专利权利要求支持段撰写

为支持权利要求1之可专利性与合法性，兹就该权利要求所涉“基于脉冲神经网络的低功耗目标检测芯片架构及训练—部署一体流程”的技术方案、技术效果与实验表征，依照《中华人民共和国专利法》第二十二条第一款至第三款（新颖性、创造性、实用性）及第二十六条第三款、第四款（清楚性、支持性与充分公开），并参照《专利审查指南》第二部分第四章（创造性判断，含意想不到的技术效果）、第二部分第九章（涉及计算机程序的发明的审查）之标准，作如下论证与说明：

一、技术方案之概括与技术问题 权利要求1主张的技术方案系以事件驱动的脉冲神经网络（SNN）为核心，提供面向目标检测任务的专用低功耗芯片架构与与之配套的训练—部署一体化流程。方案的关键技术特征包括：在芯片层面设置脉冲卷积/注意力算子与近存储计算耦合的存储层级、异步片上互连与事件队列管理机制、面向脉冲时域编码的推理时序控制，以及在算法与工具链层面提供硬件感知的脉冲化训练（含替代梯度、时序编码/解码、量化与稀疏约束）、映射-编译-校准一体流程与端到端的检测前后处理链路在芯片侧闭环执行。该组合旨在解决现有边缘端低算力、严苛能耗预算条件下目标检测精度显著退化、能效欠佳、以及训练成果与芯片部署割裂导致的不可复现与不可达标等技术问题，属于以技术手段解决特定技术问题并获得可验证技术效果的技术方案，符合《审查指南》关于计算机程序相关发明需体现“技术特征—技术问题—技术效果”的审查要点。

二、新颖性与创造性之法律与技术论证

1. 新颖性（专利法第二十二条第一款） 在不特定指向单一现有技术文献的前提下，现有公开虽各自涉及：（i）基于ANN的低比特量化检测加速器；（ii）以分类为主的SNN推理加速电路；或（iii）离线脉冲化/蒸馏的方法学，但普遍未见同时具备以下区别特征的整体性方案之公开与教导：将面向目标检测的SNN算子集（含时序卷积/稀疏注意力）与近存储计算、异步事件互连、片上前后处理管线进行硬件级耦合，并以硬件感知的训练—部署一体流程将脉冲编码、替代梯度训练、稀疏/时延约束、量化与编译映射协同优化至同一芯片目标下，进而在低算力边缘情境中稳定实现检测精度与能效的兼顾。权利要求1据此具备区别于常见单点改进或松耦合流程的整体性技术特征，具新颖性。

2. 创造性（专利法第二十二条第二款；审查指南第二部分第四章）

• 最接近现有技术一般提供ANN-INT8或稠密CNN加速器，或提供以分类为主之SNN加速器，均未解决“在低算力边缘端实现检测任务且维持有效精度与显著能效”的复合技术目标。
• 本申请区别特征的组合并非对各单元的简单并列：一方面，事件驱动NoC与近存储计算在脉冲时域下与稀疏注意力/卷积的时空特性相匹配，降低有效访存与切换功耗；另一方面，硬件感知训练将替代梯度、时序编码、稀疏与量化目标以部署约束为正则项进行联合优化，确保模型在目标芯片上的时序与能耗预算内达到预期精度。该等特征之间存在功能上的相互作用与协同增益，而非线性叠加。
• 现有技术缺乏将上述电路/系统层与训练/工具链层在同一技术问题下进行闭环设计的明确启示。依审查“三步法”，上述区别特征针对客观技术问题所获得的技术效果（见下述实验表征）具有意想不到性，足以支持创造性之认定（参见审查指南关于“意想不到的技术效果可作为创造性有力证据”的规定）。

三、实验表征与技术效果之可验证性 为满足专利法第二十六条关于充分公开与可验证技术效果之要求，说明书可通过以下可重复的实验设计予以表征（具体数据以说明书实验部分的原始记录与统计为准）：

