optimization process, an additional backup scheduling option or safety mode solution is output to address serious failures or extreme operating conditions.

[0156] Step 4: When the optimal scheduling plan When sent to the field equipment, the instructions are converted into equipment-level operation sequences through the field energy management system, and the priority allocation mechanism is used to synchronously send control instructions and collect monitoring indicator vectors. , build performance measurement function Evaluate execution deviations, when the deviations accumulate and exceed the performance threshold The rolling correction process is triggered when the local time window is optimized to generate the local optimal solution. ;

[0157] The step 4 includes the following contents:

[0158] Step 401: Dispatching instructions are issued and on-site data is transmitted back

[0159] Based on the optimal scheduling plan , extracting the operating power, start / stop status, and charge / discharge plan of various resources such as energy storage, cold / hot energy stations, adjustable loads, and distributed power sources in each time period, and converting them into specific execution instructions for the equipment;

[0160] The on-site energy management system (EMS) matches the execution points one by one according to the corresponding resource identifiers;

[0161] Integrating with the virtual power plant cloud platform, prioritizes the operation instructions for each type of equipment (e.g., energy storage devices) (e.g., energy storage dispatch precedes adjustable load peak shaving operations), ensuring that critical resources can be dispatched first in emergencies.

[0162] Use high-concurrency message queues or distributed control channels to send instructions synchronously at the same time or within the same scheduling cycle. For critical equipment prone to security risks, confirmation and response links can be added to ensure execution reliability.

[0163] After receiving and executing the command, the on-site energy management system (EMS) transmits the actual equipment operation information (such as the current energy storage charge state, cooling and heating unit power, adjustable load output, etc.) back to the cloud platform, which together with the data obtained by the sensors forms a comprehensive operation monitoring source;

[0164] During use, the cascaded model of scheduling, device mapping, and execution feedback avoids actual control confusion caused by inconsistent device location or identification. Bidirectional communication with the cloud platform and on-site energy management system (EMS) ensures real-time visibility of execution instructions. Typically, scheduling plans are issued as static documents, lacking synchronization and feedback tracking. Here, the design of high-concurrency channels and priority allocation mechanisms ensures low-latency and highly reliable execution even with large-scale distributed devices operating in parallel.

[0165] Step 402: Real-time monitoring and system operation evaluation

[0166] According to the forecast results (such as load , photovoltaic output , energy storage status ) and the optimal scheduling solution , define real-time monitoring indicators, including but not limited to:

[0167] Energy storage utilization: the ratio of the actual power output of the energy storage device in the current period to the planned power;

[0168] Load tracking degree: the difference between the actual output of the adjustable load and the predicted value;

[0169] Clean energy utilization rate: the current proportion of distributed power generation such as photovoltaics in the total energy supply;

[0170] Safety margin: the remaining space from the rated capacity or safety threshold of the equipment;

[0171] The above indicators are based on monitoring indicator vectors The data are summarized in a certain form for reference in the next step of evaluation or scheduling correction;

[0172] To compare the optimal scheduling solutions The deviation between the execution effect and the current running state, the performance measurement function is introduced , taking into account both real-time deviation and change speed through integration, the formula is as follows:

[0173]

[0174] Where: , indicating that at time When , the current execution deviation or the difference between the execution result and the scheduling plan is measured. If the scheduling is more consistent with the actual state, the The closer it is to 0, the greater the difference is. Positive increase; The scheduling instructions to be executed, For real-time monitoring of indicator vectors; , a function for measuring scheduling deviation; middle, When , the deviation is amplified superlinearly, making the large error account for a higher proportion in the overall evaluation: When , it returns to the linear situation; middle, is the weight factor of the derivative term (non-negative real number), which is used to balance the importance of the deviation amplitude and the deviation change rate. Then the extreme sudden increase of error change is given extra weight through this item; and All are control parameters; middle, , Represents the time decay factor; Set to 0, so that the entire Deviations within the interval are treated equally, To evaluate the window length;

[0175] If the performance metric function At the preset performance threshold Above, if a nonlinear jump in the energy storage / load equipment is detected (such as an extreme instantaneous deviation of power from the plan), an early warning signal will be sent to the cloud platform to prompt the scheduling layer or manual operation and maintenance to conduct further inspection; at the same time, it can trigger the rolling correction process to update the scheduling strategy in time to avoid the accumulation of deviations in subsequent time periods.

