Wind and light storage cooperative control system based on source load matching

Abstract

The invention discloses a wind and light storage cooperative control system based on source load matching. The system aligns a source load storage network and communication information and carries out credible evaluation; wind, light and load are predicted, and uncertainty levels are given; calculating a matching degree and a short plate hot spot, and combining network side hard constraints such as voltage, power flow and reverse transmission to obtain an acceptable boundary; when the matching is qualified and the complementary energy is still available, identifying the complementary energy / energy shortage and the reason, quickly checking the transferable margin and loss by the edge, generating a routing plan containing retention time and action budget, and scheduling the stored energy and the flexible load in a rolling manner; meanwhile, the risk budget and standby are written back according to feasibility, prediction uncertainty and communication quality, gray release and automatic rollback are achieved in cooperation with degradation and playback scoring, and consumption and traceability are improved.

Description

Technical Field

[0001] This invention relates to the fields of power system automation control and new energy grid-connected operation technology, and particularly to the coordinated control of wind power, photovoltaics, energy storage, flexible interconnection devices, and flexible loads in distribution networks, microgrids, and industrial park source-grid-load-storage scenarios. More specifically, it relates to a wind-solar-storage coordinated control system and its control method driven by "source-load matching assessment," bounded by the feasibility of grid-side hard constraints, and using risk budget write-back and playback consistency verification as closed-loop governance means. Background Technology

[0002] With the large-scale integration of distributed photovoltaic power, decentralized wind power, user-side energy storage, and adjustable loads into the distribution network, the operation of the distribution network exhibits characteristics such as strong fluctuations in new energy output, significant prediction errors, increased risks of power flow reversal and node voltage exceeding limits, and a rise in three-phase imbalance and power quality problems. Existing technologies typically employ a scheduling method of "day-ahead planning + intraday correction + real-time control," or a control strategy centered on energy storage smoothing and peak shaving. However, these approaches still have the following shortcomings in practical engineering implementation: First, source-load matching assessments often remain at the "paper matching" level of the energy or power layer, lacking unified and calculable indicators for different frequency band fluctuations, ramping capabilities, flexible resource coverage, and grid-side acceptability, making it difficult to quickly identify shortcomings and hotspots.

[0003] Secondly, even if many strategies show "matching qualified" in the time dimension, they will still have "surplus energy abandoned or lost" due to spatial constraints (such as node voltage limit, reverse power limit, section congestion, three-phase imbalance, power quality limit, etc.), which is the engineering pain point of "qualified but still wasted".

[0004] Third, existing technologies mostly rely on deterministic control or a single safety margin, lacking a risk budget closed loop that links with predictive uncertainty, communication quality, and feasibility verification results. This leads to either excessive conservatism and reduced absorption capacity, or frequent overreach and unstable control.

[0005] Fourth, the policy distribution, conflict arbitration, and abnormal degradation mechanisms at the main site-edge-device layer often lack verifiable closed loops and governance capabilities such as sandbox replay, A / B comparison, canary release, and automatic rollback, resulting in high policy deployment risks and difficulty in auditing and tracing.

[0006] Therefore, there is an urgent need for a collaborative control system that can connect "multi-dimensional matching assessment - grid-side hard constraints - surplus / deficient energy routing - risk budget write-back - trustworthy degradation - playback verification audit" to improve the consumption of new energy and reduce curtailment / loss while ensuring the feasibility of constraints. Summary of the Invention

[0007] The purpose of this invention is to provide a wind-solar-storage collaborative control system based on source-load matching. Under the conditions of prediction uncertainty and grid-side hard constraints, it can identify and manage scenarios where "matching is qualified but there is still surplus energy". Through edge rapid feasibility verification and routing state machine, it achieves collaborative control of "time migration priority and spatial migration supplementation". Through risk budget write-back and playback consistency scoring, it achieves policy adaptive convergence and verifiable governance, thereby improving the local consumption capacity of new energy, reducing power curtailment and losses, and improving the safety and traceability of distribution network operation.

[0008] To achieve the above objectives, this invention constructs a closed loop based on "source-load matching drive": Data reliability → Multi-timescale prediction → Multi-dimensional matching evaluation → Calculation of network-side acceptable boundary → Remaining / deficient energy identification and cause code → Edge feasibility verification → RoutePlan routing and scheduling → Risk budget write-back and backup linkage → Arbitration and degradation → Receipt and playback verification → Audit evidence chain solidification → Rewrite-back convergence.

[0009] Among them, "matching assessment" not only evaluates the supply and demand relationship in the time domain, but also combines frequency domain fluctuations, flexible resource coverage, and network-side acceptability; "network-side acceptability boundary" forms a unified executable boundary through the intersection of hard constraints; "redundant routing" uses the TransferMargin with rapid edge verification as the feasibility basis, and uses action budget and switching cost to suppress frequent operations; "risk budget write-back" incorporates feasibility scores, prediction uncertainty, and communication quality into the convergence rules; and "replay verification audit" supports gray release, automatic rollback, and traceable governance with replay consistency scores.

[0010] More specifically, the present invention adopts the following technical solutions: I. System Components: A wind-solar-storage collaborative control system based on source-load matching, the system comprising at least the following functional units (which can be deployed at a master station, substation, or edge gateway): Data processing unit: performs time alignment and reliability assessment on the source-side, load-side, energy storage, and grid-side measurements and communication status, and outputs data quality Q_score and communication quality comm_quality.

[0011] Prediction Unit: Outputs the predicted sequence of wind and solar power output and load in the prediction time domain H, and outputs the uncertainty level CI_level.

[0012] Matching evaluation unit: Calculates the overall matching degree S_total and outputs the weakest link and hotspot. Preferably, S_total is synthesized by weighting time shape matching S_shape, frequency domain matching S_freq, flexible coverage S_flex, and network-side coverage S_grid; S_freq assesses the degree of matching between net load fluctuation amplitude and resource tracking capability based on multiple frequency bands B_k.

[0013] Grid-side acceptance calculation unit: Calculates the grid-side acceptance upper limit P_accept_max and the allowable reverse power P_export_allowed based on constraints such as voltage, line load, cross-sectional power flow, reverse transmission, imbalance and power quality. Preferably, P_accept_max is calculated based on the intersection of hard constraints and is the minimum value of the upper limit of each constraint; P_export_allowed is the minimum value of the back-transmission limit and the back-transmission margin obtained by mapping the cross-section / voltage margin.

[0014] The surplus / deficit energy discrimination unit outputs the surplus energy power P_surplus and the deficit energy power P_deficit when the triggering criterion is met, and also outputs the space constraint cause code constraint_cause and the deficit value factor Value. The cause code includes at least voltage upper limit, reverse transmission limit, section congestion, imbalance constraint and power quality constraint.

