CN-122026626-A - Source-load cooperative control method for park micro-grid parallel-off-grid switching

CN122026626ACN 122026626 ACN122026626 ACN 122026626ACN-122026626-A

Abstract

The invention provides a source load cooperative control method for park micro-grid off-grid switching, which comprises the steps of arranging edge intelligent bodies with local state sensing, light dynamic prediction and neighborhood communication functions on distributed power supplies, loads and energy storage nodes, adopting state vectors after multichannel real-time acquisition, denoising and standardization processing to drive a compression version LSTM to conduct power prediction, utilizing an improved weighted consistency algorithm to realize distributed prediction error fusion and compensation through low-delay communication among the edge nodes, dynamically distributing coordination weights according to grid change, energy storage and load states, enabling each node to autonomously generate a constraint optimized local control instruction based on a prediction correction result, introducing exponential attenuation suppression of historical deviation, and strengthening closed-loop management and control of multi-period errors.

Inventors

TIAN JING
FENG BINJIE
Xu Caishen
LI TAO
LI SHENGWEI
ZENG GUO

Assignees

广州劲源科技发展股份有限公司

Dates

Publication Date: 20260512
Application Date: 20260128

Claims (10)

1. The source-load cooperative control method for the park micro-grid and off-grid switching is characterized by comprising the following steps of: s1, respectively deploying edge intelligent agents in a distributed power supply, load nodes and an energy storage device of a park micro-grid, wherein each edge intelligent agent integrates a local state sensing unit, a dynamic prediction model and a neighborhood communication interface; S2, each edge agent collects original operation data of the node through a local state sensing unit in real time, and denoising and normalizing pretreatment are carried out on the original operation data to generate a standardized local operation state vector; s3, inputting the standardized local running state vector into an LSTM dynamic prediction model in the edge intelligent agent, performing trend prediction, outputting a local power predicted value at the next moment, and calculating a local prediction deviation value based on the current actual measurement value and the local power predicted value; s4, each edge agent exchanges the local prediction deviation amount and the state confidence weight with the adjacent edge agents through a low-delay communication network, and the prediction deviation in the area is subjected to distributed fusion to generate an area cooperative error compensation factor; S5, correcting the local power predicted value based on the regional cooperative error compensation factor to generate regional cooperative predicted output after deviation compensation; and S6, when the grid-connected to off-grid switching triggering condition is detected, the main control unit starts a dynamic coupling coordination mechanism, and calculates the dynamic influence weight of each edge agent in cooperative control according to the attenuation degree of the power grid supporting capacity, the SOC level of the energy storage system and the load sensitivity index to form a dynamic coupling weight configuration matrix.
2. The method for controlling source load cooperation for on-line and off-line switching of a campus-oriented micro grid according to claim 1, wherein the step S6 further comprises: s7, each edge agent adjusts the response intensity of the edge agent to the whole control instruction according to the dynamic coupling weight configuration matrix, and generates a local control action instruction by combining the regional collaborative prediction output subjected to deviation compensation and the physical constraint condition of local equipment; And S8, introducing an error propagation inhibition function in the control execution process, and taking the influence of the historical prediction deviation on the current control action into a feedback regulation path in an exponential decay mode to continuously optimize the prediction and control behaviors in the subsequent period.
3. The method for collaborative control of source load for on-grid and off-grid switching of a campus-oriented micro-grid according to claim 1, wherein the step S1 further comprises identifying distributed power sources, adjustable loads, energy storage nodes and generating a deployment site mapping table by adopting a graph structure analysis algorithm and a node type matching method based on the electrical topology and equipment distribution characteristics of the campus-oriented micro-grid.
4. The method for collaborative control of source load for a campus-oriented microgrid and off-grid switching according to claim 1, wherein the raw operational data includes power output, load demand, environmental parameters, and operational state data.
