PidIpdtBatchRolloutScore

【免费下载链接】mat-chem-sim-pred面向工业领域，聚焦计算仿真、预测两大核心场景，构建面向流程工业"机理+数据"双轮驱动的领域计算层，推动AI for Science在材料化学领域的深度应用。项目地址: https://gitcode.com/cann/mat-chem-sim-pred

Overview

PidIpdtBatchRolloutScoreis an independent IPDT batch closed-loop rollout operator for PID candidate scoring. It evaluates a batch of IPDT process models against a shared PID candidate set on NPU and returns the best candidate per loop.

This operator is used in the tuning stage as the IPDT candidate simulation and selection kernel:

• input is process model parameters plus PID candidate arrays
• output is per-loop best score and best PID gains
• implementation is independent from the earlier model-fit operators

The plant dynamics are a pure integrator plus dead time:

y[k+1] = y[k] + b * u[k - delay]

This is thea = 1(no self-regulation) special case of the FOPDT rollout, so there is noacoefficient input. The PID control law, scoring, candidate-axis SIMD, delay ring and tiling are identical toPidFopdtBatchRolloutScore.

工程语义

IPDT rollout 与 FOPDT/SOPDT rollout 的候选评估框架相同：输入已辨识出的模型参数和一批 PID 候选，kernel 内部闭环递推，累计IAE/ISE/overshoot/settling_time/control_energy，计算固定 score 并选出每条回路的最优候选。

区别只在被控对象动态。IPDT 是积分对象：

y[k+1] = y[k] + b * u[k - delay]

它没有 FOPDT 的自回归衰减项a*y[k]，因此过程本身没有自稳能力，输入作用会持续累积到输出上。对于液位、库存、积分型热量累积等对象，这类模型比一阶自稳模型更贴近。

PID 候选不由本算子生成；本算子只接收kp[M]、ki[M]、kd[M]并评估。候选可以来自整定规则、规则附近扰动、人工网格或外部优化器。

当前 batch rollout 已经融合了候选特征、候选评分和候选选优；如果使用本算子直接输出best_result/best_idx，通常不需要再接独立候选评分/选优算子。pid_step_response_features仅用于需要保留完整候选轨迹特征表的另一条模块化链路。

Current status:

• correctness is validated against the in-process CPU reference (thebenchmarkprogram), max quality rel-err< 1e-3
• the rollout is a serial time recurrence; the candidate axis is evaluated with a wide vector SIMD lane (kLane=768) so the inner loop is throughput-bound rather than latency-bound

See benchmark report for the measured results.

Inputs And Outputs

best_resultlayout:

best_score,best_kp,best_ki,best_kd,best_iae,best_ise,best_overshoot,best_settling_time

Build

cd prediction/ProcessControl/PIDModelFit/pid_ipdt_batch_rollout_score cmake -S . -B build -DCMAKE_BUILD_TYPE=Release -DSOC_VERSION=Ascend910B3 cmake --build build -j$(nproc)

Note:

• this project now defaults toReleaseifCMAKE_BUILD_TYPEis not specified
• onnode202, runtime typically needs:

export LD_LIBRARY_PATH=/usr/local/Ascend/ascend-toolkit/latest/lib64:$PWD/build:$PWD/build/lib:${LD_LIBRARY_PATH}

Test

Python reference test:

python tests/test_pid_ipdt_batch_rollout_score.py

NPU smoke:

./build/test_aclnn_pid_ipdt_batch_rollout_score 0

NPU / CPU benchmark:

./build/benchmark_pid_ipdt_batch_rollout_score 0 64 1024 1024 0 2 0 64 # candidate_tile=0 => auto (min(C, kLane=768))

Documents

• Algorithm
• API Reference
• Benchmark Report

【免费下载链接】mat-chem-sim-pred面向工业领域，聚焦计算仿真、预测两大核心场景，构建面向流程工业"机理+数据"双轮驱动的领域计算层，推动AI for Science在材料化学领域的深度应用。项目地址: https://gitcode.com/cann/mat-chem-sim-pred