[0015] Edge feasibility verification unit: Outputs TransferMargin(i→j), incremental loss estimate loss_est and feasibility score for candidate energy surplus point i and energy deficiency point j; Preferably, the candidate migration power is checked item by item by using the sensitivity matrix or linearized power flow approximation to obtain the upper limit P_limit_ corresponding to each constraint, and TransferMargin(i→j)=min{P_limit_} is set.

[0016] Routing and Scheduling Unit: Generates RoutePlan when feasibility_score meets the threshold and loss_est does not exceed the threshold, and generates and distributes control plans including energy storage charging and discharging, flexible load adjustment, reactive power support, power generation limitation and RoutePlan on a rolling basis at the day-ahead, intraday and real-time scales.

[0017] Risk budget write-back unit: Allocate opportunity constraints risk budget ε_k and reserve requirement reserve_req, and perform write-back updates on ε_k and reserve_req based on feasibility_score, CI_level and comm_quality to achieve conservative adaptive convergence.

[0018] Arbitration and Degradation Unit: Performs hierarchical arbitration and switches between multiple degradation levels when resource conflicts or data / communication anomalies occur, and writes back degradation, rate limiting or interlocking events to trigger recalculation; Preferably, spatial migration is prohibited during high-level downgrades, with only time migration and critical load protection retained.

[0019] Replay Verification and Audit Unit: Creates a traceable record of data, models, plans, instructions and receipts, and performs strategy A / B comparison, canary release and automatic rollback based on the replay consistency score replay_score; preferably, replay_score includes at least constraint violation cost item, deviation cost item and action cost item, and triggers rollback and risk budget tightening when the degradation exceeds the threshold within the sliding window.

[0020] II. Key Triggering Criteria and RoutePlan Mechanism: The "Remaining Energy Despite Completion of Matching" trigger criterion is as follows: When S_total≥S_th, and P_surplus(t)≥P_th exists within the prediction time domain H, and constraint_cause belongs to the spatial constraint set, a remaining energy routing link is initiated. This criterion is used to specifically address "remaining energy discarding / loss caused by spatial constraints" under the background of overall successful matching.

[0021] Edge feasibility criterion: Spatial migration candidates are allowed to be written into RoutePlan only if feasibility_score≥F_th and loss_est≤Loss_th; otherwise, it degenerates into time migration or limited release strategy.

[0022] RoutePlan state machine: RoutePlan is managed using INIT / ACTIVE / RELEASE state machine; ACTIVE entry conditions include at least the above trigger criteria and feasibility criteria; RELEASE entry conditions include at least constraint touch-out, communication quality degradation, equipment rate limiting / interlocking, revenue degradation exceeding the threshold or TTL expiration; frequent switching is limited by t_hold and hysteresis threshold.

[0023] Action Budget and Vibration Suppression: RoutePlan includes an Action Budget, which limits the number of switching operations within the sliding window T_win to ≤ N_sw_max, the number of route switching operations to ≤ N_route_max, and the interconnect power change rate to ≤ (dP / dt)_max. When the action budget is insufficient, spatial migration is automatically suppressed and time migration is given priority to reduce frequent operations and secondary disturbances.

[0024] Risk budget writeback convergence: ε_k is limited to [ε_min, ε_max] and the change per period is ≤ Δε_max; when feasibility_score decreases, CI_level increases, or comm_quality decreases, ε_k is tightened and reserve_req is increased; when there are no violations of key constraints for N consecutive periods and comm_quality is not lower than the threshold, ε_k is relaxed and reserve_req is decreased to achieve an adaptive balance of "safety-absorption".

[0025] Replay Consistency-Driven Governance: When the replay_score deteriorates beyond the threshold ΔR_th within the sliding window T_eval, the policy is rolled back to a previous version, and risk budget tightening and scheduling recalculation are triggered, thereby ensuring that the policy deployment is verifiable and rollbackable.

[0026] III. Methodology Section: The control methods based on the above system include at least the following: Collect and align source-load-storage network and communication data, and perform a reliability assessment to obtain Q_score and comm_quality; Predict wind and solar power output and load, and output the prediction sequence and CI_level; Calculate S_total and output hotspot; Calculate P_accept_max and P_export_allowed based on the intersection of hard constraints; When the triggering criterion is met, output P_surplus, P_deficit, constraint_cause, and Value; Perform edge feasibility verification to obtain TransferMargin, loss_est and feasibility_score; When feasibility_score and loss_est meet the thresholds, a RoutePlan containing ttl, t_hold, ActionBudget and J_action is generated and executed on a day-ahead / day-intraday / real-time rolling schedule. Based on feasibility_score, CI_level, and comm_quality, write back and update ε_k and reserve_req; In case of conflict or anomaly, arbitration and degradation are performed, and a recalculation is triggered by writing back; Based on replay_score, A / B comparison, canary release, and automatic rollback are performed, and an audit evidence chain is recorded.

[0027] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By evaluating S_total and hotspots through multi-dimensional matching, the source-load mismatch bottleneck can be calculably located, covering both time shape and frequency domain fluctuations and flexible coverage, thereby improving the targeting of control strategies.

[0028] 2. By calculating P_accept_max and P_export_allowed through the intersection of grid-side hard constraints, voltage, line load, cross-section, reverse transmission, imbalance and power quality are incorporated into a unified executable boundary to avoid "good calculations but exceeding the limits in implementation".

[0029] 3. To address the engineering pain point of "matching qualified but still having surplus energy", a feasible energy routing mechanism is constructed by using surplus energy / deficient energy identification + reason code + edge transfer margin for rapid verification. Under the conditions of feasibility and loss threshold, surplus energy is guided to be stored locally or migrated to the energy-deficient location, thereby reducing power curtailment and loss.

[0030] 4. By using the RoutePlan state machine, t_hold hysteresis, and ActionBudget action budget, the switching cost J_action is introduced to effectively suppress frequent switching and secondary disturbances, thereby improving operational stability and equipment lifespan friendliness.

[0031] 5. By using the write-back convergence mechanism of risk budget ε_k and reserve reserve_req, feasibility scoring, uncertainty and communication quality are incorporated into a unified closed loop, which can adaptively balance between "safety and absorption", reduce the frequency of exceeding limits and improve the absorption level.

[0032] 6. The replay_score supports sandbox replay, A / B comparison, canary release and automatic rollback, and forms an audit evidence chain, making the strategy deployment verifiable, traceable and reproducible, significantly reducing the risk of engineering applications.

[0033] 7. Through layered arbitration and multi-level degradation mechanisms, critical loads and safety boundaries can still be guaranteed in the event of communication anomalies, equipment rate limiting, or interlocking, thereby enhancing the robustness and resilience of the system. Detailed Implementation

[0034] The invention will be more readily understood by referring to the following detailed description of preferred embodiments and included examples. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains, and in case of any conflict, the definitions in this specification shall prevail.