[0176] When using, with the help of performance measurement function , it can accumulate the impact on scheduling execution deviation in a short period of time, and use exponential decay to highlight the importance of recent abnormal events, which is suitable for dynamic consideration of synchronous scheduling of multiple devices; traditional real-time monitoring often uses single-point moments or basic statistical indicators for evaluation. Here, the deviation within the entire time period can be continuously observed, which enhances the adaptability to frequent fluctuation scenarios.

[0177] Step 403: Rolling correction and closed-loop collaborative optimization

[0178] When the performance metric function The performance threshold is exceeded for several consecutive monitoring periods. , or automatically trigger the dispatch re-evaluation process when there are major anomalies in load, power generation, and energy storage status, and collect the latest data in real time (including time series fusion sequence and resource status , error function , model online correction data, and scheduling plan execution deviation, etc.) will be integrated to form a new current state vector ;

[0179] Use improved particle swarm optimization or deep reinforcement learning to quickly reoptimize the scheduling plan within a local time window (which can be defined as the next 1-2 hours or less) to find a new local optimal solution Compared with the original optimal scheduling solution After comparison, if the recent operation deviation or cost / emission can be significantly reduced, the original scheduling instructions can be partially replaced, and the original plan will be used for the rest of the period. The local optimal solution after rolling correction Re-enter the instruction issuance process and accept real-time monitoring and evaluation;

[0180] When in use, with the help of small-scale, fast local re-optimization, the response speed to real-time fluctuations can be significantly improved, avoiding the high computational burden brought by large-scale global optimization. The closed-loop collaboration operates continuously under a unified data and algorithm framework, integrating equipment identification, prediction and multi-objective optimization into a mechanism that can be automatically iterated and updated, which significantly improves the practicality and robustness of the solution. While maintaining the global optimal baseline, it flexibly responds to short-term fluctuations and uncertainties, maximizing the operating benefits of the integrated virtual power plant.

[0181] Step 5: When the monitoring indicator vector Fusion Sequence with Time Series When an exception occurs, build a fault judgment function , identify device-level failure modes, when the fault decision function Exceeding the fault threshold The self-healing control is triggered when the backup scheduling scheme is adopted to start the redundant equipment and adjust the load distribution strategy. After the repair, based on the safety threshold Restore device scheduling permissions;

[0182] The step five includes the following:

[0183] Step 501: Artificial Intelligence Fault Diagnosis

[0184] Combined with time series fusion sequence (such as operating signals collected from cold and hot energy stations, adjustable load equipment, energy storage and photovoltaics), as well as monitoring indicator vectors (For example, energy storage charge state, load tracking degree, etc.; these data are uniformly recorded as comprehensive feedback data ,This tag covers multi-dimensional information such as sensor high-concurrency collection values, ,device operation history, and scheduling execution deviation;

[0185] A model combining FaultTree and time series deep learning (such as LSTM, GRU or multi-layer CNN-RNN hybrid structure) is introduced to identify potential fault signs at the device and system level online. The following fault judgment function can be defined :

[0186]

[0187] in: Indicates at time A measure of whether the system is at risk of failure; is the possible failure mode or the number of leaf nodes (obtained from fault tree analysis), represents the failure mode index; For failure mode The relevant weight coefficient, a non-negative real number, is used to balance the criticality of different fault types; is the recognition function of the time series deep learning model for the input data, Indicates the failure mode The set of trainable parameters of Indicates at time Failure Mode trigger duration or severity factor;

