PidSopdtBatchRolloutScore
【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
PidSopdtBatchRolloutScoreis an independent SOPDT batch closed-loop rollout operator for PID candidate scoring. It evaluates a batch of SOPDT 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 SOPDT 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 second-order process plus dead time:
y[k+1] = a1 * y[k] + a2 * y[k-1] + b * u[k - delay]Versus the FOPDT rollout this adds one history statey[k-1]and one coefficient (a1,a2instead of a singlea). The PID control law, scoring, candidate-axis SIMD, delay ring and tiling are identical toPidFopdtBatchRolloutScore.
工程语义
SOPDT rollout 与 FOPDT/IPDT 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。
区别在被控对象动态。SOPDT 多了一个历史输出状态:
y[k+1] = a1 * y[k] + a2 * y[k-1] + b * u[k - delay]它能表达二阶惯性、较慢过渡和更复杂的动态形态。相比 FOPDT,SOPDT 需要输入a1/a2/b/delay,并在 kernel 中维护y[k]与y[k-1]两个输出历史。
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 (the
benchmarkprogram), 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
| Tensor | Dtype | Shape | Meaning |
|---|---|---|---|
a1 | float32 | [B] | first output-history coefficient (y[k]term) |
a2 | float32 | [B] | second output-history coefficient (y[k-1]term) |
b | float32 | [B] | input gain term (u[k-delay]term) |
delay | int32 | [B] | input delay, clamped to0..31in kernel |
y0 | float32 | [B] | initial output |
sp | float32 | [B] | setpoint |
kp | float32 | [M] | PID candidate Kp |
ki | float32 | [M] | PID candidate Ki |
kd | float32 | [M] | PID candidate Kd |
best_result | float32 | [B, 8] | best candidate metrics per loop |
best_idx | int32 | [B] | best candidate index per loop |
best_resultlayout:
best_score,best_kp,best_ki,best_kd,best_iae,best_ise,best_overshoot,best_settling_timeBuild
cd prediction/ProcessControl/PIDModelFit/pid_sopdt_batch_rollout_score cmake -S . -B build -DCMAKE_BUILD_TYPE=Release -DSOC_VERSION=Ascend910B3 cmake --build build -j$(nproc)Note:
- this project now defaults to
ReleaseifCMAKE_BUILD_TYPEis not specified - on
node202, runtime typically needs:
export LD_LIBRARY_PATH=/usr/local/Ascend/ascend-toolkit/latest/lib64:$PWD/build:$PWD/build/lib:${LD_LIBRARY_PATH}Test
Python reference test:
python tests/test_pid_sopdt_batch_rollout_score.pyNPU smoke:
./build/test_aclnn_pid_sopdt_batch_rollout_score 0NPU / CPU benchmark:
./build/benchmark_pid_sopdt_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
创作声明:本文部分内容由AI辅助生成(AIGC),仅供参考