[0035] In a preferred embodiment, the wind-solar-storage collaborative control system based on source-load matching of the present invention is applicable to distribution network areas, park microgrids, or source-grid-load-storage scenarios containing distributed photovoltaic, decentralized wind power, user-side energy storage, flexible loads, and flexible interconnection devices. The system can adopt a hierarchical deployment form of master station-edge-device. The main station is used for day-ahead / intra-day plan generation, strategy configuration and audit archiving, while the edge side is used for real-time calculation, rapid feasibility verification, routing state machine advancement and arbitration degradation. The equipment side includes energy storage converters, grid-connected inverters, flexible load controllers, reactive power compensation devices and flexible interconnection devices (such as SOP or DC interconnection devices), etc., used to receive active / reactive power control commands and send back receipts. The system operates in a closed loop with a fixed control cycle Δt, which can be any configurable value from 1 second to 5 minutes. A unified data object is used to carry the measurement, prediction, decision and feedback information for each cycle.

[0036] Specifically, the system first collects data on wind power, photovoltaics, energy storage SOC and charging / discharging power, load and its adjustable bandwidth, voltage, current and cross-sectional power flow, reverse power, three-phase imbalance index and power quality index through data acquisition and time alignment. It also monitors the packet loss rate, latency and heartbeat online rate of the communication channel. After time alignment, it enters the reliable evaluation process. The preferred methods for data reliability assessment include boundary crossing detection, mutation detection, consistency verification of adjacent measurement points or energy conservation consistency verification, etc., to output a data quality score Q_score; at the same time, communication quality comm_quality is formed based on packet loss rate, latency, retransmission count, and heartbeat status. Subsequent modules prioritize Q_score and comm_quality when calculating and arbitrating to determine whether to enter restricted operation or trigger a degradation strategy, thereby avoiding uncontrollable control outputs in the event of measurement distortion or unreliable communication.

[0037] In a preferred embodiment, the data quality score Q_score is used to characterize the validity and consistency of the measurement data within the current control period, and is preferably calculated in the manner of "rule detection → sub-score normalization → weighted fusion → missing measurement penalty"; Taking a single measurement point x as an example, its original measurement value is denoted as x_raw(k), the median within the historical sliding window W_q (e.g., 10 to 120 periods) is denoted as med(x), and the median absolute deviation within the sliding window is denoted as MAD(x); Sub-rules must include at least: Out-of-bounds detection r_rng (if x_raw exceeds the physical range [x_min, x_max], then r_rng=0, otherwise r_rng=1); Mutation detection r_jmp (if |x_raw(k)-x_raw(k-1)|>Δx_max then r_jmp=0, otherwise r_jmp=1); Deviation detection r_dev (if |x_raw(k)-med(x)|>λ·MAD(x), then r_dev=0, otherwise r_dev=1, where λ can be 3 to 8); Consistency check r_cons (e.g., source-side / energy storage / load energy conservation consistency or adjacent measurement point consistency; if the consistency error e_cons exceeds the threshold E_cons, then r_cons=0; otherwise, r_cons=1). Furthermore, each rule can be mapped to a continuous sub-fraction of 0 to 1, for example, s_rng=r_rng, s_jmp=r_jmp, s_dev=max(0,1-|x_raw-med(x)| / (λ·MAD(x)+ε)), s_cons=max(0,1-e_cons / E_cons); The final quality score for measurement point x can be weighted and fused: Q_x = Σw_i·s_i, where w_i is a configurable weight and Σw_i = 1. The total quality score for the set of multiple measurement points X can be a weighted average Q_score = Σα_x·Q_x, with penalties applied for missing tests, timeouts, freezes, etc. (e.g., a penalty of (1-ρ_miss) for the missing test rate ρ_miss). When Q_score is lower than the threshold Q_th (e.g., 0.5 to 0.8), the system marks the period as "restricted trustworthy" and reduces the weight of space migration actions or directly triggers degradation in subsequent arbitrations.

[0038] In a preferred embodiment, communication quality (comm_quality) is used to characterize the availability of the communication channel. It is preferably calculated from packet loss rate (ρ_loss), round-trip time (RTT), jitter (J), and heartbeat online rate (ρ_hb), and mapped to the interval 0–1. For example, comm_quality can be: comm_quality = β1·(1-ρ_loss) + β2·clip(1-RTT / RTT_max) + β3·clip(1-J / J_max) + β4·ρ_hb, where clip(·) truncates the result to [0,1], RTT_max and J_max are engineering upper limits (e.g., RTT_max is 500ms–3000ms, J_max is 100ms–1000ms), and β1–β4 are weights that sum to 1. When comm_quality is below the threshold C_th (e.g., 0.4–0.7), the system preferably prohibits spatial migration or increases t_hold and hysteresis thresholds, entering a more conservative operating posture.

[0039] After completing the credible assessment, the prediction module outputs the predicted sequences of wind and light power and load within the prediction time domain H, and simultaneously outputs the prediction uncertainty level CI_level.

[0040] In a preferred embodiment, the prediction uncertainty level CI_level can be obtained by fusing "recent prediction residuals + meteorological confidence bandwidth + short-term fluctuation intensity". Taking the photovoltaic prediction error as an example, the normalized residual r_pv = RMSE(P_pv - P̂_pv) / (P_pv_rated + ε) is calculated within the sliding window W_e (e.g., 30 min to 24 h); similarly, the wind power residual r_w and the load residual r_l are calculated. The short-term fluctuation intensity can be represented by the energy of the high-frequency component of the net load or new energy output, e.g., v_hf = Std(HPF(P_pv)) or v_hf = Std(HPF(P_net)), where HPF is high-pass filtering / wavelet decomposition; If the meteorological prediction confidence interval width (such as cloud cover / irradiance confidence bandwidth) is available, then u_met = CI_width / CI_width_ref can be defined. The final uncertainty index U can take a weighted fusion: U = γ1·r_pv + γ2·r_w + γ3·r_l + γ4·v_hf + γ5·u_met, and U is mapped to a discrete level CI_level ∈ {L0, L1, L2, L3}. For example: U ≤ u0 is L0 (low uncertainty), u0 < U ≤ u1 is L1, u1 < U ≤ u2 is L2, U > u2 is L3 (high uncertainty), where u0 to u2 are configurable thresholds (e.g., 0.05, 0.10, 0.20). CI_level is used to drive the write-back of the risk budget ε_k and the reserve reserve_req: the higher the CI_level, the tighter ε_k and the higher reserve_req, and S_th and F_th can be increased synchronously to suppress aggressive space migration.