[0188] represents the fault attenuation factor, a non-negative real number used to balance the severity of different fault types;

[0189] When the fault judgment function Exceeding the preset fault threshold When a fault occurs, it is determined that there is a high risk of failure, the corresponding device is treated as a faulty device, and a fault diagnosis instruction is issued; if it is determined that a failure may occur at the device level or system level, it is passed to the second step of prediction model construction and real-time operation evaluation and the third step of intelligent scheduling solution generation based on multi-objective optimization in the form of a fault warning: the scheduling prediction model is allowed to consider the possibility of the device running with a fault or shutting down due to a fault, and the scheduling optimization part also increases the fault constraints or reduces the schedulable range accordingly.

[0190] The monitoring mechanism triggers backup scheduling plans to prevent equipment from continuing to operate in a high-risk state. The backup scheduling plans include:

[0191] Backup equipment activation: such as enabling backup energy storage systems, starting additional power generation equipment, or adjusting the load-side response to compensate for the absence of faulty equipment; load regulation: in some cases, load scheduling may need to be adjusted to ensure that the system's load demand can be met, especially when the faulty equipment cannot work normally; adaptive adjustment of scheduling strategy: during the generation of the backup scheduling plan, the scheduling strategy will be dynamically adjusted based on fault warning, prediction deviation and other information to reduce the impact of the fault on the overall system performance.

[0192] When used, fault trees ensure the explainability of fault mechanisms, while time-series deep learning further enhances the detection of multidimensional sensor data and hidden faults. Traditional fault detection often relies solely on static thresholds or simple statistical features. This approach introduces a comprehensive modeling approach combining fault trees, time-series deep learning, and exponential decay factors. This not only addresses complex nonlinear faults, but also dynamically quantifies the degree of fault evolution and seamlessly integrates with multi-objective scheduling systems.

[0193] Step 502: Self-healing control and backup scheduling scheme

[0194] When the fault judgment function Exceeding the preset fault threshold When a rapid upward trend occurs, the system will immediately enter the self-healing control process; re-evaluate the faulty equipment or associated loads, and issue instructions to reduce or prohibit scheduling of the faulty equipment, or increase the corresponding safety factor; suspend or terminate the issuance of scheduling instructions for the faulty equipment, and call the following backup scheduling plan;

[0195] After receiving the fault diagnosis instruction, it immediately switches to the backup strategy to quickly adapt to the changes in load and energy supply structure after the fault. It can perform secondary optimization on non-faulty equipment in the local time window and generate a temporary scheduling plan to ensure the supply of the system's core load and minimum emission requirements;

[0196] After the self-healing process is initiated, perform maintenance or isolate the faulty equipment for a period of time and gradually observe whether the risk value given by the fault tree model drops significantly:

[0197] like ,in: If the device reaches a safety threshold, indicating that the fault has been eliminated or the risk is controllable, the normal scheduling plan can be restored, and the data during the fault period can be synchronized to the data layer in step 1 to improve the fault sample library. If the device is repaired, it can be included in the second step prediction and third step scheduling. If the repair is not complete, it will remain isolated to avoid affecting the overall system operation.

[0198] During use, the risk of fault propagation and high-cost power outages is greatly reduced through the pre-deployed backup scheduling plan, as well as the immediate isolation of faulty equipment and secondary optimization of adjacent equipment in self-healing control, ensuring the continuous power supply of critical loads in the system. After the system returns to normal, the fault diagnosis information and self-healing execution records can be further fed back to the first-step data fusion and the second-step scheduling prediction model to form a fault history library, thereby improving the accuracy and timeliness of the next fault response. Traditional solutions mostly use manual troubleshooting or mechanical shutdown to respond to faults. Here, the pre-buried automation mechanism of self-healing control + backup scheduling, combined with real-time fault diagnosis and judgment, not only responds quickly, but also can reconstruct the scheduling plan in a short time, with highly intelligent fault management and control capabilities.

