一种基于神经元网络的跨临界二氧化碳热泵系统排气压力的控制方法转让专利
申请号 : CN201811643046.X
文献号 : CN109764570B
文献日 : 2020-08-18
发明人 : 曹锋 , 王静 , 殷翔 , 李明佳
申请人 : 西安交通大学
摘要 :
权利要求 :
1.一种基于神经元网络的跨临界二氧化碳热泵系统排气压力的控制方法,其特征在于,包括:采集跨临界二氧化碳热泵系统运行中的环境温度Tair、蒸发器盘管温度Te、气体冷却器的出口温度Tgc,out、热泵出水温度Twater,out,代入公式(1)中,计算获得压缩机最优排气压力Popt;根据计算获得的压缩机最优排气压力Popt,控制跨临界二氧化碳热泵系统中电子膨胀阀的开度,使压缩机的排气压力达到最优排气压力Popt;
其中:Popt——压缩机最优排气压力/MPa;
Te——蒸发器盘管温度/℃;
Tair——环境温度/℃;
Tgc,out——气体冷却器的出口温度/℃;
Twater,out——热泵出水温度/℃;
Wij——输入层到隐含层的权值;
wi——隐含层到输出层的权值;
公式(1)通过以下方法获得:
1)采集三台机组7904组跨临界二氧化碳热泵系统的运行数据,获得数组;
2)建立PSO-BP神经网格,并获得PSO-BP神经网络的权值和阈值;
3)将步骤1)处理后的数组和步骤2)获得的神经网络的权值和阈值的数组代入PSO-BP神经网格求得神经网络的确切表达式为公式(1);
步骤2)中,PSO-BP神经网格的结构选择3层结构;3层结构包括输入层,隐含层和输出层各一个;
PSO-BP神经网络的输入层和隐含层之间的节点传递函数采用了正切S型传递函数tansig;隐含层和输出层之间的节点传递函数为Purelin函数;训练函数选择梯度下降法BP算法训练函数trainlm,学习网络函数选择BP学习规则learndm,网络学习选择带动量项的BP学习规则learngm;
PSO-BP神经网络的权值和阈值的数组表为:
2.1)输入层到隐含层的权值:
Wij={-1.373645179,-0.171434748,0.180450729,-0.419344707,-1.011174395,-
1.095370306,-0.830633156,-2.375743856,-0.257030363,0.803925612,0.444660281,-
0.956847463,-1.302595066,0.60939121,1.44270544,0.896090847,-0.256660406,
1.071666471,1.091072963,-0.137175234,0.825383828,-0.554595693,1.333515141,-
1.407159615,-1.071725234,-0.805831555,1.131061235,0.357316579,-0.903785699,
0.663157186,-0.331196959,0.742801954,-0.610567156,2.525753297,0.243592494,
0.709136427,-1.105587445,-0.72669056,1.346842937,0.751014569,0.952272284,
0.177353584,1.061300908,0.875134325,0.128631765,1.449397231,0.223057781,-
0.728439985,-1.089126606,-1.385787842,2.163516832,0.017385835,-0.254999419,
1.282306975,-0.973242962,-1.156798981,-0.16573785,1.239275104,-0.86799971,-
0.567617228,-0.439380446,-1.314739974,0.912920472,-0.645901465,-1.19571486,-
1.741144827,-0.661052052,-0.565010726,-0.422493372,-0.579636988,0.960136387,
0.771403746,0.755393263,-0.328503118,-1.271603343,0.798993956,0.438676136,
1.09133534,-0.425173762,-1.134201289,-0.444948685,-1.039746691,0.35972868,
1.134079237,-0.290332065,0.407184456,-1.206843505,0.592616827,-0.19963379,-
0.861030702,0.785647446,0.706934095,-0.652098923,-0.883946838,0.849861688,
0.329638324};
2.2)输入层到隐含层的阈值
θi={0.257558476,-0.89355435,-0.930294354,0.620820834,-0.505408308,
0.323818441,2.134062922,0.698001067,-0.692487021,-0.280434446,-0.379538764,
0.042341748,-0.793686679,-0.280328752,-0.076848832,2.427111937,-0.111656369,
1.334433533,-0.003113182,-0.958630098,-0.302250969,0.477703521,-0.050519198,
0.308304082};
2.3)隐含层到输出层的权值
wi={1.069690833,0.705072106,-1.130932743,0.053217769,-0.101927472,-
0.356915054,-2.359099898,0.387526948,1.132651956,-0.798591685,-0.261955516,-
1.402381137,0.798599759,0.866208212,1.387519681,1.236713142,1.700995412,
1.576768087,1.341827887,-0.062136221,-0.150827572,2.942764548,-0.391359771,-
1.067339334};
2.4)隐含层到输出层的阈值
a1={-1.198399403}。
2.根据权利要求1所述的一种基于神经元网络的跨临界二氧化碳热泵系统排气压力的控制方法,其特征在于,将步骤1)处理后的数组具体为对原始采集的数据组进行混合处理,剔除掉相同的数组变量;然后对数组进行三次随机排序;
步骤1)中,处理后的数组包含5个变量,分别是环境温度、蒸发器盘管温度、气体冷却器的出口温度、热泵出水温度和压缩机排气压力;
步骤2)中的模型的具体建立为将7904组数据分为两部分,前7000组作为网格训练集,参与神经网格的训练,904组数据作为测试集,对训练结果进行判定,前7000组训练集中取
15%作为判定集对判定结果的网格参数进行调试;