[0041] CI_level can be comprehensively determined by rolling statistics of recent prediction residuals, meteorological confidence interval width, short-term fluctuation intensity, and abnormal weather indication, etc., and can be discretized into multiple levels or represented by a continuous index; The prediction results enter the matching evaluation module. The matching evaluation module calculates the comprehensive matching degree S_total within the prediction time domain H and outputs the short-board hotspot. Among them, S_total is preferably synthesized by multiple sub-indices according to weights, including: S_shape used to characterize the shape consistency of the net load curve and the reference trajectory in the time domain; S_freq used to characterize the matching relationship between the fluctuation amplitude Amp_net(B_k) of the net load on multiple frequency bands B_k and the system tracking ability Cap_follow(B_k); S_flex is used to characterize the SOC margin, charge / discharge power margin, and the extent to which the adjustable energy of flexible loads covers the required adjustable energy / power. The S_grid is used to characterize the coverage of renewable energy available output by the grid-side acceptable boundary; hotspots at least indicate the type of bottleneck, key time slice, and key location, so as to facilitate subsequent targeted governance.

[0042] In a preferred embodiment, the frequency domain matching sub-index S_freq uses a preset frequency band set B_k to decompose and evaluate the net load fluctuation. The frequency band set can be divided according to the time scale, for example, B1: 0.2~1min (second-level disturbance), B2: 1~5min (short-term fluctuation), B3: 5~15min (climbing fluctuation), B4: 15~60min (hourly fluctuation). Alternatively, the sampling period Δt can be set proportionally. For each frequency band B_k, the system extracts the corresponding frequency band component P_net^k(t) from the net load P_net(t) and calculates the amplitude index Amp_net(B_k)=P95(|P_net^k|) or Std(P_net^k). The resource tracking capability Cap_follow(B_k) is preferably obtained by summing the adjustable power margins of energy storage, flexible loads and interconnection devices in this frequency band. For example: Cap_follow(B_k) = Cap_bess(B_k) + Cap_flex(B_k) + Cap_link(B_k), where Cap_bess(B_k) can be determined by the PCS power limit, SOC margin and allowable rate constraint. Cap_flex(B_k) can be determined by the adjustable power and response time constant of the interruptible / peak-shiftable load; Cap_link(B_k) can be determined by the available capacity and operating budget constraints of the interconnect device; the bandwidth matching degree can be defined as: s_k=clip(1-Amp_net(B_k) / (Cap_follow(B_k)+ε)) or s_k=clip(Cap_follow(B_k) / (Amp_net(B_k)+ε)), and then sum up by weight S_freq=Ση_k·s_k. Hotspots can be output based on this. For example, when a certain frequency band k satisfies the maximum value of Amp_net(B_k) / Cap_follow(B_k*) and exceeds the threshold, hotspot=(“band bottleneck”,k*) is marked and used for subsequent priority allocation of energy storage power or restriction of spatial migration.

[0043] Meanwhile, the grid-side acceptance capacity calculation module calculates the grid-side acceptable upper limit P_accept_max and the allowable reverse transmission power P_export_allowed based on the hard constraints of distribution network operation. Preferably, the voltage upper limit, line load upper limit, cross-sectional power flow upper limit, reverse transmission limit, three-phase imbalance constraint and power quality constraint are mapped to the incremental acceptance upper limit P_limit_* respectively, and P_accept_max is calculated using the hard constraint intersection method, that is, P_accept_max=min{P_limit_U, P_limit_I, P_limit_F, P_limit_rev, P_limit_unb, P_limit_pqi}; The allowed reverse power P_export_allowed is preferably the minimum of the reverse power limit and the reverse margin obtained by mapping from the cross-sectional or voltage margin. The above calculation allows the grid-side boundary to be directly incorporated into control and routing decisions, avoiding the situation where energy balancing alone is insufficient for engineering implementation.

[0044] In a typical and critical application scenario, even if the overall matching degree S_total reaches the qualified threshold S_th, the system may still "match qualified but still have surplus energy discarded or lost" due to network-side spatial constraints. To address this, the system sets up a surplus energy / deficit energy discrimination and value assessment module, which starts the surplus energy routing link when the following triggering criteria are met simultaneously: S_total≥S_th, and there is surplus energy power P_surplus(t)≥P_th in the predicted time domain H, and the constraint reason code constraint_cause belongs to the spatial constraint set, which includes at least voltage upper limit, reverse transmission limit, cross-section congestion, imbalance constraint and power quality constraint. The surplus power P_surplus(t) can be obtained from the difference between the available renewable energy output and the grid-side absorption capacity and local absorption capacity; the deficit power P_deficit(t) can be obtained from the difference between the critical load demand and the local supply capacity; the deficit value factor Value is used to sort and select deficit points, and Value is preferably related to the critical load level, marginal cost of electricity purchase, congestion risk or power supply reliability weight; the constraint_cause is used to explain the root cause of surplus energy formation and is used for subsequent candidate deficit point screening, routing strategy selection and audit records.

[0045] When the backup power routing link is triggered, the edge fast feasibility verification module performs a fast verification of the candidate backup power point i and the power shortage point j, and outputs the transfer margin TransferMargin(i→j), the incremental network loss estimate loss_est and the feasibility score. Preferably, based on the sensitivity matrix or linearized power flow approximation, the candidate migration power is checked item by item and the power upper limit P_limit_* corresponding to each hard constraint is obtained, so that TransferMargin(i→j)=min{P_limit_*}; The system allows candidates (i→j) to be written into the energy routing plan RoutePlan only when feasibility_score≥F_th and loss_est≤Loss_th. Otherwise, the spatial migration will degenerate into time migration or a power limiting strategy. The time migration preferably includes energy storage charging to absorb surplus energy, flexible load boosting to absorb surplus energy, or adjusting reactive power support to expand the local acceptable margin. F_th and Loss_th are configurable thresholds to adapt to different levels of engineering constraints.

[0046] In a preferred embodiment, the feasibility score is used to comprehensively characterize the feasibility and robustness of the candidate migration path in the current state, and is preferably obtained by normalized weighting of components such as constraint margin, communication reliability, action budget and value benefit. For example, the normalized margin term m = clip(TransferMargin / ΔP_req) can be constructed, where ΔP_req is the candidate expected migration power; the communication term c = comm_quality; the data term q = Q_score; the action budget term a = clip(ActionBudget_remain / ActionBudget_total); and the revenue term v = clip(Value / Value_ref). The feasibility score can be taken as feasibility_score=θ1·m+θ2·c+θ3·q+θ4·a+θ5·v, and the result is truncated to [0,1]. θ1~θ5 are configurable weights and their sum is 1. To avoid false selection due to "high value but low safety margin", it is preferable to set a threshold gate: when m is lower than m_min or c and q are lower than their respective thresholds, it is directly judged as infeasible (feasibility_score is set to 0), and then a weighted score is performed.