[0199] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0200] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0201] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is only for some logical functions. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0202] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0203] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. An intelligent dispatching method for an integrated virtual power plant, characterized by: include, Preprocess the sensor data of distributed power equipment, generate a time series fusion sequence after weighted fusion, and build a comprehensive feature vector based on the equipment operating conditions and environmental data to determine the resource status; The initial feature vector is decomposed into a multi-scale vector to generate a prediction input vector. After parallel training, the prediction result is generated. The error function is used to dynamically evaluate the prediction deviation. When the error exceeds the error threshold, online parameter correction is triggered. By integrating constraint objectives through a multi-objective optimization objective function, the scheduling decision vector is optimized in multiple stages, power balance hard constraints are constructed based on equipment rated parameters, and the optimal scheduling solution is generated using an adaptive weight adjustment mechanism. When the optimal scheduling plan is issued to the field equipment, the instructions are converted into a device-level operation sequence. A priority allocation mechanism is used to synchronously issue control instructions and collect monitoring indicator vectors. A performance measurement function is constructed to evaluate execution deviations. When the accumulated deviation exceeds the performance threshold, a rolling correction process is triggered. When the monitoring indicator vector and the time series fusion sequence are abnormal and the fault judgment function exceeds the fault threshold, self-healing control is triggered, redundant equipment is started, the load distribution strategy is adjusted, and the equipment scheduling authority is restored based on the safety threshold. After receiving the fault diagnosis instruction, it immediately switches to the backup strategy to quickly adapt to the changes in load and energy supply structure after the fault. Secondary optimization is performed on non-faulty equipment in the local time window to generate a temporary scheduling plan.

2. The intelligent scheduling method for a comprehensive virtual power plant according to claim 1, characterized in that: Sensors are deployed at distributed power equipment. The raw data reported by each node and the corresponding timestamp are initially aggregated on the edge. The acquired raw data is filtered and de-redundant to obtain pre-processed data after preliminary screening. Synchronous collection and unified time sequence alignment are performed for different nodes. Import the pre-processed data into the central database, obtain the cleaned data after pre-processing, perform weighted fusion on the data of multiple nodes, and generate a unified single-point fusion value at the moment: The single-point fusion values are arranged in sequence to form a time series fusion sequence.

3. The intelligent scheduling method for a comprehensive virtual power plant according to claim 2, characterized in that: Integrate the time series fusion sequence with equipment operating condition records and meteorological data to form a comprehensive feature vector; The comprehensive feature vector is analyzed to determine whether the key resources are in a dispatchable or abnormal state, and the dispatchable state and operation mode of each distributed power device are recorded as the resource state.

4. The intelligent dispatching method for a comprehensive virtual power plant according to claim 3, characterized in that: From the time series fusion sequence and resource status, key data is selected and combined with external input to construct the initial feature vector; Wavelet transform is performed on some time series components in the initial feature vector to extract high-frequency disturbances and low-frequency trends, decompose the original time series into several multi-scale components, and merge the obtained multi-scale components with other key features to form a prediction input vector.

5. The intelligent dispatching method of the integrated virtual power plant according to claim 4, characterized in that: Implement short-term prediction based on deep neural networks and ultra-short-term prediction using lightweight time series models, outputting key prediction quantities within different prediction timeframes. The neural network model is trained and cross-validated using a rolling window method, and the scheduling prediction model is obtained after dynamically adjusting the learnable parameters.

6. The intelligent dispatching method of the integrated virtual power plant according to claim 5, characterized in that: When the scheduling prediction model is put into online operation, actual measurement values are obtained at a high frequency and compared with the model output. An error function is defined to dynamically measure the prediction accuracy. Based on the fluctuation of the error function, if the error continues to exceed the preset error threshold, the online correction program is triggered. The latest prediction input vector and target prediction value collected in real time are incrementally incorporated into the model training set, and the model parameters are updated through small-scale iterative training to complete online correction.