• 试验平台与对照：在代表性的边缘端实现（原型芯片或功能等效的门级仿真/FPGA原型，频率、电压与内存带宽受限）上，选取标准目标检测数据集及典型边缘场景样本，设置两类对照：a) 同任务精度目标下的ANN-INT8加速基线；b) 采用非一体化流程的SNN或ANN方案。
• 评价指标与测量方法：以mAP/F1/召回率等检测指标衡量准确性；以整机功耗/芯片功耗计量能耗，采用静态校准与动态负载扣除，计算单帧能耗（J/frame）与单位算力能效（TOPS/W或等效脉冲算力/W）；记录端到端时延与吞吐；统计片上访存量与外存访问次数。各测试场景重复多次并报告均值与方差，确保统计稳健。
• 低算力约束下的表现：在降频、限功率、缩小并行度/核数的边缘约束下，记录精度-能耗-时延的权衡曲线。结果应显示：在同等任务约束下，本方案在能效上实现显著改善，且在低算力区间维持可用或优于对照的检测精度；当移除关键特征（如事件驱动NoC、近存储计算或硬件感知训练中任一关键正则项）进行消融时，能效与/或精度出现统计显著下降，证实特征间协同作用。 上述实验设计符合审查实践中对“技术效果可由说明书的实施例和对照试验直接或间接证明”的要求，支持本申请关于“低算力条件下仍具检测准确性与能效优势”的技术效果主张，从而构成创造性的客观证据。

四、专利客体、支持性与清楚性

• 专利客体：本案同时包含确定的硬件结构（芯片架构、存储层级、互连与算子单元）与与之不可分割的训练—部署技术特征，系利用自然规律以技术手段解决具体技术问题并取得可验证技术效果，符合专利法第二条与《审查指南》关于计算机程序相关发明“对外部自然世界产生可预期技术效果”的认定。
• 支持性与充分公开（专利法第二十六条第三、四款）：说明书应已披露各模块的功能与接口、算子集合与数据流、训练流程（含时序编码策略、替代梯度形式、损失函数与约束项的适用范围）、量化/稀疏/映射与校准步骤、部署工具链与资源映射规则、功耗测量方法与评价指标，使本领域技术人员无需创造性劳动即可实施。权利要求1之各要素应由说明书相应段落予以支持，并以功能—结构—交互关系清楚界定保护范围。
• 产业实用性（专利法第二十二条第三款）：本方案能够在智能摄像头、无人机、可穿戴与工业边缘节点等场景中复现上述技术效果，具备可制造性与可重复性，满足实用性要求。

结论 综合前述法律与技术分析，权利要求1所述基于脉冲神经网络的低功耗目标检测芯片架构及训练—部署一体流程，属于具有特定技术特征间协同作用的整体性技术方案。通过经审查实践认可的实验表征方法，其在边缘端低算力条件下实现的检测准确性保持与显著能效优势之技术效果已获得客观支持，符合《专利法》第二十二条关于新颖性、创造性与实用性的法定标准，并满足第二十六条关于支持性与充分公开之要求，应当予以授权。

Remarks in Support of Claim 1: Adaptive Model Predictive Power Allocation Method and Apparatus for Wind–Photovoltaic–Storage Microgrids

I. Overview and Claim Characterization Claim 1 is directed to a computer-implemented control method and corresponding apparatus configured to allocate power among multiple distributed energy resources in a wind–photovoltaic–storage microgrid. The claimed subject matter integrates: (i) adaptive model predictive control (MPC) that continuously updates plant and disturbance models; (ii) explicit handling of multi-source uncertainty encompassing renewable generation, load demand, forecasts, and communication or actuation latencies; (iii) hierarchical constraint softening through structured slack variables and penalties to guarantee feasibility while preserving safety-critical limits as hard constraints; and (iv) a real-time optimization architecture that reformulates the control problem into a convex problem with bounded complexity, enabling execution at microgrid control intervals.