[0047] The incremental loss estimate, loss_est, is preferably estimated by approximating the candidate migration power and equivalent resistance, or by using linearized power flow loss sensitivity. For example, it can be approximated as loss_est≈R_eq·(ΔP)^2 / (U_base^2+ε)·Δt, where R_eq is the equivalent resistance of the candidate path or the loss coefficient identified from historical operation. Alternatively, a first-order approximation can be made as loss_est≈K_loss·ΔP. The system writes the candidate (i→j) into the RoutePlan only when feasibility_score≥F_th and loss_est≤Loss_th; otherwise, it degenerates into time migration (such as energy storage charging, flexible load absorption) or power rationing. F_th and Loss_th can be adaptively adjusted according to CI_level, constraint_cause and degradation level, but their admission logic remains consistent to ensure feasibility and verifiability.

[0048] In a preferred embodiment, the edge rapid feasibility verification uses a sensitivity matrix or linearized power flow approximation to quickly verify the candidate migration power ΔP. Taking the candidate path i→j as an example, it is equivalent to injecting ΔP at node i and absorbing ΔP (or equivalent power transfer) at node j, and the incremental response of key constraints to ΔP is calculated. Voltage constraint mapping can be expressed as ΔU≈K_U·ΔP, where K_U is the linearization sensitivity (which can be obtained from offline power flow calculation, online identification, or historical disturbance regression). Then, the upper limit of power corresponding to voltage constraint can be P_limit_U=min overnodes {(U_max-U_base) / |K_U|}. Line load / current constraint mapping can be achieved using ΔI≈K_I·ΔP, resulting in P_limit_I=min{(I_max-I_base) / |K_I|}; cross-sectional power flow constraint mapping can be achieved using ΔF≈K_F·ΔP, resulting in P_limit_F=min{(F_max-F_base) / |K_F|}; the upper limit of the reverse feed constraint P_limit_rev can be taken as the remaining margin of the reverse feed limit; three-phase imbalance and power quality constraints can be obtained by ΔK_unb≈K_unb·ΔP and ΔPQI≈K_pqi·ΔP, respectively, resulting in P_limit_unb and P_limit_pqi. In all the above upper limit calculations, ε is added to the denominator to avoid division by zero, and a conservative reduction coefficient κ∈(0,1) (e.g., 0.5~0.9) is used when the sensitivity confidence is insufficient. Finally, the TransferMargin(i→j)=min{P_limit_U, P_limit_I, P_limit_F, P_limit_rev, P_limit_unb, P_limit_pqi} is obtained by hard constraint intersection. When reactive power / voltage coordination needs to be considered, ΔP can be expanded to (ΔP,ΔQ) and the corresponding upper limit is calculated with the joint sensitivity matrix, but this does not affect the constraint intersection principle of "taking the minimum upper limit as TransferMargin".

[0049] When the feasibility verification is passed, the routing and scheduling module generates a RoutePlan and manages it using a state machine. The RoutePlan includes at least the time to live (ttl), minimum hold time (t_hold), hysteresis threshold, action budget (ActionBudget), and switching cost (J_action). The state machine includes at least an initialization state INIT, an activation state ACTIVE, and a release state RELEASE. The entry condition for ACTIVE includes at least the aforementioned trigger criterion being met and feasibility_score ≥ F_th. The entry condition for RELEASE includes at least constraint touch, communication quality comm_quality being lower than the threshold, device rate limiting or interlocking, revenue decline exceeding the threshold, or TTL expiration. Frequent state switching is limited by the minimum hold time t_hold and hysteresis threshold. The ActionBudget constrains the intensity of actions within the sliding window T_win, preferably limiting the number of switching operations to no more than N_sw_max, the number of route switching to no more than N_route_max, and the power change rate of the flexible interconnect device to no more than (dP / dt)_max. When the ActionBudget is insufficient, the system automatically suppresses spatial migration and prioritizes time migration, thereby reducing secondary disturbances and equipment wear caused by frequent switching. The switching cost J_action penalizes frequent switching in multi-candidate route comparison and scheduling decisions, prioritizing the option with fewer actions and less disturbance when the benefits are similar.

[0050] Under the aforementioned RoutePlan constraints, the system achieves day-ahead, intraday, and real-time rolling control through multi-timescale collaborative scheduling: The current scale is used to arrange energy storage energy transfer, flexible load planning and necessary reactive power support strategies, and to reserve for backup; Intraday scales are used to revise day-ahead plans and update RoutePlan candidates based on updated forecasts and constraint boundaries; The real-time scale performs rapid feasibility verification, state machine advancement, arbitration, and instruction issuance in a short cycle Δt. After execution on the equipment side, the actual execution amount, current limiting status, and protection / interlock status are returned as the basis for feasibility verification and edge judgment in the next cycle. To achieve an adaptive balance between "safety and consumption", the system sets up a risk budget write-back and reserve linkage mechanism to dynamically update the opportunity-constrained risk budget ε_k and reserve requirement reserve_req, preferably limiting ε_k to [ε_min, ε_max] and ensuring that the change in each rolling cycle does not exceed Δε_max; When feasibility_score decreases, uncertainty level CI_level increases, or comm_quality decreases, tighten ε_k and increase reserve_req; When there are no violations of the critical network-side constraints for N consecutive rolling cycles and comm_quality is not lower than the threshold, ε_k is relaxed and reserve_req is reduced, thereby forming a closed-loop convergence between feasibility and economy.

[0051] When resource conflicts, data / communication anomalies, or equipment flow limiting / interlocking occur, the system uses the arbitration and degradation module to perform layered arbitration and switch between multiple degradation levels. The preferred arbitration priority is: network-side security constraints take precedence over equipment protection, equipment protection takes precedence over energy storage safety belts, energy storage safety belts take precedence over critical load protection, and critical load protection takes precedence over economic efficiency. At higher degradation levels, the system prohibits spatial migration and retains only time migration and critical load protection strategies. Simultaneously, degradation events are written back to trigger risk budget tightening and scheduling recalculation to ensure that safety boundaries and critical power supply needs are maintained even under abnormal conditions. All degradation, current limiting, and interlocking events and their cause codes are written into the audit evidence chain to support post-event traceability and strategy optimization.

[0052] To reduce the risk of policy changes going live and to achieve verifiable governance, the system adopts replay verification, A / B comparison, canary release, and automatic rollback mechanisms. The replay consistency score (replay_score) includes at least a constraint violation cost item, a deviation cost item, and an action cost item. The constraint violation cost item is used to penalize voltage over-limit, cross-section congestion, back-feeding over-limit, imbalance over-limit, and power quality over-limit. The deviation cost term is used to penalize supply and demand deviations or tracking errors, while the action cost term is used to penalize the number of switching operations, the number of routing changes, and the power change rate. When the replay_score deteriorates beyond the threshold ΔR_th within the sliding window T_eval, the system automatically rolls back to the historical policy version and simultaneously triggers risk budget tightening and scheduling recalculation to ensure that policy iteration is controllable, rollbackable, and auditable. Through the above closed loop, the prediction residuals, constraint violations, number of actions and communication quality during the operation are solidified into statistics and continuously written back, thereby driving the parameters of ε_k, reserve_req and RoutePlan to gradually converge adaptively. Ultimately, under the premise of satisfying the feasibility of grid-side hard constraints, the renewable energy absorption capacity is improved and the curtailment and loss are reduced.