7. The intelligent dispatching method of the integrated virtual power plant according to claim 6, characterized in that: Based on the short-term / ultra-short-term forecast results of key forecast quantities and combined with resource status, a multi-objective optimization objective function is constructed; Based on the hard constraints of the scheduling decision vector within the rated operating range of the equipment, and referring to the prediction range and time resolution, power balance and safety margin should be guaranteed in each discrete scheduling period.

8. The intelligent dispatching method of the integrated virtual power plant according to claim 7, characterized in that: The system state within the scheduling period is used as the state space, the scheduling decision vector is used as the action space, and the reward or penalty corresponding to the multi-objective optimization objective function is used to learn the strategy; Convergence is determined by monitoring the iterative changes of the multi-objective optimization objective function and the satisfaction of the feasible solution constraints. If complete convergence cannot be achieved within the preset number of iterations or time, the current optimal feasible solution is taken and the optimal scheduling plan is finally output; The output optimal scheduling plan is mapped to the actual device level, the scheduling strategy is formatted and output, and the performance of the optimal scheduling plan in the objective function is decomposed.

9. The intelligent dispatching method of the integrated virtual power plant according to claim 8, characterized in that: Based on the optimal scheduling plan, the operating power, start-stop status, and charge-discharge plan of distributed power equipment in each time period are extracted and converted into specific execution instructions for the equipment; The on-site energy management system matches execution points one by one according to the corresponding resource identifiers, and combines with the virtual power plant cloud platform to establish a priority order for the operating instructions of each type of equipment, and issues the instructions synchronously at the same time or within the same scheduling cycle. After receiving the instructions and completing the execution, the on-site energy management system transmits the actual operation information of the equipment back to the cloud platform, which together with the data obtained by the sensors constitutes a comprehensive operation monitoring source.

10. The intelligent dispatching method of the integrated virtual power plant according to claim 9, characterized in that: Based on the prediction results, a monitoring indicator vector is defined to compare the deviation between the execution effect of the optimal scheduling plan and the current operating status. When the introduced performance measurement function exceeds the performance threshold for several consecutive monitoring cycles, the scheduling re-evaluation process is automatically triggered, integrating the latest data collected in real time to form the current state vector.

11. The intelligent dispatching method of a comprehensive virtual power plant according to claim 10, characterized in that: The improved particle swarm algorithm is used to quickly re-optimize the scheduling plan within the local time window. After comparing the new local optimal solution with the original optimal scheduling plan, if it can reduce the recent operating deviation or reduce costs / emissions, the original scheduling instructions can be partially replaced. The original plan will still be used for the rest of the time. The local optimal solution after rolling correction will enter the instruction issuance process again and be subject to real-time monitoring and evaluation.

12. The intelligent dispatching method of the integrated virtual power plant according to claim 11, characterized in that: After combining the time series fusion sequence and the monitoring index vector, they are uniformly recorded as comprehensive feedback data and the fault judgment function is defined; When the fault judgment function exceeds the preset fault threshold, it is determined that there is a high fault risk, the corresponding device is regarded as a faulty device, and a fault diagnosis instruction is issued; When the fault judgment function exceeds the preset fault threshold or shows a rapid upward trend, the self-healing control process is immediately entered and the backup scheduling plan is called.

13. The intelligent dispatching method of the integrated virtual power plant according to claim 12, characterized in that: After the self-healing process is initiated, the faulty equipment is maintained or isolated, and the risk value given by the fault tree model is gradually observed to see whether it decreases. If it does, it means that the equipment fault has been eliminated or the risk is controllable, and the normal scheduling plan can be returned. If the equipment is repaired, it can be added to the scheduling scope again; if the repair is not completed, it will remain isolated to avoid affecting the overall system operation.

Images