5. The source load cooperative control method for the campus-oriented micro-grid off-grid switching of claim 1, wherein the step S3 specifically includes: based on the standardized local running state vector generated in the step S2, acquiring multidimensional time sequence data serving as an input sequence of a compressed LSTM dynamic prediction model; performing feature extraction and time sequence modeling processing on the standardized local running state vector by using the compressed LSTM dynamic prediction model, and outputting a local power predicted value of the next control period; Based on the actually collected local power actual measurement value and the local power predicted value in the current control period, calculating a difference value between the local power actual measurement value and the local power predicted value to generate a local predicted deviation value, and carrying out polarity marking and amplitude normalization processing on the local predicted deviation value to obtain a standard deviation characteristic parameter; Packaging the standard deviation characteristic parameters, the corresponding time stamps, the node type identifiers and the state confidence weights, and generating a structured deviation information data packet; And synchronously caching the local power predicted value and the deviation information data packet to a local storage unit of an edge intelligent agent to form a predicted record sequence with deviation marks.
6. The method for collaborative control of source load for on-grid and off-grid switching of a campus-oriented micro grid according to claim 5, wherein the compression design in the compression version LSTM dynamic prediction model comprises reducing the number of hidden layer units to a preset low-dimensional scale, limiting the number of network layers to be not more than a limit value, adopting a weight pruning strategy to reject parameter connection with low contribution, reducing the overall operation complexity of the model and adapting the calculation force limitation of an edge calculation unit.
7. The source load cooperative control method for the campus-oriented micro-grid and off-grid switching of claim 1, wherein the step S4 specifically includes: Constructing a dual-channel data packet based on the local prediction deviation amount calculated by each edge agent in the step S3 and the state confidence weight generated by the running state data acquired by the local state sensing unit; Each edge agent sends the dual-channel data packet to adjacent edge agents which are adjacent geographically or electrically coupled through a low-delay communication network, a decentralised local communication topological structure is established, and based on the local communication topological structure, local prediction deviation amounts and corresponding state confidence weights of other edge agents in a k-hop neighborhood range are collected to form a regional multisource error observation set; Based on the regional multisource error observation set, executing an improved weighted average consistency algorithm to carry out consistency convergence, and continuing until a preset convergence threshold is met; After the consistency convergence is completed, each edge agent outputs a final fusion result thereof to generate a unified regional cooperative error compensation factor; And injecting the regional collaborative error compensation factors into a local prediction compensation loop, and simultaneously recording the time stamp and the source topology range of the regional collaborative error compensation factors to form an error fusion log.
8. The method for collaborative control of source load for on-grid and off-grid switching of a campus-oriented micro-grid of claim 7 wherein the improved weighted average consistency algorithm is specifically implemented by taking the state confidence weight of each node as a weighting coefficient, performing weighted average operation on the local prediction deviation in the neighborhood, introducing a relaxation factor to suppress high-frequency oscillation, and iteratively updating the local fusion intermediate value of each edge agent.
9. The method for controlling source load cooperation for on-line and off-line switching of a campus-oriented micro grid according to claim 1, wherein the step S5 specifically includes: based on the regional cooperative error compensation factors output in the step S4, obtaining local prediction deviation amounts obtained by exchanging all adjacent edge intelligent agents and corresponding state confidence weights thereof, and carrying out consistency alignment processing on multi-source deviation data by using a weighted residual fusion model to generate a standardized regional deviation coordination vector; Inputting the standardized regional deviation coordination vector into a distributed error inversion mapper, executing local error contribution analysis based on Jacobian iteration, calculating error propagation sensitivity coefficients of each edge agent under the current regional cooperative environment, and generating a regional error sensitivity distribution field; Constructing a dynamic compensation gain adjustment function based on the regional error sensitivity distribution field and the original local power predicted value, and applying space self-adaptive correction to the predicted output of each edge intelligent agent to generate a preliminary corrected local power predicted sequence; And performing cross-time window sliding aggregation operation on the initially corrected local power prediction sequence, and performing smooth reconstruction by using a Kalman filter in combination with the second-level and minute-level double-time-scale trend characteristics to generate a stabilized region collaborative prediction output sequence.
10. The method for collaborative control of source load for on-grid and off-grid switching of a campus-oriented microgrid of claim 1 wherein the regional error sensitivity distribution field reflects the spatial distribution of the impact of each node prediction bias on the overall prediction outcome.