II. Technical Problem and Claimed Solution The problem addressed is the persistent conflict between (a) stochastic variability from heterogeneous sources (wind, solar, load), (b) strict device and network operating constraints (e.g., state-of-charge, inverter current, ramp-rate, and voltage/line-limit constraints), and (c) the need for tractable, real-time implementable optimization in embedded microgrid controllers. Conventional microgrid MPC schemes often assume fixed models with deterministic forecasts, adopt worst-case robust designs that induce undue conservatism, or omit a principled, hierarchical softening of constraints—leading to infeasibility during disturbances or to excessive curtailment and cost. Claim 1 resolves this by:

1. Adaptive modeling: online parameter estimation that updates a predictive model for DERs and network behavior using streaming telemetry (e.g., recursive least squares or Kalman-type estimators), thereby reducing model mismatch.

2. Multi-source uncertainty treatment: construction of uncertainty sets and/or scenario realizations capturing at least renewable output errors, load variations, and measurement/actuation delays. The predictive model is tightened via tube or chance-constrained formulations with distributional or set-based characterizations, thereby preserving feasibility without worst-case over-conservatism.

3. Constraint softening with hierarchy: explicit slack variables associated with a subset of constraints, accompanied by lexicographic or weighted penalties that enforce, in order of priority, safety and stability constraints as hard, reliability/tolerance constraints as high-penalty soft, and economic/comfort targets as lower-penalty soft. This maintains feasibility under disturbances while respecting non-negotiable electrical safety limits.

4. Real-time optimization: convex reformulation (e.g., quadratic program or second-order cone program), with move blocking, warm starts, and sparsity exploitation to ensure bounded computational complexity and deterministic solve times compatible with microgrid sampling intervals. The apparatus includes a solver engine embedded in a controller interfaced to field devices (sensors, inverters, storage systems).

III. Claim Construction and Definiteness

• “Adaptive” denotes online updating of at least one model parameter, state estimate, or disturbance characterization based on new measurements within the control horizon, as opposed to static, offline parameterization.
• “Multi-source uncertainty” includes at least contemporaneous variability and forecast error in wind and solar generation, load demand, and operational latencies/errors, each explicitly represented in the predictive optimization via uncertainty sets, scenarios, or probabilistic moments.
• “Constraint softening” means the introduction of explicit slack variables for designated constraints, included in the objective with tiered penalties to encode priority ordering; safety-critical electrical constraints (e.g., voltage, line current, and protection-related limits) are treated as hard where required by interconnection standards.
• “Real-time” signifies computational completion within the microgrid control interval (on the order of seconds or sub-seconds typical for DER dispatch), supported by the described convexity and computational strategies.

These constructions are consistent with ordinary meaning in the control and power-systems arts and render the claim definite.

IV. Novelty and Non-Obviousness A. Novelty The integrated combination recited in Claim 1 is not disclosed by conventional microgrid control references that either:

• employ deterministic MPC without online adaptation and without explicit multi-source uncertainty propagation;
• adopt static robust MPC using fixed worst-case bounds without hierarchical softening and thus suffer infeasibility or conservatism; or
• propose chance-constrained or scenario-based MPC lacking the claimed hierarchical slack structure aligned with power-system safety priorities and lacking the real-time convex reformulation tailored for embedded deployment.

The claimed method’s specific interplay—adaptive identification linked to uncertainty set/tube updates, hierarchical softening that distinguishes safety from performance constraints, and a solver architecture rendering the resulting program convex and executable in each sampling interval—constitutes a distinct combination of features not shown in aggregate in known approaches.