[0053] In a preferred embodiment, the replay consistency score (replay_score) is used to measure the overall performance of the strategy within the replay or online evaluation window. It includes at least a constraint violation cost term (J_cons), a deviation cost term (J_dev), and an action cost term (J_act). The constraint violation cost term (J_cons) can be penalized separately for voltage exceeding limits, section congestion, backfeed exceeding limits, imbalance exceeding limits, and power quality exceeding limits. For example, piecewise linear or quadratic penalties can be used: for any constraint y (such as voltage U), if y ≤ y_max, then a penalty of 0 is applied; if y > y_max, then a penalty of κ_y·(y-y_max)^2 is applied, and different weights κ_y are set for different constraints.

[0054] The deviation cost term J_dev can penalize supply and demand deviations or tracking errors, for example, J_dev=κ_e·Σ|P_track-P_ref| or κ_e·Σ(P_track-P_ref)^2. The action cost term J_act can penalize the number of switching operations, route switching operations, and power change rate, for example, J_act=κ_sw·N_sw+κ_rt·N_route+κ_ramp·Σ|ΔP|. The final score can be replay_score=-(J_cons+J_dev+J_act) or mapped to the 0-100 score range after baseline normalization and within a sliding window. Aggregation is performed within T_eval by mean, quantile, or weighted sum (e.g., taking the average score within the window or the worst 10% quantile to emphasize risk). When the replay_score deteriorates more than the threshold ΔR_th relative to the historical baseline within T_eval (e.g., a decrease of 10% to 30%), a rollback is triggered: the rollback is to the previous stable policy version, and risk budget tightening (reducing ε_k and increasing reserve_req) and RoutePlan recalculation are triggered. If continuous rollbacks still cannot restore the score above the threshold, a high-level degradation with prohibited space migration is implemented as a fallback strategy, and the rollback reason code and parameter changes are written into the audit evidence chain.

[0055] Example 1 In a preferred embodiment, to balance engineering feasibility and adaptability to different scale distribution network / park microgrid scenarios, the key thresholds and periodic parameters of the system can be set within the following ranges and support online configuration and remote distribution: The control period Δt can be 1s to 300s, preferably 1s to 10s for edge fast verification and arbitration degradation, and 10s to 60s for real-time rolling scheduling. The forecast time domain H can be 5 min to 48 h, with the preferred intraday rolling H being 15 min to 6 h, and the day-ahead planning H being 24 h to 48 h; The overall matching degree threshold S_th can be 0.60 to 0.95, preferably 0.75 to 0.90; The residual power triggering threshold P_th can be 1% to 20% of the rated access capacity, preferably 3% to 10%, and can be adaptively set according to the feeder capacity, the minimum adjustment step size of the equipment and the error bandwidth. The feasibility score threshold F_th can be 0.50 to 0.95, preferably 0.70 to 0.90; The loss threshold Loss_th can be 0.5% to 10% of the energy corresponding to the candidate migration power, preferably 1% to 5%, and can be adjusted in conjunction with peak and off-peak electricity prices, congestion level, and energy shortage value.

[0056] To suppress frequent actions and secondary disturbances, the minimum hold time t_hold of RoutePlan can be 10s to 1800s, preferably 60s to 600s; The time-to-live (TTL) can be 2 min to 4 h, preferably 10 min to 60 min, and should be released early when communication quality deteriorates or when constraints are reached. The hysteresis threshold can be set at 5% to 30% of the voltage margin, reverse feed margin, or cross-sectional margin, preferably 10% to 20%. The ActionBudget preferably uses a sliding window T_win constraint, where T_win can be 5 min to 4 h, and preferably 15 min to 60 min. The maximum number of on / off / switching actions within the window, N_sw_max, can be 1 to 20 times, preferably 2 to 8 times; The maximum number of route switching attempts, N_route_max, can be 1 to 20 times, with 2 to 10 times being preferred. The upper limit of the power change rate (dP / dt)_max of the flexible interconnect device or key execution channel can be 0.1% / s to 10% / s of the rated power, preferably 0.5% / s to 3% / s; The switching cost J_action can be implemented by weighting the number of switching operations, power change rate, and switching amplitude. The weights can be configured in the range of 0.1 to 10, so that when the benefits are similar, the candidate RoutePlan with fewer actions is selected first.

[0057] Regarding risk budget write-back, the opportunity-constrained risk budget ε_k can be limited to the range of [ε_min, ε_max], where ε_min can be 0.001 to 0.05 and ε_max can be 0.05 to 0.30; the change limit Δε_max for each rolling cycle can be 0.001 to 0.05, preferably 0.005 to 0.02; When "no violation of critical constraints" is used as a relaxation condition for N consecutive cycles, N can be 3 to 200, preferably 5 to 30. Furthermore, when the prediction uncertainty level CI_level increases or the communication quality comm_quality decreases, N is automatically increased or ε_max is decreased to enhance conservatism. The reserve requirement reserve_req can be determined based on critical load classification, maximum prediction error, TransferMargin margin, and degradation level linkage. Its value can be 0% to 50% of the critical load power, preferably 5% to 25%, and is increased during high-level degradation (prohibited spatial migration) to ensure safety boundaries and critical power supply.

[0058] In terms of replay verification and automatic rollback, the replay consistency score replay_score can be calculated within a sliding window T_eval, where T_eval can be 5 min to 72 h, preferably 30 min to 12 h. The score degradation threshold ΔR_th can be set to 5% to 50% of the historical baseline, preferably 10% to 30%. When the replay_score degrades beyond ΔR_th within T_eval, a rollback to the historical policy version is triggered, along with risk budget tightening and scheduling recalculation. All of the above parameters can be fixed as configurable items in the strategy configuration file or parameter table, and their change time, change reason and effective scope can be recorded in conjunction with the audit evidence chain to facilitate reproduction and traceability.

[0059] Example 2 In a preferred embodiment, in order to enable the system to maintain an adaptive balance of "safety-absorption-controllable action" under different seasonal weather conditions, different load patterns and different communication reliability, the present invention sets adaptive update rules for key thresholds and window parameters, and writes the update results into the audit evidence chain to support reproduction and traceability.

[0060] The adaptive rule includes at least the following: Adaptive comprehensive matching threshold S_th: S_th is used to determine whether to enter the "matching qualified but still have remaining energy" governance link; Preferably, S_th is adjusted in conjunction with the prediction uncertainty level CI_level and the communication quality comm_quality: When CI_level increases or comm_quality decreases, increase S_th to reduce aggressive routing under uncertain conditions; When CI_level decreases and comm_quality remains stable, S_th can be appropriately reduced to expand the governable coverage. Furthermore, the S_th baseline can be switched according to the season / weather cluster: a higher S_th baseline is set during seasons with high incidence of extreme weather or severe convective weather clusters, and a lower S_th baseline is set during seasons with stable weather.