步骤3)中的PSO-BP神经网格的优化,主要包含两个方面:BP神经网格隐含层节点数和PSO算法粒子循环次数;通过利用步骤1)中数组对网格性能验证调试发现,隐含层节点数为
24时,算法粒子循环次数为200时,网格预测性能最佳。
3.根据权利要求1所述的一种基于神经元网络的跨临界二氧化碳热泵系统排气压力的控制方法,其特征在于,神经网络的输入层个数为4,输出层个数为1。
说明书 :
一种基于神经元网络的跨临界二氧化碳热泵系统排气压力的
控制方法
技术领域
背景技术
发明内容
0.444660281,-0.956847463,-1.302595066,0.60939121,1.44270544,0.896090847,-
0.256660406,1.071666471,1.091072963,-0.137175234,0.825383828,-0.554595693,
1.333515141,-1.407159615,-1.071725234,-0.805831555,1.131061235,0.357316579,-
0.903785699,0.663157186,-0.331196959,0.742801954,-0.610567156,2.525753297,
0.243592494,0.709136427,-1.105587445,-0.72669056,1.346842937,0.751014569,
0.952272284,0.177353584,1.061300908,0.875134325,0.128631765,1.449397231,
0.223057781,-0.728439985,-1.089126606,-1.385787842,2.163516832,0.017385835,-
0.254999419,1.282306975,-0.973242962,-1.156798981,-0.16573785,1.239275104,-
0.86799971,-0.567617228,-0.439380446,-1.314739974,0.912920472,-0.645901465,-
1.19571486,-1.741144827,-0.661052052,-0.565010726,-0.422493372,-0.579636988,
0.960136387,0.771403746,0.755393263,-0.328503118,-1.271603343,0.798993956,
0.438676136,1.09133534,-0.425173762,-1.134201289,-0.444948685,-1.039746691,
0.35972868,1.134079237,-0.290332065,0.407184456,-1.206843505,0.592616827,-
0.19963379,-0.861030702,0.785647446,0.706934095,-0.652098923,-0.883946838,
0.849861688,0.329638324};
0.042341748,-0.793686679,-0.280328752,-0.076848832,2.427111937,-0.111656369,
1.334433533,-0.003113182,-0.958630098,-0.302250969,0.477703521,-0.050519198,
0.308304082};
1.402381137,0.798599759,0.866208212,1.387519681,1.236713142,1.700995412,
1.576768087,1.341827887,-0.062136221,-0.150827572,2.942764548,-0.391359771,-
1.067339334};
附图说明
具体实施方式
0.444660281,-0.956847463,-1.302595066,0.60939121,1.44270544,0.896090847,-
0.256660406,1.071666471,1.091072963,-0.137175234,0.825383828,-0.554595693,
1.333515141,-1.407159615,-1.071725234,-0.805831555,1.131061235,0.357316579,-
0.903785699,0.663157186,-0.331196959,0.742801954,-0.610567156,2.525753297,
0.243592494,0.709136427,-1.105587445,-0.72669056,1.346842937,0.751014569,
0.952272284,0.177353584,1.061300908,0.875134325,0.128631765,1.449397231,
0.223057781,-0.728439985,-1.089126606,-1.385787842,2.163516832,0.017385835,-
0.254999419,1.282306975,-0.973242962,-1.156798981,-0.16573785,1.239275104,-
0.86799971,-0.567617228,-0.439380446,-1.314739974,0.912920472,-0.645901465,-
1.19571486,-1.741144827,-0.661052052,-0.565010726,-0.422493372,-0.579636988,
0.960136387,0.771403746,0.755393263,-0.328503118,-1.271603343,0.798993956,
0.438676136,1.09133534,-0.425173762,-1.134201289,-0.444948685,-1.039746691,
0.35972868,1.134079237,-0.290332065,0.407184456,-1.206843505,0.592616827,-
0.19963379,-0.861030702,0.785647446,0.706934095,-0.652098923,-0.883946838,
0.849861688,0.329638324}
0.042341748,-0.793686679,-0.280328752,-0.076848832,2.427111937,-0.111656369,
1.334433533,-0.003113182,-0.958630098,-0.302250969,0.477703521,-0.050519198,
0.308304082}
1.402381137,0.798599759,0.866208212,1.387519681,1.236713142,1.700995412,
1.576768087,1.341827887,-0.062136221,-0.150827572,2.942764548,-0.391359771,-
1.067339334}