Description

Source-load cooperative control method for park micro-grid parallel-off-grid switching Technical Field The invention relates to the technical field of intelligent control and prediction error compensation of micro-grids, in particular to a source-load cooperative control method for park-oriented micro-grid off-grid switching. Background Currently, the field of the cooperative control of source and load of a micro-grid in a park faces multiple challenges such as aggravation of dynamic properties of distributed energy and load fluctuation, variable system operation conditions, frequent switching of off-grid modes and the like. The main stream technical proposal in the industry mainly adopts a centralized coordination control method taking a main control unit as a core, or combines partial distributed prediction units to carry out auxiliary adjustment of source load balance. The schemes are usually focused on relying on a global load prediction model and a distributed energy output prediction model, and a dispatching master station issues control instructions uniformly to realize comprehensive coordination of power supplies and loads in the micro-grid. In order to improve the timeliness and robustness of control decisions, a part of technical schemes adopt a deep learning method (such as an LSTM neural network) or time sequence statistical analysis to predict the short time of key variables such as power, load and the like, and cooperate with algorithms such as model predictive control, integrated optimization of raw storage and the like to perform real-time scheduling. The prior art further develops a multi-time scale prediction and layering control mechanism, and the micro-grid operation is optimized through daily, hour and second progressive prediction and control instruction distribution; However, the prior art still has obvious defects under the scene of fast collaborative control of parallel-to-on-grid switching and source load, namely, the centralized or global prediction model is excessively relied on to cause the heterogeneity of the distributed node state and the response delay of local disturbance to the whole-grid control, and the single prediction model error is often amplified in the multi-period iteration or parallel-to-on-grid switching process, which is shown as the difficulty in real-time convergence between the actual control output and the preset regulation target, and the voltage frequency fluctuation and the switching transient instability probability of the micro-grid operation are increased. Meanwhile, the error independent accumulation among the distributed nodes lacks an effective fusion and correction mechanism, so that the whole cooperative control is disturbed due to the fact that the partial areas are predicted by the isolated anomalies. Some existing compensation methods cannot dynamically adjust the prediction errors of multiple time scales in real time, the error feedback path is too long or too rigid, and the distributed network structure and asynchronous state change are difficult to flexibly adapt to, so that the source-load cooperative control system has insufficient robustness to uncertain factors and local disturbance; In addition, the existing intelligent control scheme of the micro-grid in the park is large in operation load, and for the application requiring second-level dynamic response, the cloud or master control end model reasoning time delay may exceed the strict time limit of actual switching and scheduling, so that the system safety margin is reduced. For key nodes such as distributed energy storage, adjustable load, distributed energy sources and the like, the physical constraint, the operation characteristics and the equipment response capability of the key nodes are obviously different, and the current technology fails to fully utilize the local situation awareness and the neighborhood synergistic effect, so that control instruction execution is unevenly distributed or inertia adjustment is insufficient. Disclosure of Invention The invention aims to solve the technical problems and provides a source load cooperative control method for park micro-grid off-grid switching. The technical scheme of the invention is realized in that the source load cooperative control method for the park micro-grid and off-grid switching comprises the following steps: the method comprises the steps of S1, respectively deploying edge intelligent agents with local computing and communication capabilities in a distributed power supply, load nodes and energy storage devices of a micro-grid in a park, wherein each edge intelligent agent is integrated with a local state sensing unit, a lightweight dynamic prediction model and a neighborhood communication interface and is used as a basic decision execution unit for source-load cooperative control; s2, each edge agent acquires power output, load demand, environmental parameters and running state data of the node in real time through a loca