[0061] Adaptive residual energy trigger threshold P_th: P_th is used to filter false triggers caused by small residual energy noise and metering errors; Preferably, P_th is linked to the grid constraint margin, the minimum adjustment step size of the equipment, and the load level: when the grid-side margin is small (e.g., P_accept_max is close to zero margin), the equipment adjustment step size is large, or the communication quality deteriorates, P_th is increased to avoid frequent triggering. When the energy shortage value (critical load energy shortage) is high and the grid margin is sufficient, reduce P_th to more actively guide the utilization of surplus energy; Furthermore, P_th can be adaptively adjusted according to the time of day: P_th can be appropriately reduced during the midday photovoltaic peak or the evening peak energy shortage period, and P_th can be appropriately increased during periods of weak fluctuation or supply and demand balance.

[0062] Adaptation of feasibility scoring threshold F_th and loss threshold Loss_th: F_th and Loss_th are used for the admission of spatial migration candidates; Preferably, when constraint_cause is a hard constraint (such as voltage limit or cross-sectional congestion) and the constraint touches the limit frequently, increase F_th and decrease Loss_th to make the admission more stringent; When constraint_cause is a relatively mild constraint (such as slight imbalance or power quality slightly close to the threshold) and the system is running stably, F_th can be appropriately reduced and Loss_th relaxed to improve absorption. Furthermore, Loss_th is linked to the energy shortage value: the higher the value (critical load guarantee, high electricity purchase cost, high congestion risk), the allowable Loss_th can be appropriately increased; the lower the value, the tighter the Loss_th is to reduce ineffective migration and losses.

[0063] Adaptive minimum hold time t_hold, hysteresis threshold, and action budget ActionBudget: To suppress frequent switching, t_hold and hysteresis threshold can be adapted to action density and communication quality; When the number of actions within the sliding window T_win approaches its limit or comm_quality decreases, increase t_hold and the hysteresis threshold to make the state machine more stable. When the action density is low and the system is stable, moderately decrease t_hold and the hysteresis threshold to improve response speed. The ActionBudget upper limits (N_sw_max, N_route_max, (dP / dt)_max) are preferably linked to the equipment health status, temperature rise / load rate, and energy storage life constraints; When it is detected that the device is experiencing current limiting, excessive temperature rise, or rapid lifespan depletion, N_sw_max, N_route_max, or (dP / dt)_max will be automatically reduced to make the control more gentle.

[0064] Adaptive convergence rules for risk budget ε_k and reserve reserve_req: ε_k and reserve_req preferably adopt a "tightening / relaxing dual triggering" mechanism, which tightens ε_k and increases reserve_req when any tightening triggering condition occurs; Tightening trigger conditions include at least: a decrease in feasibility_score, an increase in CI_level, a decrease in comm_quality, a violation of critical constraints, or a deterioration in the replay_score of the consistency score. When the relaxation triggering conditions are met, ε_k is relaxed and reserve_req is reduced. The relaxation triggering conditions include at least: no violation of key constraints for N consecutive periods, comm_quality is not lower than the threshold, and replay_score is stable or improved. Preferably, N is adaptive to the season / weather cluster: N is increased under extreme weather clusters and decreased under stable weather clusters to achieve more robust risk control.

[0065] Adaptive linkage between degradation level and parameters: When the arbitration and degradation unit enters a higher degradation level (e.g., space migration is prohibited), the system prefers to synchronously execute parameter linkage, increase S_th and P_th, increase F_th and tighten Loss_th, increase t_hold and hysteresis threshold, lower the upper limit of ActionBudget, tighten ε_k and increase reserve_req, so as to switch to a more conservative and safer control attitude as a whole under abnormal conditions. When the degradation is lifted and the stability condition of N consecutive cycles is met, it gradually recovers in the opposite direction.

[0066] Replay consistency score replay_score driven parameter writeback: To achieve verifiable governance, the system splits replay_score into constraint violation cost item, deviation cost item and action cost item, and uses them for different parameter writebacks respectively; When the constraint violation cost increases, ε_k is tightened first, F_th is increased, and Loss_th is tightened. When the deviation cost increases, prioritize adjusting S_th and P_th and optimize RoutePlan candidate selection; When the action cost increases, prioritize increasing t_hold, increasing the hysteresis threshold, and decreasing the ActionBudget upper limit. The above write-back results, along with the corresponding trigger code, are written into the evidence chain to support post-event review and strategy iteration.

[0067] Through the above-mentioned adaptive recommendation rules, this invention can automatically adjust the thresholds and windows of "trigger-admission-execution-writeback" under different operating conditions, forming an interpretable, auditable, and reproducible engineering closed loop. This improves the utilization rate of spare energy and the critical load guarantee capability while ensuring the feasibility of network-side hard constraints, and reduces the risk of frequent actions and policy deployments.

[0068] The examples described herein are merely illustrative, intended to explain some features of the methods described herein. The appended claims are intended to claim the broadest possible scope, and the embodiments presented herein are merely illustrative of selected implementations based on combinations of all possible embodiments. Therefore, the applicant intends that the appended claims are not limited by the selection of examples illustrating the features of the invention. Some numerical ranges used in the claims also include sub-ranges within them, and variations within these ranges should be interpreted, where possible, as covered by the appended claims.

Claims

1. A wind-solar-storage collaborative control system based on source-load matching, characterized in that, For distribution networks, microgrids, or industrial park source-grid-load-storage scenarios, the system includes interconnected components: (1) Data processing unit, used to perform time alignment and reliability assessment on source-side output, load-side load, energy storage status, grid-side operation measurement and communication status, and output data quality flag Q_score and communication quality comm_quality; (2) Prediction unit, used to output the prediction sequence of wind and solar power output and load in the prediction time domain H and the uncertainty level CI_level; (3) Matching evaluation unit, used to calculate the comprehensive matching degree S_total based on the predicted sequence and output the shortcoming hotspot; (4) Grid-side acceptance calculation unit, used to calculate the grid-side acceptance upper limit P_accept_max and the allowable reverse power P_export_allowed based on voltage, line load, cross-sectional power flow, reverse transmission, imbalance and power quality constraints. (5) The surplus / deficit energy discrimination unit is used to output the surplus energy power P_surplus and the deficit energy power P_deficit when the triggering criterion is met, and output the space constraint cause code constraint_cause and the deficit value factor Value. (6) Marginal feasibility verification unit, used to output the transfer margin (i→j), incremental loss estimate (loss_est), and feasibility score for candidate energy surplus point i and energy deficit point j. (7) A routing and scheduling unit, used to generate an energy routing plan RoutePlan when the feasibility_score meets the threshold, and to generate and distribute control plans including energy storage charging and discharging, flexible load adjustment, reactive power support, power generation limitation and the RoutePlan on a rolling basis at the day-ahead, intraday and real-time scales. (8) Risk budget write-back unit, used to allocate opportunity constraint risk budget ε_k and reserve requirement reserve_req, and perform write-back update on ε_k and reserve_req according to feasibility_score, CI_level and comm_quality; (9) Arbitration and degradation unit, used to perform hierarchical arbitration and switch between multiple degradation levels when there is a resource conflict or data / communication anomaly, and to write back degradation, rate limiting or interlocking events to trigger recalculation; (10) Replay verification and audit unit, used to form a traceable record of data, models, plans, instructions and receipts, and to execute strategies A / B comparison, gray release and automatic rollback based on replay consistency score replay_score; The system satisfies the following: the prediction residuals, constraint violations, number of actions, and communication quality during operation are statistically generated by the playback verification and auditing unit, and used to drive the risk budget write-back unit to adaptively converge and update ε_k and reserve_req, so as to achieve a closed-loop balance between network-side constraint feasibility and action vibration suppression.