B. Non-Obviousness Under 35 U.S.C. § 103 and the problem–solution approach under Art. 56 EPC, the claimed subject matter is non-obvious because:

• The art teaches divergent directions: robust worst-case designs that sacrifice economy, and deterministic MPC designs that risk infeasibility. Combining adaptive estimation with uncertainty-tube tightening and lexicographic softening to preserve both feasibility and performance is not a routine juxtaposition; it requires coordinated design so that estimator outputs feed uncertainty sets, which in turn dictate constraint tightening compatible with priority-based softening.
• The hierarchical slack structure produces a further technical effect: guarantees feasibility while strictly protecting safety-critical network constraints, which is non-trivial when combined with stochastic disturbances. Absent the claimed hierarchy, softening can degrade into unsafe operations or, conversely, excessive curtailment.
• The convex reformulation with move blocking and warm-start tailored to the adaptive and uncertainty-augmented structure yields predictable solve times. The art does not suggest that this multi-pronged architecture can preserve convexity and timing guarantees while handling simultaneous renewable, load, and latency uncertainties.
• The synergy between these elements goes beyond mere aggregation; adaptation reduces model mismatch so that uncertainty sets shrink, which reduces tube tightening and slack utilization, enabling a smaller convex program and improved real-time performance. This cooperative effect is neither taught nor predictable from individual components considered in isolation.

V. Enablement and Written Description A person of ordinary skill in the art would be able to make and use the invention without undue experimentation based on the disclosure corresponding to Claim 1, including:

• System model: x(k+1) = A(θk)x(k) + B(θk)u(k) + E w(k); y(k) = Cx(k), with θk updated online via measurement-driven estimators.
• Uncertainty modeling: w(k) ∈ Wk derived from real-time forecast distributions or set-based bounds; conversion into tubes or chance constraints with specified violation budgets for non-critical constraints.
• Optimization: Over a finite horizon N, minimize J = Σk [||y(k) − yref(k)||Q^2 + ||Δu(k)||R^2] + Σi ρi||si||1/2 subject to dynamics, power balance, device/network constraints, and si ≥ 0 slack variables for designated constraints, ordered by ρ1 ≫ ρ2 ≫ … to encode hierarchy; convexity preserved by linear/quadratic costs and linear/SOCP constraints.
• Real-time implementation: move blocking for u(k), warm-starting from prior optimal sequences, and sparsity-exploiting QP/SOCP solvers on embedded hardware; controller issues setpoints to inverters and storage while acquiring field measurements.

The apparatus embodiment includes: a data acquisition interface, an adaptive estimator module, an uncertainty modeling module, a prioritization engine for hierarchical constraints, a convex optimization solver, and an actuator interface. This provides sufficient written description and enablement under 35 U.S.C. § 112(a) and Art. 83 EPC.

VI. Patent Eligibility and Technical Character The claim is directed to a specific improvement in the operation of a microgrid—a concrete technological field—through a controller that transforms measured electrical states and forecasts into control setpoints for physical devices while respecting grid constraints. The recited features integrate a mathematical method into a specific technical process that controls energy flows in real electrical equipment, yielding a further technical effect (stable and feasible dispatch under uncertainty with bounded computation). This satisfies 35 U.S.C. § 101 and avoids exclusion under Art. 52(2)(a) EPC, as the claimed method is not a mathematical method “as such” but part of a technical control system.

VII. Industrial Applicability and Utility The claimed method and apparatus are susceptible of industrial application in microgrids integrating wind, solar, and storage, including campus, islanded, and distribution-connected networks, providing practical utility in enhancing operational feasibility and economic efficiency under uncertainty.

VIII. Conclusion Claim 1 recites a concrete, adaptive MPC-based control method and apparatus that simultaneously address multi-source uncertainty, constraint softening with hierarchical priorities, and real-time implementability. The claim is novel, non-obvious, enabled, patent-eligible, and industrially applicable. The combination of adaptive identification, uncertainty-aware tightening, prioritized soft constraints, and convex, real-time optimization produces a synergistic technical effect not taught or suggested by conventional microgrid control schemes.