2. The wind-solar-storage collaborative control system based on source-load matching according to claim 1, characterized in that, The overall matching degree S_total is synthesized by weights from at least the time shape matching sub-index S_shape, the frequency domain matching sub-index S_freq, the flexible coverage sub-index S_flex, and the network side coverage sub-index S_grid. Among them, S_freq performs ratio matching between the net load fluctuation amplitude Amp_net(B_k) and the resource tracking capability Cap_follow(B_k) based on multiple frequency bands B_k, and outputs the hotspot indicating the weak link item.

3. The wind-solar-storage collaborative control system based on source-load matching according to claim 1, characterized in that, The surplus / deficient energy discrimination unit starts the surplus energy routing link when the following triggering criteria are met simultaneously: S_total≥S_th; and P_surplus(t)≥P_th exists in the prediction time domain H; and constraint_cause belongs to the spatial constraint set. The set of spatial constraints includes at least voltage upper limit, reverse transmission limit, cross-section congestion, imbalance constraint and power quality constraint, and the constraint_cause is used for energy shortage point screening or routing strategy selection.

4. The wind-solar-storage coordinated control system based on source-load matching according to claim 1, characterized in that, The grid-side acceptance calculation unit calculates P_accept_max using a hard constraint intersection method. That is, P_accept_max is the minimum value of multiple constraint upper limits, including at least the upper limit of voltage margin, upper limit of line load margin, upper limit of cross-sectional power flow margin, upper limit of reverse feed margin, upper limit of imbalance margin, and upper limit of power quality margin; and P_export_allowed is the minimum value of the reverse feed power limit and the reverse feed margin obtained by mapping the cross-sectional margin or voltage margin.

5. A wind-solar-storage collaborative control system based on source-load matching according to claim 1, characterized in that, The edge feasibility verification unit performs a constraint verification on the candidate migration power based on the sensitivity matrix or linear power flow approximation to obtain the power upper limit P_limit corresponding to each constraint. And let TransferMargin(i→j)=min{P_limit_ }; Furthermore, writing to the RoutePlan is only allowed when feasibility_score≥F_th and loss_est≤Loss_th; otherwise, spatial migration will degenerate into time migration or a limited-release strategy. F_th and Loss_th are configurable thresholds.

6. The wind-solar-storage coordinated control system based on source-load matching according to claim 1, characterized in that, The RoutePlan is managed in a state machine manner and includes at least an initialization state INIT, an activation state ACTIVE, and a release state RELEASE, and: the entry condition for ACTIVE includes at least the trigger criterion and feasibility_score≥F_th; The entry conditions for RELEASE include at least the constraint touch, comm_quality below the threshold, device current limiting / interlocking, revenue decline exceeding the threshold, or TTL expiration, and the state switching frequency is limited by the minimum hold time t_hold and hysteresis threshold. The RoutePlan includes an ActionBudget, which limits the number of switching operations within the sliding window T_win to no more than N_sw_max, the number of route switching operations to no more than N_route_max, and the interconnect power change rate to no more than (dP / dt)_max. When ActionBudget is insufficient, spatial migration is suppressed and temporal migration is prioritized.

7. A wind-solar-storage collaborative control system based on source-load matching according to claim 1, characterized in that, The risk budget write-back unit performs write-back of the opportunity-constrained risk budget ε_k according to the following rules: ε_k is limited to [ε_min, ε_max] and the change in each rolling period does not exceed Δε_max; When feasibility_score decreases, uncertainty level CI_level increases, or comm_quality decreases, tighten ε_k and increase reserve requirement reserve_req. When the critical network-side constraints are not violated and comm_quality is not lower than the threshold within N consecutive rolling cycles, ε_k is relaxed and reserve_req is reduced, where N, ε_min, ε_max and Δε_max are configurable parameters.

8. A wind-solar-storage collaborative control system based on source-load matching according to claim 1, characterized in that, The arbitration and degradation units make decisions according to the following priorities: grid-side security constraints take precedence over equipment protection, equipment protection takes precedence over energy storage safety belts, energy storage safety belts take precedence over critical load protection, and critical load protection takes precedence over economic efficiency. Furthermore, the multi-level degradation system includes at least a first level that allows spatial migration and a second level that prohibits spatial migration but only allows time migration. When entering the second level, a write-back and recalculation of the RoutePlan and risk budget is triggered.

9. A wind-solar-storage collaborative control system based on source-load matching according to claim 1, characterized in that, The replay consistency score (replay_score) includes at least a constraint violation cost item, a deviation cost item, and an action cost item, and is used for canary releases and automatic rollbacks. When the replay_score degrades beyond the threshold ΔR_th within the sliding window T_eval, a rollback to a historical policy version is triggered, along with risk budget tightening and scheduling recalculation. T_eval and ΔR_th are configurable parameters.

10. A wind-solar-storage coordinated control method based on source-load matching according to any one of claims 1-9, characterized in that, Performed by the system according to any one of claims 1-9, it at least includes: performing time alignment and reliability assessment on source-load-storage network and communication data; Output wind and solar load forecast sequences and uncertainty levels; Calculate the overall matching score and output the hotspots; Calculate the upper limit of the grid side that can be accepted and the allowable reverse transmission power based on the intersection of hard constraints; When the triggering criteria are met, output the surplus power, the power shortage, and the spatial constraint reason code, and perform edge feasibility verification. When the feasibility threshold is met, a RoutePlan containing TTL, T_hold, action budget, and switching cost is generated and executed on a rolling schedule at the day-ahead, intraday, and real-time scales. Based on feasibility scores, uncertainty levels, and communication quality, the risk budget and backup requirements are rewritten and updated. In the event of resource conflicts or anomalies, a downgrade arbitration will be performed and a recalculation will be triggered by writing back the data. Based on the replay consistency score, perform canary releases and automatic rollbacks, and form an audit evidence chain record.